HW

profilepcjoneskii12
FloridaRenewableEnergy.zip
Home>English homework help>HW

Florida Renewable Energy 1.PDF

Florida Renewable Energy 2.PDF

fenrg-09-680355 April 30, 2021 Time: 20:26 # 1

REVIEW published: 07 May 2021

doi: 10.3389/fenrg.2021.680355

Edited by: Dongran Song,

Central South University, China

Reviewed by: Jiawei Zhu,

Chang’an University, China Jian Yang,

Central South University, China Mingzhu Tang,

Changsha University of Science and Technology, China

*Correspondence: Tao Yu

[email protected]

Specialty section: This article was submitted to

Smart Grids, a section of the journal

Frontiers in Energy Research

Received: 14 March 2021 Accepted: 30 March 2021

Published: 07 May 2021

Citation: Wang K, Li Y, Wang X, Zhao Z,

Yang N, Yu S, Wang Y, Huang Z and Yu T (2021) Full Life Cycle

Management of Power System Integrated With Renewable Energy:

Concepts, Developments and Perspectives.

Front. Energy Res. 9:680355. doi: 10.3389/fenrg.2021.680355

Full Life Cycle Management of Power System Integrated With Renewable Energy: Concepts, Developments and Perspectives Kang Wang1, Yikai Li1, Xiaojun Wang1, Zengtao Zhao1, Ning Yang2, Shengcan Yu2, Yi Wang1, Zhanhong Huang2 and Tao Yu2*

1 CSG POWER GENERATION CO., LTD., Guangzhou, China, 2 School of Electric Power Engineering, South China University of Technology, Guangzhou, China

Under high-penetration of renewable energy, power grid is facing with the development problems such as production delay, wind and solar power abandoning. With the continuous growth of renewable energy installation such as wind power, photovoltaic (PV), as well as the increase of power generation capacity, it is urgent to increase peak- load and frequency regulation capacity on a large scale to alleviate the consumption problems caused by large renewable energy integration, and then requires power generation enterprises of peak-load and frequency regulation to increase relevant equipment assets. As a result, peak-load and frequency regulation enterprises must carry out scientific cost management of equipment assets. This paper introduces the concepts, developments and perspectives of life cycle cost (LCC) management of equipment assets in high-penetrated renewable energy power grid, and probes into cost collection and estimation scheme in the process of equipment asset management.

Keywords: high-penetrated renewable energy power grid, peak-load and frequency regulation, asset full life cycle cost, cost collection and estimation, full life cycle management

INTRODUCTION

Development and Consumption of New Energy With the continuous increase of environmental pressure and energy demand caused by energy development, the proportion of wind power, photovoltaic (PV) power generation and other renewable energy in the power grid is increasing year by year. Through policy guidance, preferential subsidies and other incentive policies, countries vigorously promote investment in wind power, PV and other renewable energy, develop advanced technology and architecture systems, so as to promote large-scale grid connection of renewable energy (Wen et al., 2008). According to statistics, by the end of 2020, the cumulative installed capacity of global offshore wind power has reached 32.5GW, and 162 offshore wind farms have been put into operation, an increase of 19.1% over the same period at the end of 2018, which indicates the promising prospect of renewable energy power development (Hu and Cheng, 2013; Feng et al., 2015). The variation of installed capacity of offshore wind power in the past decade is shown in Figure 1. The rapid development of renewable energy leads to major changes in the investment scale and asset management mode of power system. In 2019, the annual investment in renewable

Frontiers in Energy Research | www.frontiersin.org 1 May 2021 | Volume 9 | Article 680355

fenrg-09-680355 April 30, 2021 Time: 20:26 # 2

Wang et al. Full Life Cycle Management Review

FIGURE 1 | The installed capacity of offshore wind power in the past decade.

energy power in the world reached 53.1 billion United States dollars, and the rapid growth of asset investment greatly promoted the research and development of related technologies and implementation of projects. However, in the process of rapid development of renewable energy, the problem of renewable energy accommodation, such as abandoned wind and solar, is becoming increasingly serious with the disharmony of the spatial and temporal distribution of capacity and load (Hu and Cheng, 2013).

Since 2012, China’s installed PV capacity began to grow rapidly. In 2014, accommodation problem caused by this rapid growth began to appear, and the problem of abandoned solar aroused the attention of whole society. Therefore, the power grid needs to further improve the ability to absorb new energy (John, 2017). According to Information Brief of PV Power Generation Construction from January to September 2015 released by National Energy Administration, cumulative PV power generation in China from January to September was 30.60 billion k·Wh, and PV power abandoning was about 3.03 billion k·Wh, with a solar abandoning rate of 10%. After wind power encountered wind abandoning dilemma, PV power generation also fell into the dilemma of capacity allocation redundancy (Bird et al., 2016). In the development of wind power in the past five years, two phenomena have been accompanied by: (1) the good news of the continuous increase in installed capacity of renewable energy; (2) the dilemma of "abandoned electricity" such as abandoned wind and solar due to insufficient accommodation capacity. The statistics of China’s abandoned wind power from 2011 to 2015 are shown in Figure 2. From 2011 to the first half of 2015, China’s total wind power on grid was 561.774 billion k·Wh, the total abandoned wind power was 80.191 billion k·Wh, and the average abandoned wind rate was 14.27%. In addition, large-scale integration of new energy power generation has made power frequency imbalances increasingly frequent (Basmadjian and Meer, 2018). In the context of China’s economy entering the new normal of medium and high-speed growth,

the problem of abandoned power has become increasingly prominent (Kasis et al., 2016).

Peak-load and frequency regulation power supplies can well alleviate accommodation problems caused by large-scale grid connection of renewable energy, and Improve system operation level (Chen et al., 2009). Fujian province of China increased the average utilization hours of nuclear power by more than 700 h year-on-year, and without abandoned wind, water, and solar phenomenon, which improve the utilization of electric power production equipment and increase the return on investment (Kasis et al., 2016). Renewable energy power generation investment rise needs to pay attention to equipment management and investment effectiveness (Kasis et al., 2016; Dui et al., 2018). Hence, under the guidance of renewable power investment mode and system, the investment planning for renewable energy industries such as wind power and PV and the asset management of power enterprises should take into account the economy and reliability in full life cycle, so that the huge renewable energy power construction can get a better return on investment (Yildiz and Kazimi, 2006; Spertino and Graditi, 2014). Combined with power equipment management and related technical characteristics in large-scale renewable energy grid connection, this paper explores a refined, multi- angle and strongly related asset cost management mode, which provides an important method channel for the environmental friendliness and green economic function of renewable energy power (Billinton and Huang, 2010; Dui et al., 2018).

Organization of the Paper The rest of this article is arranged as follows: Section “Power Assets Full Life Cycle Cost Management” summarizes the development process and research status of the whole life cycle cost management of assets and power equipment at home and abroad. Section “LCC Cost Estimation Model” gives the estimation model of power equipment LCC by introducing the structure and estimation method of LCC in detail. Section

Frontiers in Energy Research | www.frontiersin.org 2 May 2021 | Volume 9 | Article 680355

fenrg-09-680355 April 30, 2021 Time: 20:26 # 3

Wang et al. Full Life Cycle Management Review

FIGURE 2 | The statistics on abandoned wind power nationwide between 2011 and 2015.

“Conclusion” summarizes the full text and gives the direction of the next stage of research.

POWER ASSETS FULL LIFE CYCLE COST MANAGEMENT

Asset Life-Cycle Cost Management According to IEC60300-3-3 standard formulated by International Electrotechnical Commission, full life cycle refers to the life cycle stage of equipment, which can be divided into concept and definition stage (Hedley-Whyte, 2000), design and development stage (Arif and Khan, 2010), manufacturing stage (Lee et al., 2016), installation stage (Asiedu and Gu, 1998), operation and maintenance stage (Solomon et al., 2000), and decommissioning disposal stage. Therefore, full life cycle cost (LCC) is the sum of all costs incurred in the above stages. Full LCC occurs in different property rights, which can be seen from the perspective of society (Ozbay et al., 2004), producers and users (Kiritsis et al., 1999; Mascitelli, 2004). From the perspective of asset users, most of the research focuses on the estimation and modeling of product design phase (Park and Simpson, 2003; Liu et al., 2008).

The concept of LCC originated in the United States in 1927 and was proposed by Department of defense (White, 1976; Bajaj et al., 2004). In 1933, the general audit office of the United States formally proposed the concept of LCC for the first time (White, 1976). In the 1960s, this concept was successfully applied to F16 fighters (Yeung et al., 2013). In 1996, the United States Department of defense began to formally study LCC theory, which was first used in the army. Later, Britain, France, Germany and other countries gradually applied LCC theory in the army. With the successful application of LCC theory in the military neighborhood, it gradually gained attention in the civil field (Zhang and Wang, 2012). With the gradual expansion of the application field of the theory, scholars began to focus on engineering design, equipment selection, equipment

maintenance, equipment decommissioning and other aspects of extensive and in-depth research.

Literature (Asiedu and Gu, 1998) described the complete steps of LCC theory and gives its reason, content and corresponding model of each step. Literature (Curry, 1989) summarized the life estimation of aerospace electronic equipment by the United States air force using LCC theory, and introduces a standardized evaluation procedure ‘STEP.’ Moreover, LCC technology is applied to the modernization of aircraft in-flight refueling and electronic system (Woodward, 1997; Seo et al., 2002). In literature (Furch, 2016), LCC theory is applied to establish the railway vehicle model, and the cost calculation formula of each stage is given. Besides, a prediction method for full life cycle scrap time of electronic components is established, a series of quantitative market or technical attributes were identified and obtained, and the scrap time of components was calculated by statistical method (Solomon et al., 2000). In addition, LCC analysis is carried out for the components of energy meter and resonant circuit in power system (Meyer and De Doncker, 2006; Cai et al., 2011). Literature (Nilsson and Bertling, 2007; Tian et al., 2011; Shafiee et al., 2016) applied LCC theory to wind power industry, and literature (Tian et al., 2011) established condition monitoring systems (CMS) based on LCC, which can reduce indirect damage in case of failure and provide favorable conditions for maintenance plans. Furthermore, it presents a LCC analysis strategy and employs CMS to improve the single wind turbine maintenance plan for onshore and offshore wind farms (Nilsson and Bertling, 2007). In literature (Shafiee et al., 2016), a wind farm investment cost regression model based on commodity price and seawater depth is proposed to estimate the accurate LCC.

In the fierce global competition environment, world-famous electric power enterprises have realized the importance of asset management for enterprise development, as well as electric power industry is also constantly exploring the full life cycle management and application of assets (Shahidehpour and Ferrero, 2005), so as to promote the maximum efficiency of

Frontiers in Energy Research | www.frontiersin.org 3 May 2021 | Volume 9 | Article 680355

fenrg-09-680355 April 30, 2021 Time: 20:26 # 4

Wang et al. Full Life Cycle Management Review

assets and serve the development of enterprises. Particularly, LCC theory has been gradually promoted and applied in the electric power industry.

National Grid Corporation of United Kingdom has integrated some intelligent management tools into daily management, established a complete set of asset management information sharing platform, among which project management, production operations, maintenance, and other fields are associated, finally achieve data integration, according to the need to generate various reports for reference of all kinds of managers.

In the late 1990s, Canada’s Hydro One introduced the concept of asset full life cycle management and established a complete set of asset evaluation methods (Danish et al., 2014). In addition, it is worth noting that this company’s asset management business has selected professional outsourcers to be responsible for the management of some assets, which not only saves the management workload of the main business personnel, but also improves the asset management level.

Ashburton Power Company of New Zealand makes a specific asset plan before making a clear investment objective, comprehensively considering all aspects of power grid planning, equipment transformation, maintenance and so on. Secondly, based on equipment condition and load forecast, the investment plan is optimized and analyzed. Finally, the information system is adopted to analyze how to achieve the optimal unit cost in the whole process of asset procurement, construction, operation and maintenance, transportation and scrapping.

Given the current situation of the power sector in Afghanistan, for improving the electricity environment in rural and remote areas, LCC theory is introduced to establish a cost-effective hybrid system. In 2004, International Power System Conference advocated that equipment manufacturers provide management reports for full life cycle of equipment and products (Lombardi, 2003; Joseph et al., 2018). As a result, world’s major electrical equipment manufacturers, such as ABB Group and Siemens, began to study full life cycle management of their products (Zhang and Cai, 2014; Zakeri and Syri, 2015). In 2005, representatives of more than 50 countries and regions, including the United Kingdom and Norway, established international organizations of asset LCC (Steen, 2005).

The asset management of international advanced electric power enterprises is aimed at minimizing the life cost of assets and maximizing the investment value: (1) pay attention to the analysis of investment in the early stage and determine the investment strategy through the optimal rating; (2) determine the health condition of the assets by rating them and determine their disposal methods to achieve the highest utilization rate; (3) the employ of information means to achieve assets procurement, construction, operation and maintenance, return and scrap of full life cycle management.

Overview of Power Equipment Life Cycle Management In literature (Shi et al., 2009), taking a 220 kV heavy load substation as an example, considering the loss of social output value, the comprehensive economy of full life cycle of substation

construction is calculated and compared, which concludes that the scheme has the best reliability and economy is obtained.

Literature (Kim et al., 2010) establishes a two-dimensional model of power system LCC for the drawback that the application of power system LCC is traditionally limited to specific equipment or stage, in which research status of LCC technology is reviewed. The cost breakdown structure is described in detail from the device level and the system level. Besides, the maintenance cost of the combination of reliability-centered maintenance (RCM) and fault repair (overhaul) is also analyzed. On this basis, some studies have also proposed a component-cost- time 3D model (Luo et al., 2011).

Literature (Cai et al., 2011) analyzed the latest progress in LCC technology, problems that should be paid attention to and several suggestions for the management of LCC. A full LCC-benefit model for energy-saving transformation of distribution network is established in literature (Karamouz et al., 2017), in which a decision method considering financial and technical constraints is proposed. It is effective to apply the model and method to the actual distribution network transformation. Besides, in order to overcome the problem of neglecting the medium and long-term cost and underestimating the short-term investment in the current economic evaluation of power system, a three- dimensional LCC model of the whole power system is established from the perspective of component dimension, cost dimension and time dimension. Through the analysis of the structure of the device layer, a series of economic evaluation strategies based on LCC are proposed, and the transformation of devices with different lifetime is studied Change the cycle (Liu et al., 2012).

At present, the practice of asset LCC management mainly focuses on the following two forms:

The first is to seek a new breakthrough in asset management mode based on the cost management of full life cycle of assets. It mainly standardizes asset management through management means such as internal rules and regulations or norms of enterprises, and integrates cost management concepts into daily management, such as optimizing design schemes, equipment selection, cost schemes and other efforts to achieve the goal of the lowest asset cost of equipment or system.

Second, relying on the information system to achieve asset life cycle cost management. The labor of asset management is huge, which makes it cumbersome, time-consuming and prone to human errors. The establishment of a new information platform can break the departmental barriers, connect the processes of asset planning, design, construction, operation and maintenance, return and scrap, so that realize the cooperation of each module. The full life cycle management of assets is realized. Through the strict management of the full life cycle cost of equipment, the links of material procurement, equipment collection, financial settlement and payment are strictly controlled. The original mode of extensive management and information isolation is changed, and the equipment asset management information system of real-time dynamic and fine management is established. However, due to equipment defect state maintenance, risk assessment lacks support, which cannot provide decision-making basis for equipment overhaul and technical reform. At the same time, because of lack of effective assessment and assessment

Frontiers in Energy Research | www.frontiersin.org 4 May 2021 | Volume 9 | Article 680355

fenrg-09-680355 April 30, 2021 Time: 20:26 # 5

Wang et al. Full Life Cycle Management Review

means, as well as lack of quantitative assessment indicators for equipment and management, resulting in most of the current power enterprises still cannot solve the contradiction between low cost and high utilization of assets.

Asset management is no longer the management of a certain link, but the management of full life cycle of the equipment, which makes the asset management more scientific and makes the equipment achieve the optimal cost in full life cycle (Li et al., 2018; Orfanos et al., 2019). At present, the management mechanism of power grid companies has been difficult to meet the needs of the rapid development of power grid. It is urgent to transform and improve the asset management model, promote the information and digitization of asset management, and accelerate the construction of world-class power grid. Therefore, the research significance of asset life cycle management is as follows:

Firstly, it should not only implement the national strategy and enterprise strategy and continuously improve the sustainable development ability of power grid, but also vigorously promote the high-quality development of power grid and build a world-class energy internet enterprise. The concept of asset management is the source of exerting asset management efficiency and improving asset management capabilities. It plays an important role in improving the overall asset management performance of the enterprise. Enterprises must introduce advanced asset life cycle management concepts.

Secondly, enterprise asset life cycle management is not only a process of concept innovation, but also a process of technology application. By employing asset life cycle management technology, the asset life cycle resource value and asset utilization efficiency of power grid enterprises can be improved.

Thirdly, asset life cycle management concept determines the direction and content of asset management decision-making of power Grid Company, which has macro guiding significance.

Fourthly, the large scale, wide distribution and variety of assets of power grid companies increase the difficulty of life cycle management of assets of power grid companies. Power grid companies must innovate the concept of asset management, apply the full life cycle theory to asset management, take the full cost management of assets within the full life cycle as the basis for management decisions, and pay attention to the long-term nature and efficiency of asset management, which is an innovation of the traditional asset management concept (Liu et al., 2012).

LCC COST ESTIMATION MODEL

Since the concept of LCC is applied soon, there is no correlation between the historical data of relevant asset management systems of all enterprises, and the data of full life cycle is missing. Besides, the cost of equipment assets operation and maintenance can only be counted according to the management units, and cannot be collected into the equipment and the current situation of basic data cannot meet the requirements of efficient asset management (Xu and Wang, 2011). Therefore, it is necessary to estimate the assets of power system, and then estimate LCC (Liu et al., 2012;

Lee et al., 2020). LCC estimation also estimates the possible LCC in the future, which is related to the establishment of budget (Savoretti et al., 2017), quotation generation (Govil, 1984) and development strategy (Govil, 1985), which is an indispensable part of the cost efficiency (reliability) evolution model in engineering (Yang et al., 2017).

The LCC estimation method is mainly established and implemented around the cost structure of the research target equipment. A reasonable estimate of the cost of the entire life cycle of the equipment is made in the form of engineering standards, and a targeted estimation method is used in conjunction with the composition of the entire life cycle of the equipment. Through the integration of various methods and cost components, a unified and comprehensive LCC estimation model is formed. In the process of establishing and revising the estimation model, the cost calculation of a single device is realized through accounting, statistics, and apportionment of various costs based on the current data situation. Taking into account the importance and value of the equipment, the LCC estimation work uses the method of setting the correction coefficient matrix to characterize the equipment difference, and the LCC estimation model is constantly revised and improved in the long-term work by mathematical checking and empirical judgment.

Structure Composition of LCC Based on the full life cycle management requirements and operational characteristics of the research equipment, the cost structure of each part of the equipment LCC is decomposed. In terms of selecting equipment to be studied, this article takes important equipment such as transformers, generators, and circuit breakers in the power supply equipment of the power system as the main research objects. The above important equipment occupies an important position in the operation of the power grid and has a relatively comprehensive and highly targeted Cost management process. According to the equipment survey results, combined with the current status of equipment cost management in the power grid and advanced LCC theory, the cost structure based on the LCC management method mainly includes investment costs, operating costs, maintenance costs, failure costs, scrap costs, etc. Decompose the cost of each part of LCC, and its cost structure is shown in Figure 3.

Particularly, relevant cost breakdown of LCC can be computed by:

LCC = CI+CO+CM+CF+CD (1)

where CI represents cost of investment; CO means operating cost; CM denotes maintenance cost; CF stands for fault cost; CD is abandon cost.

These costs can also be further broken down into sub costs, as mentioned above, LCC is related to future costs, and when the time value of funds is considered (Bastian, 2011), the cost incurred in the future shall be reduced and corrected multiply by (

1+r 1+Kcpi

)yeari , where Kcpi represents CPI index, which reflects

inflation; r denotes discount rate, and yeari means service years.

Frontiers in Energy Research | www.frontiersin.org 5 May 2021 | Volume 9 | Article 680355

fenrg-09-680355 April 30, 2021 Time: 20:26 # 6

Wang et al. Full Life Cycle Management Review

purchase cost

Installation cost

service fee

Construc�on cost

Energy cost

On-call cost

Labor costs

Blackout cost

Replacement cost

Elimination of defect costs

Return cost

Demoli�on fee

Residual value

Spare parts costLCC

Investment cost(CI)

Operating cost(CO)

Failure cost(CF)

Scrap cost(CD)

Maintenance cost (CM)

Resolve

Solve

Engineering es�ma�on method

Parameter es�ma�on method

Analogy es�ma�on method

Engineering es�ma�on method

LCC

FIGURE 3 | Cost breakdown structure.

Some studies also put forward the environmental cost (Li et al., 2018), but for power supply of peak-load and frequency, it is relatively clean, so this item cannot be considered.

Estimation Method The research object of LCC estimation is capital invested in the future (Lee et al., 2020). Through estimation of capital to be invested, it provides an important basis for economic judgment and final decision-making of the scheme (de Jong and Declercq, 2012). LCC estimation is the key content of LCC technology research and the basis of technology application.

Some studies believe that there are two main channels for data collection: professional manufacturers and suppliers, and historical data (Schneiderova-Heralova, 2018). The amount of data and information obtained determines the cost estimation method used. Besides, there are two kinds of estimation methods: certainty and uncertainty. For the former, there are more evaluation models (Cole and Sterner, 2000; Vahdat-Aboueshagh et al., 2014); for the latter, there are Monte Carlo method (Ammar et al., 2013; Goh and Sun, 2016), fuzzy set method (Shahata and Zayed, 2013; Plebankiewicz et al., 2020) and neural network (NN) (Ilg et al., 2017). For the existing asset estimation methods, the accuracy of LCC model depends on the choice of calculation method and the certainty of data. LCC of power system in

estimation scheme at present is very diverse. In view of the different stages of development and enterprise data types can be estimated according to the advantages and disadvantages of the scheme selection. Defects of technological level is that the barriers to a large amount of data is temporarily unable to break in the process of cost separating, and partly by artificial decomposition cost reimbursement need to upload the data to the cloud, the time cost and the statistical labor costs rose, as a result, the unity of the intelligent data analysis big data platform construction is the core content of LCC technology, in combination with the discussion of data and analysis the process needed a big enterprise data platform construction. Therefore, the content of data application construction is not discussed. This paper only discusses the system framework and model application established by LCC asset management. The estimation methods applied in the big data analysis platform are introduced as follows:

Gray Fuzzy Estimation Method Gray fuzzy estimation method (Chen and Ren, 2018) is that when some evaluation indexes cannot be accurately quantified, interval fuzzy method is usually used to evaluate the risk of contractor selection and installation construction unit selection. Fuzzy algorithm combines expert evaluation, fuzzy interval setting and other methods to integrate and calculate the evaluation opinions

Frontiers in Energy Research | www.frontiersin.org 6 May 2021 | Volume 9 | Article 680355

fenrg-09-680355 April 30, 2021 Time: 20:26 # 7

Wang et al. Full Life Cycle Management Review

given by multiple experts by the evaluation registration (Miah et al., 2017), so as to obtain the comprehensive evaluation of the risk of contractor or supplier selection, and formulate the cost quantification strategy based on the risk level.

Parameter Estimation Method Parameter estimation, which is based on a lot of historical cost data of similar equipment, selection of sensitive to cost several main physical parameters and performance features, then employing the regression analysis, gray system and neural network data processing method to set up the mathematical relationship of cost between the parameters, so as to estimate LCC or estimate the cost of a main unit. The first task of establishing the cost parameter relationship is to determine which characteristic quantity the cost is related to. In this method, the most important link is the database (Wang et al., 2013). Database must meet some specific requirements, such as establishing related connections for similar power equipment, in which each similar data unit should be composed of similar components and be processed consistently in the same way, that is, to ensure comparability. Otherwise, it will lead to obvious deviations in the estimation relationship, or even unreliable.

With the continuous accumulation of data, the model can be modified, and the more the model is used, the higher the accuracy will be. Therefore, this method is most widely used in LCC estimation, in which full LCC can be approximately related to quality, yield, performance and other characteristic variables. Compared with analogy estimation method, parameter estimation method reflects the relationship between cost and attribute. Therefore, as long as get the value of some of attribute parameters of complex system, it only needs to input the characteristic quantity to calculate the cost of the equipment at this stage. Then according to equipment reliability, maintainability and other parameters calculate the cost of operation, maintenance and scrap recovery phase, so as to obtain LCC. This method is the most commonly used cost estimation method in the early stage of full LCC analysis, especially in the absence of detailed planning and design specifications. Another advantage of establishing this kind of cost parameter relation is that it can quickly estimate the influence of the change of power equipment performance or some characteristic parameters on cost, so as to evaluate the influence of cost when the scheme is chosen during the planning and design and when the scheme is changed.

The calculation equation of parameter estimation method can be expressed as follows:

C2 = C1 ( S2 S1

)n CF (2)

where C1 represents actual engineering costs of similar projects; C2 denotes costs required for the proposed project; S1 means production scale of similar projects; S2 stands for production scale of the proposed project; CF represents price conversion index; and n denotes production scale index.

The value of n: the scale is enlarged, mainly with the increase of equipment capacity, n is 0.8∼0.9; for high-pressure equipment, n is 0.3∼0.5, and usually the average value of n

is about 0.6. Therefore, this estimation method is also called 0.6 index method. However, there are obvious drawbacks to this approach. First of all, it needs a lot of historical data, which is almost impossible to obtain detailed historical data since the LCC management of power companies started late. Secondly, the model established by this method only represents the law of changes in the past costs. The period of power system engineering is large, and the geographical gap between regions is also large. These differences will lead to an increase in the error of estimation model and affect decision-making. Furthermore, comparing the various attribute parameters of a complex system, parameter estimation model only relies on limited and easily measurable parameters for cost estimation, and does not consider various situations in detail. It is generally used in the early development stage of complex system, when there are only system specifications but no detailed planning and design specifications, especially when the power equipment is not standardized. The reasonable degree of the model established by the parameter estimation method depends on the staff ’s understanding of the system and their modeling experience and skills. The prediction accuracy is highly subjective, and this method is no longer applicable when the new system adopts advanced development and production technology.

Engineering Estimation Method Engineering estimation method is a traditional cost estimation method, also known as detailed estimation method or bottom- up method. It uses work breakdown structure to calculate each cost unit item by item from bottom to top, and then add item by item to get the total LCC (Ilg et al., 2017). Besides, engineering estimation method divides the research object into different sub- parts, and carries on the cost estimation, respectively, according to different characteristics of the parameters of each part. Finally, the estimated value of each part is summed up to obtain the total LCC. This estimation method is generally used in planning, development and production of research objects. With the accumulation of analysis, more and more data support for estimation, and the result of estimation becomes more and more accurate. It starts from the lowest level work unit, calculates LCC item by item from bottom to top by using work breakdown structure, sums up the cost of each work unit in the system, and then obtains the value of the upper-level cost unit item by item, and finally obtains LCC.

When applying engineering estimation method to calculate cost of each unit, it is necessary to collect detailed data information about the relevant costs. It is not difficult to observe that the advantages of this method are detailed and specific, with high estimation accuracy, but the disadvantages are cumbersome, time-consuming, heavy workload and complex calculation process. Therefore, this method can only be adopted after detailed design and mastering the relevant information of the equipment and the cost of its use and maintenance, which can be used to estimate the cost of some decision-making problems in the later stage.

Its mathematical model can be expressed as:

C = C1 +C2 +···+Cn (3)

Frontiers in Energy Research | www.frontiersin.org 7 May 2021 | Volume 9 | Article 680355

fenrg-09-680355 April 30, 2021 Time: 20:26 # 8

Wang et al. Full Life Cycle Management Review

where C represents LCC; and Ci denote cost of each unit at different stages.

The cost of each unit can be further divided into sub-units to form a complete cost breakdown structure diagram of the equipment, so as to obtain the estimated value of the total cost. This method is a detailed estimation method, its accuracy is directly related to the amount of information obtained, and is generally used in the later stage of the project.

Analogy Estimation Method Analogy estimation method is a method to estimate LCC of equipment by referring to the known cost information and other data of similar equipment. Based on the existing data of similar equipment, the equipment to be built is compared with it, and the fixed coefficient value is taken according to the characteristics of the latter to estimate its cost. Analogical estimation method is generally used in the early stage of equipment life, and its accuracy depends on expert experience.

When the data of similar power equipment is relatively reliable and the database is complete, this method is a more suitable estimation method. In most cases, it is used in the early planning and design stage of LCC to preliminarily estimate full LCC of power equipment.

It is a method to estimate the cost of the target equipment by comparing known information of the same type of equipment. Particularly, the implementation steps are as follows: firstly, select the sampling equipment, and the key parameters of the sampling equipment shall be the same as the existing equipment, and then compare the existing equipment with the sampling equipment. During the comparison, the characteristic parameters of the existing equipment can be set according to the different points between the equipment. Finally, LCC is obtained through comparing the characteristic parameters with the historical values of the sampling equipment. As the key part of the analogy method, the characteristic parameters in the actual use, usually call some experts, comprehensive research and judgment after the value, so the analogy method is also known as the expert method. This method is mainly employed in the planning and feasibility study stage at the initial stage of engineering construction, which applies to the situation that the data of similar projects are more accurate and detailed.

Moreover, it is a method to estimate LCC by referring to the cost data of completed projects similar to construction projects (Angelis and Stamelos, 2000; Steinert, 2009). Employing the analogy method to calculate the cost is mainly to use the cost data of similar projects, and select the correlation coefficient to correct according to the specific situation, so as to accurately estimate the cost of the proposed project. The selection of correlation coefficient is very important, which is generally determined by consulting experts. Its mathematical model can be expressed as:

C = C0 n∑

i = 1

aiKi (4)

where C denotes cost of proposed project; C0 represents cost of similar projects; ai stands for the proportion of labor cost, material cost and procurement cost in the total cost of similar

projects; Ki means the correlation coefficient of labor cost, material cost and purchase cost between the proposed project and similar project.

The analogy method estimates the cost based on the cost of similar products or technologies in the past. Besides, this method updates the historical data to reflect the impact of rising costs and technological progress, which is suitable for cost estimation with historical data and actual data reference.

Neural Network Method Artificial neural network (ANN) has been studied since the early 1940s, which is an intelligent computing system that simulates a biological NN with a computer network system. Furthermore, ANN can simulate some unique behaviors of the human brain, such as learning, memory, and recall, through self-learning, self- organization, self-adaptation, and nonlinear dynamic processing. The main advantages of NN estimation method are as follows: because ANN owns self-learning function, through the network data training, it can quickly and accurately simulate the results, so it does not need to establish a specific mathematical model of the cost. There are many uncertain factors, such as different electrical parameters, equipment operating conditions, climate, policies, etc., in LCC estimation of power transformers, so it is more accurate and objective to use neural network to calculate.

However, NN estimation method also has certain shortcomings: a large amount of historical data of power transformers is required during model training, and this data is often lacking in practical applications; the choice of hidden layers does not have a very scientific basis, which can be determined after trial calculation; it is not easy to obtain the sensitivity of the key factors of LCC of power transformers.

Activity-Based Costing Method Activity-based costing (ABC) is employed to calculate the cost of the equipment by summing up the activities related to the power equipment (Özbayrak et al., 2004; Karim et al., 2012). Based on historical information or estimated data, it first calculates the unit cost of each activity, and then calculates the activity consumed by new equipment, multiplying the two to get the total cost of power equipment (Waghmode and Sahasrabudhe, 2012; Bierer et al., 2015). It is mainly used in the later stage of LCC. The practical operation steps are as follows: À select the main activity; Á collect the cost of resources to the homogeneous cost base; Â select the cost driver; Ã calculate the allocation rate of each cost base; Ä allocate the collected cost in each cost base to power equipment according to allocation rate of cost base; Å summarize and calculate the total cost of power equipment. The main problem of this method is that it is not easy to obtain unit activity cost (Ben-Arieh and Qian, 2003).

Case-Based Reasoning Method In short, case-based reasoning method (CBR) adopts past problem-solving methods to deal with new problems (Ji et al., 2012). Its main spirit lies in how to systematically preserve and deal with the previous problem-solving knowledge and experience, in order to solve the new or repeated problems encountered, so as to reduce the mass of information, avoid

Frontiers in Energy Research | www.frontiersin.org 8 May 2021 | Volume 9 | Article 680355

fenrg-09-680355 April 30, 2021 Time: 20:26 # 9

Wang et al. Full Life Cycle Management Review

repeated process load. At the same time, CBR can accumulate experience. Each time a problem is solved, the new experience is saved. Nearest neighbor technique is probably the most widely used technique in CBR. For each case attribute, determine the similarity between the problem (target) case and the cases in the case base. This measure can be multiplied by a weighting factor. The similarity sum of all attributes is then calculated to provide a similarity measure between the case in the library and the target case.

The estimation range of LCC in CBR is limited by sample value and cannot be extrapolated, but its characteristic parameter quantity is independent of sample size. That is, the number of feature attributes can be increased a lot. Besides, it doesn’t need to judge the number of feature coefficients by considering the number of samples. Moreover, CBR method is simple and can obtain superior estimation results.

Expert Estimation Method Expert estimation method is the estimated value of full LCC of equipment based on expert experience judgment, which is the application of Delphi method in the prediction technology in cost estimation (Steinert, 2009). When using the expert estimation method, a certain number of experts independently estimate the corresponding equipment, and then synthesize them to obtain the estimated cost of equipment. Particularly, expert estimation method is generally adopted in the absence of data or the difficulty of collection, and the insufficient number of statistical samples, as well as employed as an auxiliary estimation of other estimation methods. Its mathematical model can be expressed as follows:

C = ∑n

i = 1 Ci n

(5)

where C denotes the estimated value of cost unit, here the average value is taken; Ci is the estimated value of the ith expert for cost unit; and n is the number of experts participating in the estimation.

Blind Number Theory Estimation Method Blind number theory estimation method is to comprehensively consider the attributes and characteristics of various uncertain information from the initial purchase to the later decommissioning process, and make a reasonable evaluation of the uncertain information, so as to determine the blind number expression of full LCC (Liming and Bo, 2020).

Blind number expressions are expressed by the following equation:

f (x) = { ak, x = xk (k = 1, 2, · · · , n)

0, otherwise (6)

where ak indicates the reliability of a blind number; x = xk represents a possible value or range of possible values for a blind number.

Compared with traditional deterministic LCC calculation method, LCC method based on blind number theory can’t only calculate the expected value of LCC, but also obtain possible distribution intervals of different costs and corresponding

credibility information, so as to improve the rationality of estimation results.

Comparison of Various Estimation Methods The purpose and precision of LCC estimation vary greatly in different stages due to variety and complexity of power equipment. Therefore, LCC estimation methods and models are not invariable, and different estimation algorithms are needed according to characteristics of collected historical data. Particularly, it concluded characteristics of different estimation methods, as shown in Table 1.

The focus of this part is to analyze and evaluate the most suitable and accurate LCC evaluation method according to different equipment categories, different equipment life stages and different data conditions.

In big data era, LCC estimated model at technical level can be combined with the deep learning framework to forecast, should not satisfy with the traditional estimation method, cyber-physical system and knowledge graph is the future of artificial intelligence in an important direction in big data analysis scenarios, the LCC estimation in the new method of artificial intelligence can be used on innovation breakthrough.

LCC Estimation Model of Power Assets According to cost estimation model of different life cycle stages of power assets, LCC of power assets is estimated. The following is the detailed scheme of LCC cost structure separation to point out the data path and separation ideas, and the LCC cost reduction formula for a supplementary explanation. The cost mapping relationship of each cycle in cost collection and estimation is explained. As for the content of mutual diffraction in the cost of each life cycle, it is a reasonable method to establish the standard of cost calculation system within the stage to solve the superposition effect of cost. The content of LCC cost estimation standard is introduced as follows.

Investment Costs The cost of investment and construction from start of planned construction to formal operation (excluding subsequent technical reform) mainly includes:

CI = Cpurchase +Cinstallation +Cconstruction +Cfield service (7)

where Cpurchase denotes purchase cost; Cinstallation represents installation cost; Cconstruction means construction cost; and Cfield service stands for field service cost.

Operating Costs Operating costs can be expressed as follows:

CO = Cenergy +Cduty (8)

where Cenergy represents energy cost; and Cduty denotes on- duty cost.

Frontiers in Energy Research | www.frontiersin.org 9 May 2021 | Volume 9 | Article 680355

fenrg-09-680355 April 30, 2021 Time: 20:26 # 10

Wang et al. Full Life Cycle Management Review

TABLE 1 | Summary of characteristics of different estimation methods.

Methods Uncertainty Application phase Accuracy Historical data

Parametric estimation method Medium Early stage Medium Many

Engineering estimation method Low Mid-late stage High Many

Analogy estimation method High Early stage Low Average

Neural Network method medium early stage high many

Expert estimation method High Early stage Low Few

Blind number theory estimation method Medium Early stage Low Fewer

Maintenance Costs Maintenance costs can be computed by:

CM = Clabor(CM) +Csupply(CM) +Cspare(CM) (9)

where Clabor represents labor costs; and Csupply denotes supply cost; and Cspare means spare cost.

Fault Costs Fault costs can be expressed by:

CF = Cblackout +Creplace +Cdeficiency +Cwithdrawal (10)

where Cblackout means power blackout cost; Creplace denotes equipment replace cost; Cdeficiency represents deficiency cost; and Cwithdrawal means withdrawal cost.

Discard Costs Discard costs can be computed by:

CD = Cscrap +Cresidual value (11)

where Cscrap means scrap cost; and Cresidual value denotes residual value.

Recoverable costs can be estimated by parameter estimation: residual value is approximately equivalent to the product of the weight of steel and its price, or the original value of the equipment is added with a proportional coefficient.

Asset management system of project cost detail work, there are detailed classification is too rough form, it fails to meet the project cost to carry out the elaboration to the equipment and component levels of business requirements, therefore the present solutions for the project was obtained from the financial management system of financial course code and cost detail, after the model to further improve the account of the project cost, for the enterprise of the management system of data interface, after the export project cost detail, project data cleansing, obtain the LCC cost collects the required at all levels in different stages of the data.

Reliability Evaluation of LCC Estimation Method In the construction of the theoretical model, must be based on historical data, using the total life cycle cost calculation model, calculation and analysis on specific historical node balance data, and combining with the investment cost is the subsequent comprehensive analysis comparison, the calculation model of correction methods, to ensure the effectiveness and

reliability of the final model in the actual production. That is, with the application of the system, the model algorithm can be modified in real time by comparing the difference between the actual cost accounting and the data platform estimation model.

The model constructed by historical data predicts the situation of the plant or similar projects in the same area in the same period, and compares the actual data to verify whether the error between the calculation results of the model and the actual results is within a certain allowable range. If so, the model is accurate. On the contrary, it is necessary to consider revising or even changing the model, and study two ways of revising the model based on the revision method of years and the revision method based on economic parameters.

The cost structure split based on the whole life cycle management method basically includes the economic investment generated in equipment management. The LCC management method implemented around the cost structure needs to be adjusted in conjunction with the data foundation and the characteristics of the equipment object, and the cost structure is increased or decreased and optimized when necessary to ensure that the economic indicators of the equipment are comprehensively and systematically considered in the implementation of the LCC management.

CONCLUSION

Combined with LCC analysis of power assets, full LCC management method is adopted for peak-load and frequency regulation enterprises, which can’t only be adopted for project cost budget, but also better guide a safe production and reliable operation, such as procurement, maintenance and scrap planning, as well as risk control.

In the application of full life cycle management, necessary data and information should be obtained to decompose each stage according to different costs, calculate appropriate estimation model, and collect from upper-level step by step to estimate final full LCC. Considering the time value of the capital, future-oriented full LCC needs to be converted into current- oriented value to facilitate the comparison and analysis of different schemes.

Since the transparency of data and information in different stages of full life cycle is different, various estimation algorithms should be carefully selected according to the characteristics of each period. At the same time, there are errors in estimation,

Frontiers in Energy Research | www.frontiersin.org 10 May 2021 | Volume 9 | Article 680355

fenrg-09-680355 April 30, 2021 Time: 20:26 # 11

Wang et al. Full Life Cycle Management Review

so the value of such uncertainty needs to be recognized. In the research, it is necessary to focus on the important systems and equipment, take the 20–80 principle as the analysis standard, and reasonably consider the research scope and precision of the equipment cost. The equipment content of asset management should be classified as primary and secondary in the classification of the equipment tree, and the importance level of the equipment should be divided. The equipment content with higher importance should focus on the evolution process of cost efficiency of LCC. The overall goal of power equipment cost model estimation is to have the lowest LCC of power equipment in the whole life cycle, and the goal of each stage is to have the lowest cost of each stage under the condition of meeting the total goal. Therefore, the target value of the consumption cost of each stage can be established to provide reference for the management mode of each stage of power equipment.

To sum up, the implementation of LCC needs to establish a complete database, a scientific decomposition mechanism, an appropriate estimation model and cooperation between various departments, which aims to improve accuracy of full LCC estimation and guide a satisfactory management of assets with a high assets efficiency.

Therefore, future research will focus more on improving the asset management level and the accuracy of the estimation model, and on this basis, consider a better combination of cost and efficiency. This is the next direction of work.

AUTHOR CONTRIBUTIONS

KW contributed to conceptualization. YL contributed to data curation and writing. XW contributed to formal analysis. NY contributed to funding acquisition. SY contributed to methodology. ZZ contributed to project administration. YW and TY contributed to resources. ZH contributed to visualization. All authors contributed to the article and approved the submitted version.

ACKNOWLEDGMENTS

The authors gratefully acknowledge the support of the analysis and research of power generation equipment management decision model based on LCC (STKJXM20190123).

REFERENCES Ammar, M., Zayed, T., and Moselhi, O. (2013). Fuzzy-based life-cycle cost model

for decision making under subjectivity. J. Constr. Eng. Manag. 139, 556–563. doi: 10.1061/(asce)co.1943-7862.0000576

Angelis, L., and Stamelos, I. (2000). A simulation tool for efficient analogy based cost estimation. Empir. Softw. Eng. 5, 35–68.

Arif, M., and Khan, M. E. (2010). Design and life cycle cost analysis of a SAPV system to electrify a rural area household in India. Curr. World Environ. 5, 101–106. doi: 10.12944/cwe.5.1.15

Asiedu, Y., and Gu, P. (1998). Product life cycle cost analysis: state of the art review. Int. J. Prod. Res. 36, 883–908. doi: 10.1080/002075498193444

Bajaj, A., Kekre, S., and Srinivasan, K. (2004). Cost and schedule performance in design and manufacturing management science. Manag. Sci. 50, 527–536. doi: 10.1287/mnsc.1030.0177

Basmadjian, R., and Meer, H. (2018). A heuristics-based policy to reduce the curtailment of solar-power generation empowered by energy-storage systems. Electronics 7, 349–370. doi: 10.3390/electronics7120349

Bastian, N. D. (2011). Optimizing army sustainability at fort bragg: a case study connecting life-cycle cost analysis with leadership in energy and environmental design for existing buildings. Eng. Manag. J. 23, 42–54. doi: 10.1080/10429247. 2011.11431894

Ben-Arieh, D., and Qian, L. (2003). Activity-based cost management for design and development stage. Int. J. Prod. Econ. 83, 169–183. doi: 10.1016/s0925-5273(02) 00323-7

Bierer, A., Götze, U., Meynerts, L., and Sygulla, R. (2015). Integrating life cycle costing and life cycle assessment using extended material flow cost accounting. J. Clean. Prod. 108, 1289–1301. doi: 10.1016/j.jclepro.2014. 08.036

Billinton, R., and Huang, D. (2010). Wind power modelling and the determination of capacity credit in an electric power system. Proc. Inst. Mech. Eng. 224, 1–9. doi: 10.1243/1748006xjrr266

Bird, L., Lew, D., Milligan, M., Carlini, E. M., Estanqueiro, A., and Flynn, D. (2016). Wind and solar energy curtailment: a review of international experience. Renew. Sustain. Energy Rev. 65, 577–586.

Cai, Y. Z., Liu, L., Cheng, H. Z., Ma, Z. L., and Zhu, Z. L. (2011). Application review of life cycle cost (LCC) technology in power system. Power Syst. Protect. Control 39, 149–154.

Chen, H., Cong, T. N., Yang, Y., Tan, C., Li, Y., and Ding, Y. (2009). Progress in electrical energy storage system: a critical review. Prog. Nat. Sci. 19, 291–312. doi: 10.1016/j.pnsc.2008.07.014

Chen, L., and Ren, J. (2018). Multi-attribute sustainability evaluation of alternative aviation fuels based on fuzzy ANP and fuzzy grey relational analysis. J. Air Trans. Manag. 68, 176–186. doi: 10.1016/j.jairtraman.2017.10.005

Cole, R. J., and Sterner, E. (2000). Reconciling theory and practice of life-cycle costing. Build. Res. Inform. 28, 368–375. doi: 10.1080/096132100418519

Curry, E. E. (1989). STEP: a tool for estimating avionics life cycle costs. Aerosp. Electron. Syst. Magazine IEEE. 4, 30–32. doi: 10.1109/62.16986

Danish, M. S. S., Yona, A., and Senjyu, T. (2014). Pre-design and life cycle cost analysis of a hybrid power system for rural and remote communities in Afghanistan. J. Eng. 2014, 438–444. doi: 10.1049/joe.2014.0172

de Jong, M., and Declercq, K. (2012). Economic evaluation of urban track systems: integration of life cycle costs and socio-economic assessment. Proc. Soc. Behav. Sci. 48, 1264–1273. doi: 10.1016/j.sbspro.2012.06.1102

Dui, X., Zhu, G., and Yao, L. (2018). Two-stage optimization of battery energy storage capacity to decrease wind power curtailment in grid-connected wind farms. IEEE Trans.Power Syst. 33, 3296–3305. doi: 10.1109/tpwrs.2017.2779134

Feng, Y., Lin, H., Ho, S. L., Yanc, J., Dong, J., and Fang, S. (2015). Overview of wind power generation in China: status and development. Renew. Sustain. Energy Rev. 50, 847–858.

Furch, J. (2016). A model for predicting motor vehicle life cycle cost and its verification. Trans. FAMENA. 40, 15–26.

Goh, B. H., and Sun, Y. (2016). The development of life-cycle costing for buildings. Build. Res. Inform. 44, 319–333.

Govil, K. K. (1984). New analytical models for logistics support cost and life cycle cost vs reliability function. Microelectron. Reliabil. 24, 61–63. doi: 10.1016/ 0026-2714(84)90638-3

Govil, K. K. (1985). Optimum design of reliable systems for specified life cycle cost. Microelectron. Reliabil. 25, 239–241. doi: 10.1016/0026-2714(85)90007-1

Hedley-Whyte, J. (2000). Standardization of interface design for medical devices: international electrotechnical commission and international organization for standardization medical alarm systems. Hum. Fact. Ergon. Soc. Annu. Meeting Proc. 44, 215–218. doi: 10.1177/154193120004400158

Hu, Y., and Cheng, H. (2013). Development and bottlenecks of renewable electricity generation in China: a critical review. Environ. Sci. Technol. 47, 3044–3056. doi: 10.1021/es303146q

Frontiers in Energy Research | www.frontiersin.org 11 May 2021 | Volume 9 | Article 680355

fenrg-09-680355 April 30, 2021 Time: 20:26 # 12

Wang et al. Full Life Cycle Management Review

Ilg, P., Scope, C., Muench, S., and Guenther, E. (2017). Uncertainty in life cycle costing for long-range infrastructure. Part I: leveling the playing field to address uncertainties. Int. J. Life Cycle Assess. 22, 277–292. doi: 10.1007/s11367-016- 1154-1

Ji, S. H., Park, M., and Lee, H. S. (2012). Case adaptation method of case-based reasoning for construction cost estimation in Korea. J. Construct. Eng. Manag. 138, 43–52. doi: 10.1061/(asce)co.1943-7862.0000409

John, E. (2017). Economic and technical challenges of flexible operations under large-scale variable renewable deployment. Energy Econ. 64, 363–372. doi: 10.1016/j.eneco.2017.04.012

Joseph, T., Ugalde-Loo, C. E., Liang, J., and Coventry, P. F. (2018). Asset management strategies for power electronic converters in transmission networks: application to HVDC and FACTS devices. IEEE Access. 6, 21084– 21102. doi: 10.1109/access.2018.2826360

Karamouz, M., Yaseri, K., and Nazif, S. (2017). Reliability-based assessment of lifecycle cost of urban water distribution infrastructures. J. Infrastruct. Syst. 23:04016030. doi: 10.1061/(asce)is.1943-555x.0000324

Karim, H., Magnusson, R., and Natanaelsson, K. (2012). Life-cycle cost analyses for road barriers. J. Trans. Eng. 138, 830–851. doi: 10.1061/(asce)te.1943-5436. 0000391

Kasis, A., Devane, E., Spanias, C., and Lestas, I. (2016). Primary frequency regulation with load-side participation part I: stability and optimality. IEEE Trans. Power Syst. 32, 3505–3518. doi: 10.1109/tpwrs.2016.2636286

Kim, G. T., Kim, K. T., Lee, D. H., Han, C. H., Kim, H. B., and Jun, J. T. (2010). Development of a life cycle cost estimate system for structures of light rail transit infrastructure. Autom. Construct. 19, 308–325. doi: 10.1016/j.autcon.2009.12. 001

Kiritsis, D., Neuendorf, K. P., and Xirouchakis, P. (1999). Petri net techniques for process planning cost estimation. Adv. Eng. Softw. 30, 375–387. doi: 10.1016/ s0965-9978(98)00126-4

Lee, J., Yang, H., Lim, J., Hong, T., Kim, J., and Jeong, K. (2020). BIM-based preliminary estimation method considering the life cycle cost for decision- making in the early design phase. J. Asian Arch. Build. Eng. 19, 384–399. doi: 10.1080/13467581.2020.1748635

Lee, S. Y., Jo, C., Bergan, P., Pettersen, B., and Chang, D. (2016). Life-cycle cost- based design procedure to determine the optimal environmental design load and target reliability in offshore installations. Struct. Saf. 59, 96–107. doi: 10.1016/j.strusafe.2015.12.002

Li, X., Chalvatzis, K. J., and Stephanides, P. (2018). Innovative energy islands: life-cycle cost-benefit analysis for battery energy storage. Sustainability 10, 1–19.

Liming, S., and Bo, Y. (2020). Nonlinear robust fractional-order control of battery/SMES hybrid energy storage systems. Power Syst. Protect. Control 48, 76–83.

Liu, H., Gopalkrishnan, V., Quynh, K. T. N., and Ng, W. (2008). Regression models for estimating product life cycle cost. J. Intell. Manuf. 20, 401–408. doi: 10.1007/s10845-008-0114-4

Liu, L., Wang, H., Cheng, H., and Liu, J. (2012). Economic evaluation of power systems based on life cycle cost. Autom. Electr. Power Syst. 36, 45–50. doi: 10.4324/9781315611730-12

Lombardi, L. (2003). Life cycle assessment comparison of technical solutions for CO2 emissions reduction in power generation. Energy Convers. Manag. 44, 93–108. doi: 10.1016/s0196-8904(02)00049-3

Luo, X., Li, L., and Wei, Z. (2011). Applications of life cycle cost theory in decision- making of investment for distribution transformers renovation. Power Syst. Technol. 35, 207–211.

Mascitelli, R. (2004). A simple cost evcluator for product design. Mach. Des. 76, 98–102.

Meyer, C., and De Doncker, R. W. (2006). LCC analysis of different resonant circuits and solid-state circuit breakers for medium-voltage grids. IEEE Trans. Power Deliv. 21, 1414–1420. doi: 10.1109/tpwrd.2005.861334

Miah, J. H., Koh, S. C. L., and Stone, D. (2017). A hybridised framework combining integrated methods for environmental life cycle assessment and life cycle costing. J. Clean. Produ. 168, 846–866. doi: 10.1016/j.jclepro.2017.08.187

Nilsson, J., and Bertling, L. (2007). Maintenance management of wind power systems using condition monitoring systems-life cycle cost analysis for two case studies. IEEE Trans. Energy Convers. 22, 223–229. doi: 10.1109/tec.2006.8 89623

Orfanos, N., Mitzelos, D., Sagani, A., and Dedoussis, V. (2019). Life- cycle environmental performance assessment of electricity generation and transmission systems in Greece. Renew. Energy 139, 1447–1462. doi: 10.1016/j. renene.2019.03.009

Ozbay, K., Jawad, D., Parker, N., and Hussain, S. (2004). Life-cycle cost analysis: state of the practice versus state of the art. Trans. Res. Record J. Trans. Res. Board 1864, 62–70. doi: 10.3141/1864-09

Özbayrak, M., Akgün, M., and Türker, A. K. (2004). Activity-based cost estimation in a push/pull advanced manufacturing system. Int. J. Prod. Econ. 87, 49–65. doi: 10.1016/s0925-5273(03)00067-7

Park, J., and Simpson, T. W. (2003). Production cost modeling to support product family design optimization. Math. Problems Eng. 2003, 165–174.

Plebankiewicz, E., Meszek, W., Zima, K., and Wieczorek, D. (2020). Probabilistic and fuzzy approaches for estimating the life cycle costs of buildings under conditions of exposure to risk. Sustainability 12, 226–236. doi: 10.3390/ su12010226

Savoretti, A., Mandolini, M., Raffaeli, R., and Germani, M. (2017). Analysis of the requirements of an early life-cycle cost estimation tool: an industrial survey. Procedia Manuf. 11, 1675–1683. doi: 10.1016/j.promfg.2017.07.291

Schneiderova-Heralova, R. (2018). Importance of life cycle costing for construction projects. Eng. Rural Dev. 17, 1223–1227.

Seo, K. K., Park, J. H., Jang, D. S., and Wallace, D. (2002). Approximate estimation of the product life cycle cost using artificial neural networks in conceptual design. Int. J. Adv. Manuf. Technol. 19, 461–471. doi: 10.1007/s001700200049

Shafiee, M., Brennan, F., and Espinosa, I. A. (2016). A parametric whole life cost model for offshore wind farms. Int. J. Life Cycle Assess. 21, 961–975. doi: 10.1007/s11367-016-1075-z

Shahata, K., and Zayed, T. (2013). Simulation-based life cycle cost modeling and maintenance plan for water mains. Struct. Infrastruct. Eng. 9, 403–415. doi: 10.1080/15732479.2011.552509

Shahidehpour, M., and Ferrero, R. (2005). Time management for assets: chronological strategies for power system asset management. IEEE Power Energy Magazine. 3, 32–38. doi: 10.1109/mpae.2005.1436498

Shi, J. N., Han, H. L., and Xu, T. (2009). Application of life cycle costs analysis in planning design of power transformation projects. Power Syst. Technol. 9, 63–66.

Solomon, R., Sandborn, P. A., and Pecht, M. G. (2000). Electronic part life cycle concepts and obsolescence forecasting. IEEE Trans. Compon. Packaging Technol. 23, 707–717. doi: 10.1109/6144.888857

Spertino, F., and Graditi, G. (2014). Power conditioning units in grid-connected photovoltaic systems: a comparison with different technologies and wide range of power ratings. Solar. Energy. 108, 219–229.

Steen, B. (2005). Environmental costs and benefits in life cycle costing. Manag. Environ. Qual. Int. J. 16, 107–118. doi: 10.1108/14777830510583128

Steinert, M. (2009). A dissensus. Technol. Forecast. Soc. Chang. 76, 291–300. Tian, Z., Jin, T., Wu, B., and Ding, F. (2011). Condition based maintenance

optimization for wind power generation systems under continuous monitoring. Renew. Energy. 36, 1502–1509. doi: 10.1016/j.renene.2010.10.028

Vahdat-Aboueshagh, H., Nazif, S., and Shahghasemi, E. (2014). Development of an algorithm for sustainability based assessment of reservoir life cycle cost using fuzzy theory. Water Resour. Manag. 28, 5389–5409. doi: 10.1007/s11269-014- 0808-7

Waghmode, L. Y., and Sahasrabudhe, A. D. (2012). Modelling maintenance and repair costs using stochastic point processes for life cycle costing of repairable systems. Int. J. Comput. Integr. Manuf. 25, 353–367. doi: 10.1080/0951192x. 2010.551783

Wang, H. S., Wang, Y. N., and Wang, Y. C. (2013). Cost estimation of plastic injection molding parts through integration of PSO and BP neural network. Exp. Syst. Appl. 40, 418–428. doi: 10.1016/j.eswa.2012.01.166

Wen, F., Hua, D., Wang, Q., and Singh, S. N. (2008). Wind power generation in china: present status and future prospects. Int. J. Energy Technol. Policy. 6, 736–744.

White, G. (1976). The life cycle cost of an item is the sum of all funds expended in support of the item from its conception and fabrication through its operation to the end of its useful life. Manag. Accounting 57, 39–42.

Woodward, D. G. (1997). Life cycle costing-theory, information acquisition and application. Int. J. Project Manag. 15, 335–344. doi: 10.1016/s0263-7863(96) 00089-0

Frontiers in Energy Research | www.frontiersin.org 12 May 2021 | Volume 9 | Article 680355

fenrg-09-680355 April 30, 2021 Time: 20:26 # 13

Wang et al. Full Life Cycle Management Review

Xu, W. R., and Wang, W. M. (2011). Summary and analysis of asset life cycle cost supported by the fixed grid asset operation and maintenance cost. East China Electric Power. 39, 1588–1591.

Yang, J., Liu, Q., Li, X., and Cui, X. (2017). Overview of wind power in China: status and future. Sustainability 9, 1–12. doi: 10.1002/9781118910054.ch1

Yeung, A. A., Yoho, K. D., and Arkes, J. (2013). Estimates of unit cost reductions of the F-16 fighter as a result of U.S. arms export production. J. Cost Anal. Parametr. 6, 3–22. doi: 10.1080/1941658x.2013.766547

Yildiz, B., and Kazimi, M. S. (2006). Efficiency of hydrogen production systems using alternative nuclear energy technologies. Int. J. Hydrogen Energy. 31, 77–92. doi: 10.1016/j.ijhydene.2005.02.009

Zakeri, B., and Syri, S. (2015). Electrical energy storage systems: a comparative life cycle cost analysis. Renew. Sustain. Energy Rev. 42, 569–596. doi: 10.1016/j.rser. 2014.10.011

Zhang, G., and Wang, W. (2012). The research of comprehensive evaluation model for thermal power equipment based on life cycle cost. Syst. Eng. Procedia 4, 68–78. doi: 10.1016/j.sepro.2011.11 .051

Zhang, Y., and Cai, M. (2014). Overall life cycle comprehensive assessment of pneumatic and electric actuator. Chin. J. Mech. Eng. 27, 584–594. doi: 10.3901/ cjme.2014.03.584

Conflict of Interest: KW, YL, XW, ZZ, and YW were employed by the company CSG POWER GENERATION CO., LTD.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2021 Wang, Li, Wang, Zhao, Yang, Yu, Wang, Huang and Yu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

Frontiers in Energy Research | www.frontiersin.org 13 May 2021 | Volume 9 | Article 680355

  • Full Life Cycle Management of Power System Integrated With Renewable Energy: Concepts, Developments and Perspectives
    • Introduction
      • Development and Consumption of New Energy
      • Organization of the Paper
    • Power Assets Full Life Cycle Cost Management
      • Asset Life-Cycle Cost Management
      • Overview of Power Equipment Life Cycle Management
    • Lcc Cost Estimation Model
      • Structure Composition of LCC
      • Estimation Method
        • Gray Fuzzy Estimation Method
        • Parameter Estimation Method
        • Engineering Estimation Method
        • Analogy Estimation Method
        • Neural Network Method
        • Activity-Based Costing Method
        • Case-Based Reasoning Method
        • Expert Estimation Method
        • Blind Number Theory Estimation Method
      • Comparison of Various Estimation Methods
      • LCC Estimation Model of Power Assets
        • Investment Costs
        • Operating Costs
        • Maintenance Costs
        • Fault Costs
        • Discard Costs
      • Reliability Evaluation of LCC Estimation Method
    • Conclusion
    • Author Contributions
    • Acknowledgments
    • References

Florida Renewable Energy 3.PDF

ORIGINAL RESEARCH published: 10 June 2021

doi: 10.3389/fenrg.2021.671279

Frontiers in Energy Research | www.frontiersin.org 1 June 2021 | Volume 9 | Article 671279

Edited by:

Yulong Ding,

University of Birmingham,

United Kingdom

Reviewed by:

Daofan Cao,

University of Birmingham,

United Kingdom

Zac Cesaro,

University of Oxford, United Kingdom

*Correspondence:

Mathias Berger

[email protected]

Specialty section:

This article was submitted to

Process and Energy Systems

Engineering,

a section of the journal

Frontiers in Energy Research

Received: 23 February 2021

Accepted: 19 April 2021

Published: 10 June 2021

Citation:

Berger M, Radu D, Detienne G,

Deschuyteneer T, Richel A and Ernst D

(2021) Remote Renewable Hubs for

Carbon-Neutral Synthetic Fuel

Production.

Front. Energy Res. 9:671279.

doi: 10.3389/fenrg.2021.671279

Remote Renewable Hubs for Carbon-Neutral Synthetic Fuel Production Mathias Berger1*, David Radu1, Ghislain Detienne2, Thierry Deschuyteneer2,

Aurore Richel3 and Damien Ernst1

1 Department of Electrical Engineering and Computer Science, University of Liège, Liège, Belgium, 2 Fluxys SA, Brussels,

Belgium, 3 Laboratory of Biomass and Green Technologies, Gembloux Agro-Bio Tech - University of Liège, Gembloux,

Belgium

This paper studies the economics of carbon-neutral synthetic fuel production from

renewable electricity in remote areas where high-quality renewable resources are

abundant. To this end, a graph-based optimisation modelling framework directly

applicable to the strategic planning of remote renewable energy supply chains is

proposed. More precisely, a hypergraph abstraction of planning problems is introduced,

wherein nodes can be viewed as optimisation subproblems with their own parameters,

variables, constraints and local objective. Nodes typically represent a subsystem such

as a technology, a plant or a process. Hyperedges, on the other hand, express the

connectivity between subsystems. The framework is leveraged to study the economics of

carbon-neutral synthetic methane production from solar and wind energy in North Africa

and its delivery to Northwestern European markets. The full supply chain is modelled in an

integrated fashion, which makes it possible to accurately capture the interaction between

various technologies on an hourly time scale. Results suggest that the cost of synthetic

methane production and delivery would be slightly under 150 e/MWh (higher heating

value) by 2030 for a system supplying 10 TWh annually and relying on a combination

of solar photovoltaic and wind power plants, assuming a uniform weighted average

cost of capital of 7%. A comprehensive sensitivity analysis is also carried out in order

to assess the impact of various techno-economic parameters and assumptions on

synthetic methane cost, including the availability of wind power plants, the investment

costs of electrolysis, methanation and direct air capture plants, their operational flexibility,

the energy consumption of direct air capture plants, and financing costs. The most

expensive configuration (around 200 e/MWh) relies on solar photovoltaic power plants

alone, while the cheapest configuration (around 88 e/MWh) makes use of a combination

of solar PV and wind power plants and is obtained when financing costs are set to zero.

Keywords: optimisation, renewable energy, carbon neutral, synthetic fuels, remote supply chain, linear

programming, structured models, graph

Berger et al. Renewable Hubs for Fuel Production

1. INTRODUCTION

Electricity generation from renewable resources combined with wide-ranging electrification has been a mainstay of European climate and energy policies, with the primary goal of decarbonising the power sector as well as other carbon- intensive sectors.

Major obstacles to such endeavours have nevertheless surfaced in recent years. Firstly, sectors like aviation, shipping, heating or industry have proved difficult to fully electrify. Indeed, feedstocks and energy carriers with specific properties such as a high energy density are typically required (Eveloy et al., 2021). Hence, the production of carbon-neutral synthetic fuels and feedstocks from renewable electricity has been the focus of a growing body of literature. For example, the synthesis of carbon-neutral hydrogen (Borgschulte, 2016), methane (Biswas et al., 2020), methanol (Centi et al., 2020), and ammonia (Ghavam et al., 2021) have all been considered. A number of demonstration projects have been carried out as well (Wulf et al., 2020). Secondly, it has become clear that the technical renewable potential of some European countries (i.e., the maximum amount of renewable electricity that may be produced within a country’s borders and exclusive economic zone, while accounting for a variety of land eligibility constraints Ryberg et al., 2018) is insufficient to supply current energy demand levels (e.g., in densely-populated countries like Belgium Berger et al., 2020; Limpens et al., 2020 or the United Kingdom MacKay, 2008). It is still unclear whether pooling renewable resources at the European level would alleviate the problem. On the other hand, it is well documented that social acceptance issues tend to compound it (Segreto et al., 2020).

A simple solution consists in harvesting renewable resources in remote areas where they are abundant, synthesising carbon- neutral fuels or feedstocks using renewable electricity and transporting them back to demand centres (Fasihi and Bogdanov, 2015; Chapman et al., 2017; Heuser et al., 2019). However, two conditions must be satisfied for such an approach to be worth pursuing. Firstly, transport should be energy-efficient and cost-effective. This will often depend on the physics of the commodity considered and the maturity of technologies available to handle it. Secondly, very-high-quality renewable resources should be tapped. The quality of such resources is typically estimated via the annual capacity factor of a given technology harnessing them, which directly reflects the amount of electricity that may be produced per unit capacity. Since renewable power generation technologies usually have very low operating costs, the higher the capacity factor, the lower the electricity cost. Regions with outstanding resources and vast technical potential include Patagonia (wind) (Heuser et al., 2019), North Africa (sun and wind) (Fasihi and Bogdanov, 2015), and Greenland (wind) (Radu et al., 2019). Providing an accurate quantitative assessment of the economics and efficiency of such remote renewable energy supply chains and pathways is critical to evaluate future sustainable energy supply options available to policy makers and society at large as well as to identify where to direct future research and development efforts.

From a conceptual standpoint, a supply chain can be viewed as a networked system composed of dynamical subsystems

interacting with each other. In order to tackle the problem at hand, the collection of processes and technologies forming a remote renewable energy supply chain must be analysed in an integrated fashion, which makes it possible to properly capture the interactions between subsystems in space and time. In addition, a sufficient level of technical detail and temporal resolution should be used to properly model their operation (Poncelet et al., 2016). This paper formalises these considerations and proposes a graph-based optimisation modelling framework directly applicable to the strategic planning and analysis of remote renewable energy supply chains. More precisely, a hypergraph abstraction of planning problems is introduced, wherein nodes can be viewed as optimisation subproblems with their own parameters, variables, constraints and local objective, and typically represent a subsystem such as a technology, a plant or a process. Hyperedges, on the other hand, express the connectivity between subsystems. The framework is then leveraged to study the economics of carbon-neutral synthetic methane production from renewable electricity and atmospheric carbon dioxide in North Africa and its delivery to Northwestern European markets. Synthetic methane is an appealing carbon-neutral energy carrier, as some downstream transport infrastructure is readily available in Northwestern European countries, and the liquefied methane chain is mature and cost-effective (Timera Energy, 2018). It would also obviate the need for replacing or upgrading appliances and processes presently used for residential heating and in industry that a switch to other fuels would entail. In this paper, the carbon-neutral synthetic methane supply chain is modelled end-to-end, from power generation in North Africa to methane regasification in Northwestern Europe. A detailed description of each process and technology is provided, along with comprehensive data resources. The modelling framework also served as a basis for the development of an open source optimisation modelling language (Berger et al., 2021) and tool (Miftari et al., 2021). In the interest of transparency, the input files and full data enabling others to reproduce the analyses presented in this paper are also available in the associated repository (Miftari et al., 2021).

This paper is structured as follows. Section 2 reviews the relevant literature. Section 3 details the proposed modelling framework, while Sections 4 and 5 describe the case study and discuss results, respectively. Finally, Section 6 concludes the paper and discusses future work directions.

2. LITERATURE REVIEW

To the best of the authors’ knowledge, Hashimoto et al. (1999) were the first to suggest the production of hydrogen from renewable electricity in remote areas followed by the synthesis of hydrocarbons using captured carbon dioxide as a means of producing carbon-neutral fuels. The paper, however, did not provide a quantitative techno-economic analysis of the proposed supply chains. By contrast, Zeman and Keith (2008) performed one of the first quantitative economic analyses of carbon-neutral synthetic fuel production using carbon-neutral hydrogen and atmospheric carbon dioxide. Production cost estimates for this

Frontiers in Energy Research | www.frontiersin.org 2 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

route were found to be between 23.5 and 30.0 US$/GJ (which would roughly correspond to 74.1 and 94.6 e/MWh, using the 2020 average exchange rate of $1.142 for 1.0e). The production of carbon-neutral synthetic methane and liquid fuels in remote areas with abundant renewable resources has been considered in Fasihi and Bogdanov (2015) and Fasihi et al. (2017), respectively. In the first study, the authors estimate that the cost of producing synthetic methane from renewable electricity in North Africa (specifically in central and southern Algeria) and delivering it to Japan could be around 65–75 e/MWh by 2030 for a hybrid solar-wind system, assuming a uniform weighted average cost of capital (WACC) of 7%. It is not specified whether the higher heating value (HHV) or the lower heating value (LHV) of methane was used to compute these costs. In the second study, the cost of producing synthetic methane in the same region and delivering it to Finland is found to be between 100 and 110 e/MWh (HHV) by 2030 and between 90 and 100 e/MWh (HHV) by 2040, respectively, using a WACC of 7%. Finally, the economics of carbon-neutral fuel production is also analysed in Agora Verkehrswende et al. (2018). Cost estimates close to 140– 150 e/MWh (LHV) by 2030 and 110 e/MWh (LHV) by 2050 (using a WACC of 6% in both cases) are found for synthetic methane production in North Africa (specifically in central and southern Algeria) and delivery to Germany based on both solar energy alone and hybrid systems combining solar and wind power plants.

It is also informative to review the modelling approaches followed in these studies. Firstly, Zeman and Keith (2008) do not specify the technologies used to implement the various conversion processes, and instead rely on a set of assumptions about conversion efficiencies and the cost of producing input commodities (in stoichiometric proportions) to come up with a cost estimate for the final product. Then, Fasihi and Bogdanov (2015) resort to a so-called annual-basis model estimating the annual number of equivalent full load hours of renewable power production in order to calculate electricity and synthetic methane costs based on a set of techno-economic assumptions. This method is equivalent to estimating annual power production and costs using an average capacity factor value, and the model is therefore not temporally-resolved. A so-called hourly-basis model enabling the sizing of solar photovoltaic (PV) and wind power plants is mentioned in Fasihi and Bogdanov (2015) and Fasihi et al. (2017), but no mathematical model is explicitly described and no computer code implementing it is made available, which makes the approach difficult to interpret and scrutinise. Somewhat surprisingly, very minor differences in cost estimates are observed between the annual-basis and hourly-basis models in Fasihi and Bogdanov (2015). In Agora Verkehrswende et al. (2018), an annual full load hour model similar to that of Fasihi and Bogdanov (2015) is used. For systems driven by variable renewable energy resources, it has been shown that using a high temporal resolution (e.g., hourly) and adopting a proper level of technical detail (i.e., representing the flexibility of technologies, or lack thereof) is key for accurately sizing plants and estimating both investment and operating costs properly (Poncelet et al., 2016). It is worth noting that the aforementioned papers rely on models that have both a very low level of technical

detail and a very low temporal resolution. Furthermore, none of these models makes it possible to design the supply chain in an integrated fashion while properly accounting for interactions between subsystems. Similar shortcomings can be found in studies focussing on other energy carriers such as hydrogen (Dagdougui, 2012; Heuser et al., 2019).

The design and analysis of energy systems and supply chains has often been tackled using mathematical programming techniques in the literature (Garcia and You, 2015; Conejo et al., 2016). Different classes of models may be used, ranging from linear and mixed-integer linear programs (LPs and MILPs) to non-linear and mixed-integer (possibly non-convex) non-linear programs (NLPs and MINLPs) (Biegler and Grossmann, 2004). Parameter uncertainty may also be taken into account (Sahinidis, 2004). The type of model used typically depends on the research scope, the available computational resources and the data at hand. For example, the design of a single piece of equipment used in a process may require NLP or MINLP models to accurately represent its physics and operating modes (Grossmann, 2002). On the other hand, supply chains can be viewed as collections of interconnected plants or processes, which themselves rely on a variety of complex pieces of equipment. Representing each of them in their full complexity would require vast amounts of data and result in intractable models. Thus, for the purpose of strategic or high-level system design analyses, aggregate plant models are typically employed (Chen and Grossmann, 2017; Montastruc et al., 2019). In such models, mass and energy conservation laws are enforced at plant level while accounting for basic operational constraints. Mass and energy balances are also enforced between interconnected plants in order to guarantee consistency of flows at system level. Such approaches, which usually rely on LP or MILP models, have for instance been applied to the design of integrated biorefineries (Kokossis et al., 2015), the design of power-to-syngas processes (Maggi et al., 2020) and power-to- chemicals networks (Schack et al., 2016; Liesche et al., 2019). Such an approach is adopted in this paper, as discussed next.

3. METHODOLOGY

This section formally introduces the abstract graph-based optimisation modelling framework and describes how it can be applied in the context of strategic energy supply chain planning.

3.1. Graph-Based Optimisation Modelling Framework In this paper, supply chain planning problems are formulated as structured linear programs. These problems typically involve the optimisation of discrete-time dynamical systems over a finite time horizon and exhibit a natural block structure that may be encoded by a sparse graph or hypergraph (Gallo et al., 1993). A hypergraph abstraction is therefore employed to represent them, wherein nodes model optimisation subproblems, while hyperedges express the relationships between nodes. A global discretised time horizon and associated set of time periods common to all nodes are also defined. Each node is equipped with a set of so-called internal and external (or coupling) variables.

Frontiers in Energy Research | www.frontiersin.org 3 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

A set of constraints is also defined for each node, along with a local objective function representing its contribution to a system-wide objective. Finally, for each hyperedge, constraints involving the coupling variables of the nodes to which it is incident are defined. In the following paragraphs, we formally define variables, constraints, objectives and formulate the abstract model that encapsulates the class of problems considered.

Let T ∈ N be the time horizon considered, let T = {0, 1, . . . , T − 1} be the associated set of time periods, and let G = (N, E) be a (possibly directed) hypergraph encoding the block structure of the problem considered, with node set N and hyperedge set E ⊆ 2N (i.e., each hyperedge corresponds to a subset of nodes). Let Xn ∈ Xn and Zn ∈ Zn denote the collection of internal and coupling variables defined at node n ∈ N. Note that all variables are assumed to take values in continuous sets (i.e., Xn and Zn are continuous). In addition, for any hyperedge e ∈ E, let Ze = {Zn|n ∈ e} denote the collection of coupling variables associated with the nodes to which this hyperedge is incident.

Let Fn denote the function defining the local objective at node n ∈ N. In this paper, we consider scalar objectives of the form

Fn(Xn, Zn) = f n0 (X n, Zn) +

∑

t∈T

f n(Xn, Zn, t), (1)

where f n0 and f n are (scalar) affine functions of Xn and Zn.

Both equality and inequality constraints may be defined at each node n ∈ N. More precisely, an arbitrary number of constraints that can each be expanded over a subset of time periods may be defined. Hence, we consider equality constraints of the form

hn k (Xn, Zn, t) = 0, ∀t ∈ Tn

k , (2)

with (scalar) affine functions hn k and index sets Tn

k ⊆ T, k =

1, . . . , Kn, as well as inequality constraints

gn k (Xn, Zn, t) ≤ 0, ∀t ∈ T̄

n k, (3)

with (scalar) affine functions gn k and index sets T̄

n k ⊆ T, k =

1, . . . , K̄n. Likewise, both equality and inequality constraints may be

defined over any hyperedge e ∈ E. These constraints, however, can only involve the coupling variables of the nodes to which hyperedge e ∈ E is incident (i.e., nodes such that n ∈ e). More precisely, let He and Ge be affine functions of Ze used to define the equality and inequality constraints associated with a given hyperedge e ∈ E.

Using this notation, the class of problems that can be represented in this framework reads

min ∑

n∈N F n(Xn, Zn)

s.t. hn k (Xn, Zn, t) = 0, ∀t ∈ Tn

k , k = 1, . . . Kn, ∀n ∈ N

gn k (Xn, Zn, t) ≤ 0, ∀t ∈ T̄

n k, k = 1, . . . K̄

n, ∀n ∈ N

He(Ze) = 0, ∀e ∈ E

Ge(Ze) ≤ 0, ∀e ∈ E

Xn ∈ Xn, Zn ∈ Zn, ∀n ∈ N.

(4)

Figure 1 schematically illustrates the class of problems that can be modelled in this framework.

3.2. Application to Energy Supply Chains The framework presented in Section 3.1 can be readily leveraged to model energy systems and supply chains. In this case, nodes typically represent a technology, a plant or a process, while hyperedges may be used to enforce some coupling between plants. Introducing a few generic (parametrised) nodes and hyperedges often suffices to model a broad range of system configurations. In the following, some key modelling assumptions are introduced, along with two generic nodes and one generic hyperedge, namely conversion and storage nodes, as well as conservation hyperedges.

3.2.1. Modelling Assumptions Central Planning and Operation. Investment decisions are made by a single entity that also operates the system, and whose goal is to minimise total system costs.

Perfect Foresight and Knowledge. The entity planning and operating the system has perfect foresight and knowledge, that is, future weather events and demand patterns, as well as all technical and economic parameters are assumed to be known with certainty.

Investment and Operational Decisions. A static investment model is used, whereby investment decisions are made at the beginning of the time horizon and assets are immediately available. Operational decisions are made at hourly time steps. The investment and operational problems are solved simultaneously.

Technology and Process Models. The sizing and operation of technologies are modelled via a set of affine input-output relations that typically express mass and energy balances at plant or process level. Input or output dynamics are considered for some technologies, but only storage technologies have a simple state space representation.

3.2.2. Nodes Preliminaries. In the following developments, Latin letters denote optimisation variables and indices, while Greek letters indicate parameters.

Conversion. Let n ∈ N be a node representing a so- called conversion technology that processes a set of commodities (e.g., an electrolysis plant that splits water into hydrogen and oxygen using an electric current and therefore processes four commodities). Commodity flows are modelled as external variables, and an index i ∈ In is assigned to each commodity (i.e., each i ∈ In corresponds to a time-indexed external variable). The processing of commodities by technology n is modelled via a set of linear equations linking the flow of a reference commodity r ∈ In (e.g., hydrogen may be taken as the reference commodity for electrolysis plants) to the flow of all other commodities i ∈ In \ {r}, which read

qnrt − φ n i q

n i(t+τ ni )

= 0, ∀i ∈ In \ {r}, ∀t ∈ Tn, (5)

where qnit ∈ R+ represents the flow of commodity i at time t, φni ∈ R+ is the so-called conversion factor between commodity r

Frontiers in Energy Research | www.frontiersin.org 4 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

FIGURE 1 | Hypergraph abstraction of a hypothetical problem whose block structure is represented by four nodes (i.e., N = {n1, n2, n3, n4}) and two hyperedges (i.e.,

E = {e1, e2}, with e1 = {n1, n2} and e2 = {n2, n3, n4}). Note that e1 only has equality constraints while e2 only has inequality constraints.

and i (which may be derived, e.g., from stoichiometric coefficients or the enthalpy of the underlying reaction), while τ ni ∈ N is the amount of time that may be required for the conversion process to take place and Tn ⊆ T is a suitable subset of time periods. The capacity of a technology is typically modelled as an internal variable and defined as the maximum flow of a reference commodity r′ ∈ In according to which the technology is sized. Note that r′ may be different from r (e.g., the size of electrolysis plants is typically expressed in terms of their electrical capacity, although hydrogen may be the reference commodity used in Equation (5)). Since a static investment model is considered, capacity deployments occur at the beginning of the time horizon and remain constant throughout, i.e.,

Kn0 − K n t = 0, ∀t ∈ T \ {0}, (6)

where Knt ∈ R+ denotes the new capacity of technology n. In the following, Kn will be used as shorthand for Kn0 . Thus, the total capacity of technology n is defined via

qnr′t − π n t (κ

n + Kn) ≤ 0, ∀t ∈ T, (7)

where πnt ∈ [0, 1] indicates the availability of technology n at time t and κn ∈ R+ represents the existing capacity. The so- called availability parameter πnt may for instance represent the instantaneous capacity factor of a renewable power plant. The maximum capacity of a technology may be bounded, which leads to the introduction of an additional constraint,

(κn + Kn) − κ̄n ≤ 0, (8)

with κ̄n ∈ R+ the maximum capacity of technology n that may be installed. A variety of operational constraints may also

be considered. For instance, some conversion technologies may have a limited operating range, and may only work if a minimum flow of commodity i ∈ In is maintained, which can be expressed as

µ n(κn + Kn) −

φni

φn r′ qnit ≤ 0, ∀t ∈ T, (9)

where µn ∈ [0, 1] represents the minimum operating level (as a fraction of the installed capacity). Since the technology is sized with respect to the flow of commodity r′, the flow of a commodity i 6= r′ must be scaled by the ratio of conversion factors in Equation (9). The rate at which the flow of commodity i ∈ In can vary may also be limited, leading to the introduction of so-called ramping constraints,

φni

φn r′ (qnit − q

n i(t−1)

) − 1ni,+(κ n + Kn) ≤ 0, ∀t ∈ T \ {0}, (10)

and

φni

φn r′ (qn

i(t−1) − qnit) − 1

n i,−(κ

n + Kn) ≤ 0, ∀t ∈ T \ {0}, (11)

with 1ni,+ ∈ [0, 1] and 1 n i,− ∈ [0, 1] the maximum rates at which

flows can be ramped up and down (as a fraction of the installed capacity per unit time), respectively. Finally, the local objective function associated with this node reads

Fn = ν(ζ n + θ

n f )Kn +

∑

t∈T

θ n t,vq

n r′tδt, (12)

Frontiers in Energy Research | www.frontiersin.org 5 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

where ν ∈ N is the number of years spanned by the optimisation horizon, ζ n ∈ R+ represents the (annualised) investment cost (also known as capital expenditure, CAPEX), θn

f ∈ R+ models

fixed operation and maintenance (FOM) costs, θnt,v ∈ R+ represents variable operation and maintenance (VOM) costs, which may be time-dependent, and δt ∈ R+ is the duration of each time period.

Storage. Let n ∈ N be a node representing a storage technology. A storage technology is assumed to hold one commodity, although its operation may involve other commodities (e.g., a compressed hydrogen storage system stores hydrogen but requires electricity to drive compressors). The inventory level of the storage system is defined as an internal variable, while the charge and discharge flows are defined as external variables, respectively. Let i ∈ In and j ∈ In be the indices of the in/outflows of the commodity stored in technology n, respectively. Then, the basic equation governing the operation of storage systems describes the inventory level dynamics and reads

ent+1 − (1 − η n S)e

n t − η

n +q

n it +

1

ηn− qnjt = 0, ∀t ∈ T \ {T − 1}, (13)

where ent ∈ R+ is the inventory level at time t, q n it ∈ R+

and qnjt ∈ R+ represent commodity in- and outflows at time t,

respectively, ηnS ∈ [0, 1] is the self-discharge rate, η n + ∈ [0, 1] is

the charge efficiency and ηn− ∈ [0, 1] is the discharge efficiency. Charging a storage system may also require the consumption of another commodity l ∈ In, l 6= i, j (e.g., electricity consumed by compressors), which is typically modelled via an additional external variable qn

lt ∈ R+ and equations

qn lt − φ

n i q

n it = 0, ∀t ∈ T. (14)

In order to avoid spurious transient effects in storage operation, inventory levels are typically required to be equal at the beginning and at the end of the optimisation horizon,

en0 − e n T−1 = 0. (15)

The stock capacity of the storage technology is modelled as an internal variable and it is defined by the maximum inventory level. Since a static investment model is used, the stock capacity is constant throughout the entire time horizon, i.e.,

En0 − E n t = 0, ∀t ∈ T \ {0}, (16)

where Ent ∈ R+ is the new capacity. In the following, E n will be

used as shorthand for En0. The total storage capacity is therefore defined via

ent − (ǫ n + En) ≤ 0, ∀t ∈ T, (17)

where ǫn ∈ R+ denotes the existing stock capacity. Note that the total stock capacity itself may be constrained,

(ǫn + En) − ǭn ≤ 0, (18)

with ǭn ∈ R+ the maximum stock capacity that may be deployed. In addition, some storage technologies may require a minimum inventory level to be maintained, which can be expressed as

σ n(ǫn + En) − ent ≤ 0, ∀t ∈ T, (19)

where σ n ∈ [0, 1] represents the minimum inventory level (as a fraction of the stock capacity). The maximum inflow capacity is sized independently of the stock capacity. For example, in the case of battery storage systems, this implies that the energy-to- power ratio is not fixed a priori. The maximum inflow is modelled using an internal variable that is also constant throughout the time horizon considered, as in Equation (6). It is defined as follows

qnit − (κ n + Kn) ≤ 0, ∀t ∈ T, (20)

where κn ∈ R+ denotes the existing flow capacity and K n ∈ R+

is used as shorthand for the new capacity. The maximum in- and outflows may be asymmetric, depending on the properties of the underlying technology, which is modelled via

qnjt − ρ n(κn + Kn) ≤ 0, ∀t ∈ T, (21)

where ρn ∈ R+ represents the maximum discharge-to-charge ratio. Finally, the local objective function associated with this node reads

Fn = [

ν(ςn + ϑn f )En +

∑

t∈T

ϑ n t,ve

n t δt

]

+

[

ν(ζ n + θn f )Kn+

∑

t∈T

θ n t,vq

n itδt

]

. (22)

where ςn ∈ R+ and ζ n ∈ R+ represent the stock and flow

components of CAPEX, ϑn f

∈ R+ and θ n f

∈ R+ model the

stock and flow components of FOM costs, while ϑnt,v ∈ R+ and θnt,v ∈ R+ represent the stock and flow components of VOM costs, which may be time-dependent.

3.2.3. Hyperedges Conservation. Let e ∈ E be a so-called conservation hyperedge that enforces local flow conservation of some commodity. More precisely, one commodity is associated with a given conservation hyperedge. Let i ∈ ∩n∈eI

n denote this commodity (i.e., each node n ∈ e has an external variable representing a flow of commodity i). In addition, let us assume that hyperedge e is directed (i.e., it can be partitioned into two disjoint subsets eT and eH that are called its tail and head, respectively). Roughly speaking, e can be interpreted as “going from nodes in eT to nodes in eH”. Then, flow conservation of commodity i over hyperedge e can simply be expressed as

∑

n∈eT

qnit − ∑

n∈eH

qnit − λ e t = 0, ∀t ∈ T, (23)

where the first two sums on the left-hand side represent the aggregate flows from the nodes in eT and eH (whose signs depend

Frontiers in Energy Research | www.frontiersin.org 6 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

on the orientation of e), while λet ∈ R represents exogenous withdrawals or injections that may take place over e at each time period t. Note that Equation (23) may sometimes be relaxed to a greater-than-or-equal-to inequality constraint (in which case the net flow must exceed the injections/withdrawals at each time period).

3.3. Implementation The graph-based modelling framework discussed in Section 3.1 has been used as a basis for developing an optimisation modelling language for structured linear and mixed-integer linear programs called the graph-based optimisation modelling language (GBOML) (Berger et al., 2021). The language blends elements of both algebraic (Kallrath, 2012) and object-oriented (Schichl, 2004) modelling languages in order to facilitate problem encoding and post-processing, promote model re-use and improve portability. The full description of GBOML, which is beyond the scope of this paper, is detailed in a separate tutorial paper (Berger et al., 2021). A parser for GBOML, called the GBOML compiler, has also been implemented in Python 3.8 (using the PLY library), and is released as open source software (Miftari et al., 2021). The GBOML compiler directly interfaces with both commercial and open source LP and MILP solvers (namely Gurobi, CPLEX, and Clp/Cbc), enabling users to model problems, interact with solver APIs, query solutions and retrieve post-processed results in an integrated fashion. For the sake of transparency and reproducibility, the input file that encodes the model and full data allowing one to reproduce the case study and results discussed in Sections 4 and 5 are also provided in the GBOML repository (Miftari et al., 2021). One instance of the resulting linear programming model can be solved in about ten minutes with the homogeneous barrier algorithm (cross-over disabled) of Gurobi 9.1.1 on a laptop with 16 GB of RAM and a Quad-Core Intel i7 processor clocking at 2.6 GHz.

4. CASE STUDY

This case study aims to analyse the economics of producing carbon-neutral methane from renewable electricity in areas of North Africa where abundant and high-quality renewable resources are readily available, and exporting it to Northwestern European markets. More specifically, the entire supply chain is modelled and optimised in an integrated fashion over a time horizon of five years with hourly resolution (i.e., T = 43, 824, since 2016 is a leap year), from the remote generation of electricity to the synthesis and liquefaction of carbon- neutral methane in North Africa, to its eventual delivery and regasification at a Northwestern European gas terminal. Figure 2 shows a schematic representation of the supply chain considered.

4.1. System Configuration A more detailed representation of the system configuration considered in this study is shown in Figure 3, where icons correspond to conversion or storage nodes, while bullets and arrows schematically represent conservation hyperedges. For the sake of readability, the set of nodes is split into three clusters, which also correspond to the different geographical areas

FIGURE 2 | Remote carbon-neutral methane supply chain: electricity is

produced in a remote inland cluster in central Algeria and transported to a

coastal cluster where carbon-neutral methane is synthesised and liquefied for

export to Northwestern European markets.

displayed in Figure 2. The nodes and hyperedges used to model this system are described in the following subsections.

4.1.1. Conversion Nodes Conversion nodes are discussed in this subsection. Tables 1, 2 gather the techno-economic data (2030 estimates) used to model conversion nodes along with the original data sources and complement the descriptions below. In the model, power flows are measured in GW (GWh/h), energy is measured in GWh, mass flows are measured in kt/h, mass is measured in kt, and money is measured in Me.

Solar PV. Solar photovoltaic panels are used for power generation. The plants are modelled with one external variable representing the output power and one internal variable representing the plant capacity, respectively. Constraints (7) and (8) are used along with the local objective function (12). In order to construct the capacity factor time series πnt , five years (2015– 2019) of hourly-sampled irradiance data are retrieved from the ERA5 database (European Center for Medium Range Weather Forecasts (ECMWF), 2020) for each grey point in Figure 2 and converted into capacity factors using a generic transfer function (HOMER, 2020) and TrinaSolar Tallmax M tilted module data (TrinaSolar, 2017). Sites with a five-year average capacity factor value exceeding 24.5% are retained (eleven in total, shown by red crosses in Figure 2) and the associated time series are then aggregated (spatially averaged) into a single time series πnt , which is illustrated in Figure 4 for a set of weekly periods in 2016.

Wind Turbines. Wind turbines are used for generating power as well. Wind power plants are modelled in a fashion similar to solar PV plants, that is, with one external variable representing the power output and one internal variable representing the plant

Frontiers in Energy Research | www.frontiersin.org 7 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

FIGURE 3 | Remote hub system configuration. Icons represent conversion or storage nodes, while bullets and arrows schematically represent conservation

hyperedges.

capacity, respectively. Constraints (7) and (8) are used along with the local objective function (12). In order to construct the capacity factor time series πnt , five years (2015–2019) of hourly-sampled wind speed data are retrieved from the ERA5 database (European Center for Medium Range Weather Forecasts (ECMWF), 2020) for each grey point in Figure 2 and converted into capacity factors using the transfer function of the Vestas V90 turbine available in the windpowerlib library (Haas et al., 2019). Sites with a five-year average capacity factor value exceeding 50% are retained (five in total, shown by blue crosses in Figure 2) and the associated time series are then aggregated (spatially averaged) into a single time series πnt , which is also displayed in Figure 4.

HVDC Interconnection. Ultra-high-voltage direct current (HVDC) overhead lines (800 or 1, 100 kV) are assumed to be used for bulk power transmission from the first cluster (inland) to the second one (coastal hub) (CIGRE C1.35 Working Group, 2019). Note that the yellow area containing the solar and wind sites in Figure 2 is assumed to be a copper plate for the purpose of this study, which implies that solar PV and wind power plants feed directly into the electricity interconnection and the cost of the infrastructure connecting power plants to the HVDC interconnection is neglected. Voltage source

converters (VSC) are well-suited for remote applications, as they are self-commutated and are much more controllable than typical line-commutated (LC) alternatives, although they are more expensive and have higher conversion losses (Xiang et al., 2016). In this case, two VSC stations are placed on each side of an overhead HVDC cable whose length is assumed to be 1, 000 km. Losses in each converter station roughly amount to 1.8% of the power flowing through it, while approximately 1.5% of the power transiting through the HVDC cable is lost. Combining these figures yields the overall efficiency reported in Table 1. Economic data shown in Table 2 include the costs of both converter stations and the cable. The interconnection is modelled using two external variables and one internal variable. The external variables represent the power flows into (i.e., leaving the first cluster) and out of (i.e., reaching the second cluster) the HVDC interconnection, respectively. The internal variable is the capacity of the converter-line pair. Investment costs in lines and converter stations are accounted for in the local objective (12), along with operating costs.

Electrolysis Plants. Proton exchange membrane (also called polymer electrolyte membrane, PEM) electrolysis plants (Carmo et al., 2013) are used for producing hydrogen. This technology makes it possible to split water into hydrogen and oxygen by the

Frontiers in Energy Research | www.frontiersin.org 8 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

TABLE 1 | Technical parameters used to model conversion nodes.

φ1 φ2 φ3 µ 1+,−

HVDC Interconnection 0.9499

IEA ETSAP, 2014; Xiang et al., 2016 -

Electrolysis 50.6 9.0 8.0 0.05 1.0

Gotz et al., 2016 GWhel/ktH2 ktH2O /ktH2 ktO2

/ktH2 - -/h

Methanation 0.5 2.75 2.25 1.0 0.0

Gotz et al., 2016; Roensch et al., 2016 ktH2/ktCH4 ktCO2

/ktCH4 ktH2O

/ktCH4 - -/h

Desalination 0.004 1.0 0.0

International Renewable Energy Agency (IRENA), 2012 GWhel/ktH2O - -/h

Direct Air Capture 0.1091 0.0438 5.0 1.0 0.0

Keith et al., 2018 GWhel/ktCO2 ktH2/ktCO2

ktH2O /ktCO2

- -/h

CH4 Liquefaction 0.616 0.0 1.0

Pospisil et al., 2019 GWhel/ktLCH4 - -/h

LCH4 Carriers 0.994

Howard Rogers, 2018 -

LCH4 Regasification 0.98

Pospisil et al., 2019 -

passage of an electric current. Hence, the plants are modelled with four external and one internal variables. The external variables represent the power and water inflows as well as the hydrogen and oxygen outflows, while the internal variable is the plant capacity. The reference commodity r used in Equation (5) is hydrogen, while the commodity r′ according to which the technology is sized in Equation (7) is the power input. Electrolysis plants are also assumed to operate at 20 bar and 40◦C. This technology is flexible and can ramp up and down very quickly (usually within seconds). Hence, no ramping constraints are used. However, a minimum hydrogen production level around 5 − 10% of the nominal capacity must be maintained when the plant is switched on. Constraints (9) are therefore used to model plant operation. The usual objective function (12) is also used.

Methanation Plants. The carbon dioxide methanation (Sabatier) reaction enables the conversion of carbon dioxide and hydrogen into methane and water (steam) and is highly exothermic (i.e., the production of 1 kg of methane releases approximately 2.867 kWh of high-temperature heat Roensch et al., 2016). In this paper, cooled fixed-bed (catalytic) reactors operating at 300◦C and 20 bar are assumed to be used to produce synthetic methane via the Sabatier reaction. Furthermore, carbon dioxide and hydrogen are assumed to be fed in stoichiometric proportions, and the conversion of reactants is assumed to be complete (which is facilitated by the use of, e.g., alumina- supported nickel catalysts Mills and Steffgen, 1974 that offer good selectivity and are relatively cheap). Plants are modelled using four external variables and one internal variable. The external variables represent the hydrogen and carbon dioxide inflows as well as the methane and water (steam) outflows, while

the internal variable is the plant capacity. Methane is taken as the reference commodity r used to describe the process in Equation (5) as well as the reference commodity r′ used for sizing the plant in Equation (7). Owing to the exothermicity of the reaction, cooled fixed-bed reactors are very sensitive to changes in operating parameters such as the feed and coolant temperatures (Schlereth and Hinrichsen, 2014) or the feed flow rate (Theurich, 2019), and can suffer from pronounced hot-spots or thermal runaway (Schlereth and Hinrichsen, 2014), which can in turn lead to catalyst sintering and deactivation. Hence, reactors usually have a (very) narrow operating range, although promising ways of improving this have been proposed (Bremer and Sundmacher, 2019). Dynamic operation may also involve shutting down the reactor, keeping it in hot standby and starting it up again (Gorre et al., 2020), which nevertheless leads to inefficiencies (e.g., the reactor must be flushed with hydrogen Gorre et al., 2020). Finally, note that maintaining product quality is typically more difficult in unsteady operation. In the light of these observations, in this paper, it is assumed that methanation reactors operate in steady state. Constraints (9), (10), (11) are therefore used to model plant operation, while investment and operating costs are modelled via (12).

Water Desalination Plants. Reverse osmosis (RO) plants are employed to desalinate seawater and produce freshwater (Caldera et al., 2016). This technology essentially pumps seawater into a chamber featuring a porous membrane and produces a pressure differential across the membrane, enabling dead- end filtration and the recovery of freshwater on the other side of the membrane. The plants are modelled with two external variables and one internal variable. The external variables are

Frontiers in Energy Research | www.frontiersin.org 9 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

TABLE 2 | Economic parameters used to model conversion nodes (2030 estimates).

CAPEX FOM (θf) VOM (θv) Lifetime

Solar Photovoltaic Panels 380.0 7.25 0.0 25.0

Danish Energy Agency, 2020b Me/GWel Me/GWel-yr Me/GWhel yr

Wind Turbines 1040.0 12.6 0.00135 30.0

Danish Energy Agency, 2020b Me/GWel Me/GWel-yr Me/GWhel yr

HVDC Interconnection 480.0 7.1 0.0 40.0

EIA, 2018; CIGRE C1.35 Working Group, 2019 Me/GWel Me/GWel-yr Me/GWhel yr

Electrolysis 600.0 30.0 0.0 15.0

Danish Energy Agency, 2020c Me/GWel Me/GWel-yr Me/GWhel yr

Methanation 735.0 29.4 0.0 20.0

International Energy Agency (IEA), 2019 Me/GWCH4 (HHV) Me/GWCH4

-yr (HHV) Me/GWhCH4 (HHV) yr

Desalination 28.08 0.0 0.000315 20.0

CMI Marseille, 2016 Me/(ktH2O /h) Me/(ktH2O

/h)-yr Me/ktH2O yr

Direct Air Capture 4801.4 0.0 0.0207 30.0

Keith et al., 2018 Me/(ktCO2 /h) Me/(ktCO2

/h)-yr Me/ktCO2 yr

CH4 Liquefaction 5913.0 147.825 0.0 30.0

Brian Songhurst, 2018 Me/(ktLCH4 /h) Me/(ktLCH4

/h)-yr Me/ktLCH4 yr

LCH4 Carriers 2.537 0.12685 0.0 30.0

Economic Research Institute for ASEAN and East Asia (ERIA), 2018 Me/ktLCH4 Me/ktLCH4

-yr Me/ktLCH4 yr

LCH4 Regasification 1248.3 24.97 0.0 30.0

Dongsha et al., 2017 Me/(ktCH4 /h) Me/(ktCH4

/h)-yr Me/ktCH4 yr

FIGURE 4 | Capacity factor time series πnt used for solar PV and wind power plants in 2016. For all other nodes except liquefied methane carriers, π n t = 1 over the

entire time horizon.

the power required to drive pumps and the freshwater outflow, whereas the internal variable represents the plant capacity. The reference commodity r′ according to which the plant is sized is the freshwater flow out of the system. For mechanical reasons, membranes are usually designed to operate under constant pressure and plants therefore operate more or less continuously. Hence, constraints (5), (7), (9), (10), (11) are used to model plant sizing and operation, while investment and operating

costs are modelled via (12). Note that the seawater inflow and brine discharge are not modelled. The implicit assumptions are that seawater is freely available and that the brine by- product can be disposed of at no cost, without any restriction on pumped volumes.

Direct Air Capture Units. Direct air capture units extract carbon dioxide from the atmosphere (Kiani et al., 2020). The process used in this paper is the one proposed by Keith

Frontiers in Energy Research | www.frontiersin.org 10 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

et al. (2018). Roughly speaking, this process relies on four main chemical reactions, which are combined to form two chemical loops. In the first loop, aqueous sorbents are used in an air contactor to chemically bind carbon dioxide and form dissolved compounds. These compounds then react with pellets in a fluidised-bed reactor, making it possible to recover the aforementioned sorbents and trap carbon in solid compounds. The second loop essentially recovers carbon dioxide by calcining the solid compounds and replenishes the pellet stock by hydrating (slaking) the solid product of the calcination reaction. The process requires electricity to power fans driving air through the contactors, pumps maintaining the flow of aqueous solutions as well as compressors compressing the output carbon dioxide stream from atmospheric pressure to 20 bar (the associated energy expense is approximated via the polytropic compression work, assuming a polytropic efficiency of 80%). The net power consumption is obtained as the difference between the total consumption of these subsystems and the power produced by a steam turbine recovering slaking heat. A sustained water supply is also necessary to form aqueous solutions, counter natural evaporation in the air contactors and produce steam used in the slaker. Furthermore, a source of heat at around 900◦C is required for the calcination reaction. In the original design, natural gas is burnt via an oxy-fuel combustion process at the bottom of the calciner to provide this heat, and the off-gases fluidise the reactor. In this paper, it is assumed that the high temperature heat (approximately 1.46 MWh per ton of carbon dioxide) is provided by burning hydrogen (assuming a lower heating value of 33.3 MWh per ton of hydrogen burnt). Hence, the process is modelled using four external variables and one internal variable. The external variables represent the power, water and hydrogen inflows as well as the carbon dioxide outflow, while the internal variable is the plant capacity. The reference flow according to which the plant is sized is the carbon dioxide outflow. None of the technologies implementing the various reactions really lend themselves to highly variable operation. Constraints (5), (7), (9), (10), (11) are therefore used to model plant sizing and operation, while investment and operating costs are modelled via (12).

Methane Liquefaction Units. Liquefaction units turn gaseous methane into liquefied methane (Pospisil et al., 2019). This technology typically relies on compressors and pumps in order to progressively compress and cool the methane inflow, which is eventually throttled and liquefied via the Joule-Thomson effect. In this case, three external variables and one internal variable are used. The external variables represent the methane inflow, the power consumption of compressors and pumps as well as the liquefied methane outflow (which is the reference commodity), while the internal variable represents the plant capacity. Constraints (5), (7), (9), (10), (11) are used to model plant sizing and operation, while investment and operating costs are modelled via (12).

Liquefied Methane Carrier Vessels. Liquefied methane is transported to market with large ocean-going vessels powered by dual fuel diesel electric (DFDE) engines (Howard Rogers, 2018). These engines are particularly efficient and can run solely on natural boil-off gas (i.e., gaseous methane resulting from the natural evaporation of liquefied methane stored on board in

insulated cargo tanks). This allows vessels to sail at a speed of 19 knots, with approximately 0.1% of their cargo evaporating due to natural boil-off per day spent at sea, which is used for propulsion (i.e., no other fuel is needed). The liquefied methane heel that must usually be maintained for the return journey to guarantee that the onboard tanks remain cool (roughly 4–5% of the total cargo) is neglected in this paper. Two external variables and one internal variable are used to describe a stylised carrier vessel. The external variables represent the flow of liquefied methane loaded at the coastal hub and the flow of liquefied methane unloaded at the destination, respectively. The internal variable is the vessel capacity. Equation (5) is used to model the transport of liquefied methane, with τ = 116 h, as the berthing and travel time between the coastal hub and the destination is assumed to take slightly less than 5 days. The conversion factor φ ≈ 0.994 represents the transport efficiency, computed from the boil-off consumption (0.125% of cargo per day) and trip duration (116 h). In addition, loading and unloading may only be possible when the vessel is moored at the coastal hub and destination, respectively. This is enforced via Equation (7) and time series πnt (with values equal to 0 or 1), which defines a berthing, mooring, loading and unloading schedule (loading or unloading take place when πnt = 1). For the sake of simplicity, πnt represents an aggregate schedule constructed from 7 different, non-overlapping schedules corresponding to individual carrier vessels. Some of these schedules are shown in Figure 5 (loading and unloading is assumed to take 24 h). The standard local objective (12) is used for the stylised carrier.

Liquefied Methane Regasification Units. Regasification units are used to transform liquefied methane into gaseous methane at the destination (Dongsha et al., 2017). The heat required to do so can come from a variety of sources. In this case, it is assumed to come from the combustion of a fraction of the methane (around 2%). Thus, two external variables and one internal variable are used. The external variables represent the liquefied methane inflow as well as the gaseous methane outflow, and the internal variable is the plant capacity. Constraints (5), (7) are used to model plant sizing and operation, while investment and operating costs are modelled via (12).

4.1.2. Storage Nodes Storage nodes are discussed in this subsection. Tables 3–5 gather the techno-economic data (2030 estimates) used to model storage nodes along with the original data sources and complement the descriptions below. In the model, power flows are measured in GW (GWh/h), energy is measured in GWh, mass flows are measured in kt/h, mass is measured in kt, and money is measured in Me.

Stationary Battery Storage. Nickel manganese cobalt (NMC) oxide lithium-ion batteries are used for short-term electricity storage (Danish Energy Agency, 2020a). Power in- and outflows are modelled using external variables. The state of charge, power capacity and energy capacity, on the other hand, are modelled as internal variables. Constraints (13), (15), (17), (18), (20), (21) are used, while the local objective function is given in (22).

Frontiers in Energy Research | www.frontiersin.org 11 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

FIGURE 5 | Subset of non-overlapping schedules used to construct the aggregate schedule πnt of stylised liquefied methane carriers. These time series are summed

to obtain the aggregate schedule πnt . For all other nodes except wind and solar PV power plants, π n t = 1 over the entire time horizon.

TABLE 3 | Technical parameters used to model storage nodes.

ηS η+ η− σ ρ φ

Battery Storage 0.00004 0.959 0.959 0.0 1.0

Danish Energy Agency, 2020a - - - - -

Compressed H2 Storage 1.0 1.0 1.0 0.05 1.0 1.3

Danish Energy Agency, 2020a GWhel/ktH2

Liquefied CO2 Storage 1.0 1.0 1.0 0.0 1.0 0.105

Mitsubishi Heavy Industries, 2004 GWhel/ktCO2

Liquefied CH4 Storage 1.0 1.0 1.0 0.0 1.0

Assumed - - - - -

H2O Storage 1.0 1.0 1.0 0.0 1.0 0.00036

Caldera et al., 2016 GWhel/ktH2O

TABLE 4 | Economic parameters used to model storage nodes (stock component, 2030 estimates).

CAPEX FOM (ϑf) VOM (ϑv) Lifetime

Battery Storage 142.0 0.0 0.0018 10.0

Danish Energy Agency, 2020a Me/GWh Me/GWh-yr Me/GWh yr

Compressed H2 Storage 45.0 2.25 0.0 30.0

Danish Energy Agency, 2020a Me/kt Me/kt-yr Me/kt yr

Liquefied CO2 Storage 1.35 0.0675 0.0 30.0

Mitsubishi Heavy Industries, 2004 Me/kt Me/kt-yr Me/kt yr

Liquefied CH4 Storage 2.641 0.05282 0.0 30.0

Interior Gas Utility, 2013 Me/kt Me/kt-yr Me/kt yr

H2O Storage 0.065 0.0013 0.0 30.0

Caldera et al., 2016 Me/kt Me/kt-yr Me/kt yr

Hydrogen Storage Tanks. Compressed hydrogen storage tanks are considered in this paper. More precisely, overground, man-made steel storage vessels (type I) withstanding pressure

levels around 200 bar and suitable for stationary applications are used (Danish Energy Agency, 2020a). A minimum inventory level of 5% is assumed to represent the cushion gas (effectively

Frontiers in Energy Research | www.frontiersin.org 12 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

TABLE 5 | Economic parameters used to model storage nodes (flow component, 2030 estimates).

CAPEX FOM (θf) VOM (θv) Lifetime

Battery Storage 160.0 0.5 0.0 10.0

Danish Energy Agency, 2020a Me/GW Me/GW-yr Me/GWh yr

Liquefied CO2 Storage 48.6 2.43 0.0 30.0

Mitsubishi Heavy Industries, 2004 Me/(kt/h) Me/(kt/h)-yr Me/kt yr

H2O Storage 1.55923 0.0312 0.0 30.0

Caldera et al., 2016 Me/(kt/h) Me/(kt/h)-yr Me/kt yr

reducing the working volume). Since hydrogen at 20 bar and 40◦C is produced by electrolysis plants, the hydrogen inflow must be compressed to 200 bar using electric compressors for storage purposes. The associated energy expense is approximated via the polytropic compression work (assuming a polytropic efficiency of 80%) (JRC, 2003). Thus, three external variables and three internal variables are used. The external variables represent the hydrogen inflow, the electricity consumption and the hydrogen outflow, while the state of charge, the power capacity and the energy capacity are modelled as internal variables. Constraints (13), (14), (15), (17), (18), (20), (21) are used, while the local objective function is (22).

Liquefied Carbon Dioxide Storage Tanks. Liquefied carbon storage tanks are used to store carbon dioxide. Liquefaction and regasification units are also required (Mitsubishi Heavy Industries, 2004). Liquefaction units consume electricity, while regasification units are assumed to use ambient heat to recover gaseous carbon dioxide. Hence, in this case, three external variables and five internal variables are used. The external variables are the gaseous carbon dioxide inflow, the power consumption of the liquefaction units and the gaseous carbon dioxide outflow, while the internal variables represent the state of charge, the tank capacity, capacities of liquefaction and regasification units, and the flows of liquefied carbon dioxide in and out of the tanks. Constraints (13), (14), (15), (17), (18), (20), (21) are used, while the local objective function is (22).

Liquefied Methane Storage Tanks. Liquefied methane is stored in full containment tanks (i.e., tanks with both inner and outer containment walls and such that the annular gap between both walls is sealed to prevent any gaseous leaks Interior Gas Utility, 2013). It is assumed that the boil-off gas keeping the content of the storage tanks cold is re-liquefied and pumped back into the tanks but the electricity consumption required to do so is neglected. Two external variables and two internal variables are used. The external variables are the liquefied methane in- and outflow, while internal variables represent the state of charge and the storage capacity. Constraints (13), (15), (17), (18), (20), (21) are used, while the local objective function is (22).

Water Storage Tanks. Water is stored in tanks equipped with electric pumps (Caldera et al., 2016). Three external variables and three internal variables are used. The external variables correspond to the water inflow, the power consumed by pumps

and the water outflow. The internal variables represent the state of charge, the tank capacity and the flow capacity of pipes feeding into the tank. Constraints (13), (15), (17), (18), (20), (21) are used, while the local objective function is (22).

4.1.3. Conservation Hyperedges Inland Power Balance. This hyperedge enforces active power flow conservation (which derives from Kirchhoff’s current law) in the inland cluster. It therefore guarantees that the sum of power flows from the solar PV plant, the wind power plant and the battery is equal to the sum of power flows to the HVDC interconnection and the battery. Note that both in and outflows are used for the battery, which correspond to charge and discharge flows, respectively. No exogenous power injections and withdrawals take place over this hyperedge, hence λet = 0, ∀t ∈ T. This is assumed to be the case for all other hyperedges as well, unless otherwise stated.

Coastal Power Balance. This hyperedge enforces active power flow conservation in the coastal cluster. It therefore guarantees that the power flow from the HVDC interconnection is equal to the sum of power flows to the direct air capture plant, the electrolysis plant, the hydrogen storage system, the methane liquefaction units, the desalination plant, the water storage system and the liquefied carbon dioxide storage system.

Coastal Hydrogen Balance. This hyperedge enforces conservation of hydrogen mass flows in the coastal cluster. Hence, it guarantees that the sum of flows from the electrolysis plants and the storage system is equal to the sum of flows to the direct air capture plants, the methanation plants and the storage system.

Coastal Carbon Dioxide Balance. This hyperedge enforces conservation of (gaseous) carbon dioxide mass flows in the coastal cluster. Thus, it guarantees that the sum of flows from the direct air capture units and the storage system is equal to the sum of flows to methanation plants and the storage system.

Coastal Water Balance. This hyperedge guarantees that the aggregate flow of freshwater generated by desalination and methanation plants in the coastal cluster exceeds the aggregate flow consumed by electrolysis and direct air capture plants. Hence, it is assumed that any freshwater surplus may be released into the environment without harm or used in other applications (e.g., cooling), and the equality constraint in Equation (23) is relaxed to a greater-than-or-equal-to inequality.

Frontiers in Energy Research | www.frontiersin.org 13 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

Coastal Methane Balance. This hyperedge enforces conservation of (gaseous) methane mass flows in the coastal cluster, and guarantees that flows from the methanation plants and to the liquefaction units are equal.

Coastal Liquefied Methane Balance. This hyperedge enforces conservation of liquefied methane mass flows in the coastal cluster. Thus, it guarantees that the sum of flows from the liquefaction units and the storage system is equal to the sum of flows to the storage system and the liquefied methane carriers.

Destination Liquefied Methane Balance. This hyperedge enforces conservation of liquefied methane mass flows at the destination. Hence, it guarantees that the sum of flows from the liquefied methane carriers and the storage system is equal to the sum of flows to the regasification units and storage system.

Destination Methane Balance. This hyperedge guarantees that the exogenous demand for methane at the destination is satisfied by the flows from the regasification units. The gas demand is set to 10 TWh (HHV) per annum, but no specific assumptions about the end-uses or sectors relying on synthetic methane are made. Furthermore, the liquefied methane terminal located in Northwestern Europe is assumed to be connected to some existing gas network infrastructure. Although the latter is not explicitly modelled, it is assumed to be able to absorb both short-term and seasonal demand variability (e.g., via its line pack and seasonal storage facilities Correa-Posada and Sanchez- Martin, 2015, as is the case in most systems). The demand profile, which essentially represents injections from the terminal into the gas network, is therefore assumed to be flat. Hence, assuming that synthetic methane has a HHV of 15.441 kWh/kg, the demand profile is obtained as λet = (10 × 10

3/8, 760) × (1/15.441) ≈ 0.07393 kt/h, ∀t ∈ T. Note that using the LHV would have resulted in a higher mass flow rate.

4.2. Scenarios One reference scenario and a comprehensive sensitivity analysis are presented. The reference scenario studies a system configuration that relies on a combination of solar PV and wind power plants for electricity generation. A uniform weighted average cost of capital (WACC) of 7% is assumed for all technologies, which represents the case where the funds required to finance the system are borrowed on capital markets. Under these assumptions, for a conversion or storage technology n, the CAPEX values in Tables 2, 4, 5 are used to compute

ζ n = CAPEXn ×

w

(1 − (1 + w)−Ln) , (24)

with Ln the lifetime of technology n and w the WACC. Hence, ζ n

represents the annualised cost of investing in technology n. The sensitivity analysis investigates the impact of a number

of techno-economic parameters and assumptions on synthetic methane cost, which include the availability of wind power plants, the investment costs of electrolysis, direct air capture and methanation plants, the operational flexibility of the latter two technologies as well as the energy consumption of direct air capture plants and the financing costs. More specifically, a hypothetical situation where the cost of financing the system

is zero is studied, such that the cost of synthetic methane production and delivery solely reflects the cost and efficiency of technologies in the supply chain. In this set-up, Equation (24) cannot be used and annualised investment costs are instead computed via

ζ n =

CAPEXn

Ln . (25)

5. RESULTS

5.1. Reference Scenario In the reference scenario, a system configuration relying on solar and wind power plants is studied assuming a WACC of 7%. In this set-up, synthetic methane is delivered to market in gaseous form at 149.7 e/MWh (HHV), which is computed as the ratio of total (annualised) system cost to methane volume delivered (10 TWh HHV per year). It is worth noting that using the LHV would have increased the cost per MWh, as this would have effectively reduced the amount of energy that could have been retrieved per unit mass of methane delivered.

The synthetic methane cost breakdown is provided in Figure 6, where each bar represents the contribution (in e/MWh) of the corresponding technology to synthetic methane cost. Each bar can also be interpreted as representing the contribution of the corresponding technology to total system cost. Wind turbines account for the lion’s share of synthetic methane cost (roughly 28.9%). Electrolysis (19.6%) and solar PV plants (11.3%) come in second and third, respectively, although they each represent a much smaller proportion of costs than wind turbines (taken together, they contribute slightly more than wind turbines to total system cost). Overall, the technologies used to generate, transport and store electricity (shown in gold in Figure 6) represent the largest share of costs (around 56.6%). Hydrogen storage plants, which are used as a buffer between flexible electrolysis and inflexible methanation plants, make up approximately 5% of total cost. Hence, the technologies producing and storing hydrogen (shown in light blue in Figure 6) account for roughly 25% of total system cost. It is worth noting that the plants upstream of the inflexible plants (i.e., methanation, direct air capture, and desalination plants) make up almost 80% of total system cost. On the other hand, methanation plants make up a minor share of total cost (approximately 7.7%), and the full methane chain (i.e., production, liquefaction, storage, transport, and regasification, shown in light orange in Figure 6) accounts for roughly 12.5% of final product cost. Direct air capture plants also represent a minor fraction of system cost (around 7.9%, shown in green in Figure 6). Water desalination and storage technologies are deployed in moderate quantities, resulting in a very small share of total cost (well under 1%), while carbon dioxide storage is not deployed.

Analysing mass and energy balances provides some insight into system design and operation. Figure 7 displays mass and energy balances (flow values are integrated over the full optimisation horizon of five years and divided by the number of years) along with technology capacities. Firstly, as can be seen in Figure 7, the average annual electricity production

Frontiers in Energy Research | www.frontiersin.org 14 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

FIGURE 6 | Breakdown of synthetic methane cost at destination for reference scenario. All contributions roughly sum to 149.7 e/MWh (HHV).

of solar PV and wind power plants is equal to 8.90 and 13.87 TWh, respectively, which suggests that the full supply chain has a conversion efficiency of roughly 43.9%. However, the amount of curtailment is substantial and stands at 5.26 TWh, which represents slightly less than one quarter of the useful power production. This effectively decreases the capacity factors of the photovoltaic and wind power plants from their theoretical maximum of 24.6 and 50.0% over 2015–2019 (i.e., corresponding to the case where all electricity produced is used) to 23.8 and 36.8% (taking only useful power production into account). However, the high share of curtailment attributed to wind power plants must be taken with caution. Indeed, since curtailment is not penalised in the objective, some natural symmetry exists in the model when solar PV and wind power plants jointly produce more than the system can absorb (which typically occurs during hours of peak solar PV production). More specifically, in such situations, solutions curtailing a given share of solar PV output would yield the same objective values as that of other solutions curtailing the same given share of wind power output, and curtailment could therefore be attributed to one technology or the other without any impact on the objective. The high overall rate of curtailment can nevertheless be explained by the difficulty of effectively absorbing the highly variable aggregate power input from renewable power plants. This is a direct consequence of the fact that the operating regimes of several key conversion technologies are inflexible, which has two further implications. Firstly, battery and hydrogen storage systems are deployed at great cost in order to smooth the variability of the power supply as much as possible. Secondly, plants located upstream of the inflexible ones are typically oversized, as the level of

smoothing required to guarantee steady power and hydrogen flows cannot be economically provided by storage plants alone. Additional evidence supporting this analysis is provided in the following section.

5.2. Sensitivity Analysis The sensitivity analysis investigates the impact of a number of techno-economic parameters and assumptions on synthetic methane cost. More precisely, the impacts of (i) being unable to deploy wind power plants, (ii) the operational flexibility of direct air capture, methanation, and desalination plants, (iii) the investment costs of electrolysis, direct air capture, and methanation plants, (iv) the energy consumption of direct air capture plants, (v) financing costs are assessed. The cost share, capacities and capacity factors of the technologies that were found to contribute the most to total system cost in Section 5.1 are computed and analysed in this section. Hence, Figure 8 displays the breakdown of synthetic methane costs obtained under various techno-economic assumptions, while Figure 9 gathers the capacities and average capacity factors of key conversion and storage technologies. These results are elaborated upon below.

Solar PV System. This analysis assumes that wind power plants cannot be deployed, and the full electricity supply must therefore come from solar PV power plants. As can be seen in Figure 8, the cost of synthetic methane for this configuration is around 202 e/MWh, which is almost 35% more expensive than that found in the reference scenario. In this case, electrolysis plants contribute the most to total system cost, followed by solar PV power plants. In addition, it is clear from Figure 8 that the cost of each technology located upstream of the inflexible plants

Frontiers in Energy Research | www.frontiersin.org 15 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

FIGURE 7 | Material and energy balance diagram for the reference scenario, along with technology capacities. All energy-equivalent flows of energy carriers other

than electricity were computed using their HHV. Flow values represent yearly averages (i.e., flows were integrated over the full time horizon and divided by the number

of years) and all values were rounded up to keep significant digits only.

(i.e., methanation, direct air capture, and desalination plants) increases in absolute terms compared with the reference scenario. The only technologies whose cost remains the same are direct air capture and methanation. This claim is supported by the results shown in Figure 9. Indeed, the capacities of conversion and storage technologies located upstream of the inflexible ones are larger in the solar-only case, while the capacities of inflexible technologies are equal in both cases. Furthermore, the average capacity factors of key conversion technologies (e.g., electrolysis plants) are much lower in the solar-only case. This observation suggests that these plants are oversized to help absorb the highly variable input from solar PV power plants, as smoothing it with storage systems alone would be uneconomical. Overall, relying on solar PV power plants alone results in a system design that is much less efficient and much more expensive than the one identified in the reference scenario.

System Flexibility. In the reference scenario, several technologies were assumed to be inflexible, namely methanation (MT), direct air capture (DAC) and desalination (DS) plants. Their inflexibility, combined with the fact that the renewable

power supply is highly variable, was found to have an impact on synthetic methane cost. Although these assumptions are well-founded, the minimum level and ramping constraints of the three aforementioned technologies are now relaxed (i.e., µ = 0.0 and 1+ = 1− = 1.0) in order to evaluate the sensitivity of our results. Figure 8 shows that shifting to a system with fully flexible methanation, direct air capture and desalination plants can lead to cost savings around 6%, and has a substantial impact on the capacities of several technologies. More specifically, the capacity of solar PV power plants decreases by 10%, while the capacities of battery and hydrogen storage systems shrink by 40 and 80%, respectively. On the other hand, the capacities of methanation and direct air capture plants increase by 20 and 11%, respectively, which slightly offsets the cost savings made elsewhere. Although not shown in Figure 9, the capacity of liquefied methane storage tanks in the coastal hub more than doubles. This can be explained by the fact that the transport of liquefied methane does not occur on a continuous basis, which introduces some inflexibility in the supply chain. Since liquefied methane storage is much cheaper than battery or hydrogen

Frontiers in Energy Research | www.frontiersin.org 16 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

FIGURE 8 | Breakdown of synthetic methane costs obtained under various techno-economic assumptions.

FIGURE 9 | Capacities and average capacity factors (shown as greyed fractions of capacity bars) of key conversion and storage technologies under various

techno-economic assumptions.

storage, the buffer absorbing the variability of renewable power generation is moved downstream in the supply chain. It is worth noting that even if the transport of liquefied methane took

place continuously, the mismatch between the production and demand profiles would need to be balanced by some storage capacity, which will always come at a cost.

Frontiers in Energy Research | www.frontiersin.org 17 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

Investment Costs. A number of technologies used in the proposed remote carbon-neutral synthetic methane supply chain have not yet reached full maturity. In particular, electrolysis (EL), direct air capture and methanation plants are still undergoing development, and their costs thus remain highly uncertain to 2030. The cost figures used for electrolysis plants in the reference scenario lean toward moderately optimistic for 2030, and achieving them would require both sustained research and development efforts and a commercial scale-up (Schmidt et al., 2017). Likewise, the cost figures used for direct air capture plants are on the fairly optimistic side for 2030. Indeed, the cost of each ton of carbon dioxide captured from the atmosphere is between 65 and 70 e (energy costs excluded) in the reference scenario, which is close to long-term cost targets of $100/ton (energy costs included) that Keith et al. (2018) seek to reach. Methanation cost estimates, however, are rather conservative (Agora Verkehrswende et al., 2018; International Energy Agency (IEA), 2019). The uncertainty around these costs and their impact on synthetic methane cost is resolved as follows. Firstly, the CAPEX and FOM of electrolysis and direct air capture plants are increased by 50%, which yields synthetic methane cost estimates around 168.1 e/MWh (which is approximately 12% higher than that of the reference scenario). Then, the CAPEX and FOM of electrolysis, direct air capture and methanation plants are decreased by 50%, first on an individual basis and then all at once. Although such drastic cost reductions seem very unlikely, they nevertheless make it possible to estimate the sensitivity of synthetic methane cost to these assumptions and provide a lower bound on costs that may realistically be achieved. The least sensitive of these parameters is the cost of direct air capture plants, followed by the cost of methanation plants. Decreasing them individually only leads to methane cost reductions of 5% or less. Decreasing the cost of electrolysis plants by 50%, however, has a much greater impact and leads to cost reductions around 10%. Decreasing the costs of all three technologies at once leads to cost reductions of roughly 16% and yields a strong lower bound of 125.1e/MWh on the reductions in synthetic methane cost that may be achieved in this fashion. It is worth noting that these cost savings are achieved with virtually no change in deployed capacities (as shown in Figure 9), which suggests that the latter mostly depend on renewable production profiles, the flexibility of technologies and the demand that must be satisfied.

DAC Energy Consumption. The direct air capture process used in this analysis requires high-temperature heat to calcine calcium carbonate compounds and release the carbon dioxide that they trap. In this paper, it is assumed that this heat is provided by burning hydrogen, and slightly less than 20% of the total hydrogen production is used for this purpose. Hence, the fact that this hydrogen must be produced from renewable electricity leads to the deployment of additional power generation, storage, transport, electrolysis and hydrogen storage capacity, which directly translates into a lower overall efficiency of the full supply chain and a higher synthetic methane cost. Instead, a different capture process that only uses low to medium- temperature heat and electricity could be used (Wurzbacher et al., 2011). It has been estimated that this process consumes approximately 0.5 MWh of electricity per ton of carbon dioxide

captured and requires roughly 2.5 MWh of heat at 100◦C. Since the production of one ton of synthetic methane by the Sabatier reaction releases 2.87 MWh of high temperature heat (and requires 2.75 tons of carbon dioxide), some of the heat required by the direct air capture process could be supplied by nearby methanation plants. The impact of such a change is analysed by increasing the electricity consumption of the direct air capture process fivefold (i.e., setting it to roughly 0.5 MWh/ton instead of 0.1 MWh/ton in the reference scenario) and setting the hydrogen consumption to zero. The same cost figures as those of the reference scenario are used. Results shown in Figures 8, 9 confirm the intuition that using hydrogen to satisfy the heat requirements of a high-temperature direct air capture process has a substantial impact on synthetic methane cost and the capacities of various key technologies. More precisely, the cost of synthetic methane in this configuration is approximately 10% lower than that found in the reference scenario, while the capacities of power generation and electrolysis plants are 10 and 20% smaller, respectively. Although promising, a detailed analysis of the heat integration potential and the cost of this process should be performed in order to confirm these findings.

Financing Costs. A weighted average cost of capital of 7% has so far been used. In order to evaluate the impact of financing costs, a hypothetical situation where the cost of financing the system is set to zero is now studied (i.e., annualised CAPEX values are calculated using Equation (25)). Thus, in this set-up, the cost of synthetic methane production and delivery solely reflects the costs and efficiencies of technologies in the supply chain, and provides an absolute lower bound on costs that may be realistically achieved. Results in Figure 8 suggest that neglecting financing costs leads to a synthetic methane cost of 88.3e/MWh (which corresponds to a 40% reduction compared with the reference scenario), which is by far the lowest observed in this paper and highlights the influence of weighted average cost of capital assumptions. It is also worth noting that this cost decrease is achieved with little change in the capacities of conversion and storage technologies compared with the reference scenario.

5.3. Discussion Discrepancies exist between the results presented in Sections 5.1 and 5.2 and synthetic methane production cost estimates published elsewhere in the literature. Indeed, recall that Zeman and Keith (2008) provide cost estimates ranging from 74.1 to 94.6 e/MWh. Furthermore, in Fasihi and Bogdanov (2015), the cost of producing synthetic methane from renewable electricity in central and southern Algeria and delivering it to Japan is estimated to be around 65–75 e/MWh in 2030 for a hybrid solar-wind system using a WACC of 7%. In Fasihi et al. (2017), the cost of producing synthetic methane in the same region and delivering it to Finland is estimated to be between 100 and 110 e/MWh (HHV) by 2030 and between 90 and 100 e/MWh (HHV) by 2040, respectively, using a WACC of 7%. Finally, in Agora Verkehrswende et al. (2018), a uniform WACC of 6% is used, yielding cost estimates around 140 e/MWh (LHV) for a solar PV configuration and around 150 e/MWh (LHV) for a hybrid solar-wind configuration. Notably, the hybrid solar-wind

Frontiers in Energy Research | www.frontiersin.org 18 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

configuration is slightly more expensive than the solar-powered system in their reference cost scenario.

The methods used in the aforementioned papers, which are discussed in Section 2, suffer from several shortcomings. More precisely, they use a very low temporal resolution (one time period per year in the so-called full load hour model) that completely smoothes out the variability of power production signals. Furthermore, their models have a very low level of technical detail. The combination of these two features makes it very difficult to capture the interaction between subsystems and accurately model the supply chain in an integrated fashion. Hence, this typically removes the need for oversizing renewable power generation technologies or deploying flexibility options such as storage systems to balance the variable power supply and satisfy operating constraints, in spite of the fact that the operation of some technologies further down the chain is inflexible or discontinuous (e.g., methanation plants or transport by liquefied methane carrier vessels). Oversizing plants or deploying storage technologies is relatively expensive and both account for a non- negligible share of the final methane cost, as discussed in Sections 5.1 and 5.2. For example, the fact that solar PV variability has been completely smoothed out by the full load hour model used in Agora Verkehrswende et al. (2018) explains the fact that solar-only and hybrid solar-wind configurations yield very close methane cost estimates, while the solar-only configuration is almost 35% more expensive than the hybrid wind-solar configuration considered in this paper. Thus, the aforementioned papers underestimate final product cost as a result of inadequate modelling choices. In addition, some of the techno-economic assumptions made in Fasihi and Bogdanov (2015) and Fasihi et al. (2017) seem particularly optimistic. For example, the CAPEX of electrolysis and methanation plants is approximately two and three times lower than the values used in the reference case presented in this paper. These assumptions clearly lead to low methane cost estimates but are poorly supported. Indeed, to the authors’ best knowledge, such assumptions do not appear elsewhere in the literature or in publicly-accessible databases and are therefore difficult to cross-check.

6. CONCLUSION AND FUTURE WORK

This paper studies the economics of carbon-neutral synthetic fuel production in remote areas where high-quality renewable resources are abundant. With this goal in mind, a hypergraph- based optimisation modelling framework directly applicable to the strategic planning of remote renewable energy supply chains is proposed. The method is leveraged to study the economics of carbon-neutral synthetic methane production from solar and wind energy in North Africa.

The full supply chain is modelled and optimised in an integrated fashion over five years (2015–2019) with hourly time resolution. Essential operational constraints are taken into account, which is key for accurately capturing interactions between subsystems. Results suggest that the cost of synthetic methane delivery to northwestern European consumers would be around 149.7 e/MWh (HHV) by 2030 for a system

that relies on a combination of solar photovoltaic and wind power plants, assuming a uniform weighted average cost of capital of 7%. A comprehensive sensitivity analysis has also been carried out in order to evaluate the impact of various techno-economic parameters and assumptions on synthetic methane cost, including the availability of wind power plants, the investment costs of electrolysis, methanation and direct air capture plants, their operational flexibility, the energy consumption of direct air capture plants, and financing costs. The most expensive configuration (around 200 e/MWh) relies on solar photovoltaic power plants alone, while the cheapest configuration (around 88 e/MWh) makes use of a combination of solar PV and wind power plants and is obtained when financing costs are set to zero. The cost estimates found for the reference scenario and the configuration relying solely on solar PV power plants are much higher than those previously published in the literature. This discrepancy can be partly explained by the fact that the models used in previous studies had a very low temporal resolution and failed to properly capture the interactions between highly variable power generation plants (especially solar photovoltaic units) and inflexible conversion technologies (such as methanation plants) and demand profiles.

Several research directions can be pursued in future work. Firstly, quantitatively analysing some of the options suggested for cost reductions would provide more insight into the economic potential of an energy supply pathway based on carbon- neutral methane synthesis in remote areas. Then, leveraging the framework to study different pathways involving different regions (and thus resource types and profiles) and energy carriers (e.g., hydrogen, methanol, or ammonia), would allow one to draw a complete picture of energy supply options and to compare their respective merits. Finally, the graph-based modelling framework could be expanded in different ways. For instance, the class of problems that can be represented could be broadened by introducing non-linear expressions. The graph representation could also be exploited to facilitate preprocessing tasks and the analysis of model properties, eventually enabling the deployment of more efficient solution methods that better exploit problem structure (Jalving et al., 2019).

DATA AVAILABILITY STATEMENT

Publicly available datasets were analysed in this study. This data can be found here: GBOML repository (Miftari et al., 2021).

AUTHOR CONTRIBUTIONS

MB and DE designed the research. MB, DR, and GD collected the data. MB performed the research and drafted the manuscript. DR, GD, TD, AR, and DE provided feedback on the research and manuscript. All authors contributed to the article and approved the submitted version.

ACKNOWLEDGMENTS

The authors would like to thank Adrien Bolland, Hatim Djelassi, and Virginie Pison for providing feedback on an

Frontiers in Energy Research | www.frontiersin.org 19 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

earlier version of this manuscript. The authors would also like to thank Steffi Theurich for insightful discussions about the physics and operation of methanation reactors as well as Julien Confetti for his precious help with the design of

figures and diagrams used in this paper. Finally, the authors would like to gratefully acknowledge the support of the Federal Government of Belgium through its Energy Transition Fund and the INTEGRATION project.

REFERENCES

Agora Verkehrswende, Agora Energiewende, and Frontier Economics (2018). The

Future Cost of Electricity-based Synthetic Fuels. Available online at: https://

www.agora-energiewende.de/en/publications/the-future-cost-of-electricity-

based-synthetic-fuels-1/ (accessed February 22, 2021).

Berger, M., Bolland, A., Miftari, B., Djelassi, H., and Ernst, D. (2021). Graph-Based

Optimization Modelling Language: A Tutorial. Available online at: http://hdl.

handle.net/2268/256705 (accessed February 22, 2021).

Berger, M., Radu, D., Fonteneau, R., Deschuyteneer, T., Detienne, G., and Ernst,

D. (2020). The role of power-to-gas and carbon capture technologies in

cross-sector decarbonisation strategies. Electric Power Syst. Res. 180:106039.

doi: 10.1016/j.epsr.2019.106039

Biegler, L. T., and Grossmann, I. E. (2004). Retrospective on optimization. Comput.

Chem. Eng. 28, 1169–1192. doi: 10.1016/j.compchemeng.2003.11.003

Biswas, S., Kulkarni, A. P., Giddey, S., and Bhattacharya, S. (2020). A review

on synthesis of methane as a pathway for renewable energy storage with a

focus on solid oxide electrolytic cell-based processes. Front. Energy Res. 8:229.

doi: 10.3389/fenrg.2020.570112

Borgschulte, A. (2016). The hydrogen grand challenge. Front. Energy Res. 4:11.

doi: 10.3389/fenrg.2016.00011

Bremer, J., and Sundmacher, K. (2019). Operation range extension via hot-spot

control for catalytic co2 methanation reactors. React. Chem. Eng. 4, 1019–1037.

doi: 10.1039/C9RE00147F

Brian Songhurst (2018). LNG Plant Cost Reduction 2014-2018. Available online at:

https://www.oxfordenergy.org/publications/lng-plant-cost-reduction-2014-

18/ (accessed February 22, 2021).

Caldera, U., Bogdanov, D., and Breyer, C. (2016). Local cost of seawater RO

desalination based on solar PV and wind energy: a global estimate. Desalination

385, 207–216. doi: 10.1016/j.desal.2016.02.004

Carmo, M., Fritz, D. L., Mergel, J., and Stolten, D. (2013). A comprehensive

review on PEM water electrolysis. Int. J. Hydrogen Energy 38, 4901–4934.

doi: 10.1016/j.ijhydene.2013.01.151

Centi, G., Perathoner, S., Salladini, A., and Iaquaniello, G. (2020).

Economics of CO2 utilization: a critical analysis. Front. Energy Res. 8:241.

doi: 10.3389/fenrg.2020.567986

Chapman, A. J., Fraser, T., and Itaoka, K. (2017). Hydrogen import pathway

comparison framework incorporating cost and social preference: case studies

from Australia to Japan. Int. J. Energy Res. 41, 2374–2391. doi: 10.1002/er.3807

Chen, Q., and Grossmann, I. (2017). Recent developments and challenges in

optimization-based process synthesis. Annu. Rev. Chem. Biomol. Eng. 8,

249–283. doi: 10.1146/annurev-chembioeng-080615-033546

CIGRE C1.35 Working Group (2019). Global Electricity Network: Feasibility

Study. Available online at: https://orbi.uliege.be/handle/2268/239969 (accessed

February 22, 2021).

CMI Marseille (2016). Desalination Technologies and Economics. Available

online at: https://www.cmimarseille.org/knowledge-library/desalination-

technologies-and-economics-capex-opex-technological-game-changers-0

(accessed February 22, 2021).

Conejo, A. J., Baringo, L., Kazempour, S. J., and Siddiqui, A. S. (2016). Investment

in Electricity Generation and Transmission. Cham: Springer International

Publishing. doi: 10.1007/978-3-319-29501-5

Correa-Posada, C. M., and Sanchez-Martin, P. (2015). Integrated power and

natural gas model for energy adequacy in short-term operation. IEEE Trans.

Power Syst. 30, 3347–3355. doi: 10.1109/TPWRS.2014.2372013

Dagdougui, H. (2012). Models, methods and approaches for the planning and

design of the future hydrogen supply chain. Int. J. Hydrogen Energy 37,

5318–5327. doi: 10.1016/j.ijhydene.2011.08.041

Danish Energy Agency (2020a). Technology Data for Energy Storage. Available

online at: https://ens.dk/en/our-services/projections-and-models/technology-

data/technology-data-energy-storage (accessed November 30, 2020).

Danish Energy Agency (2020b). Technology Data for Generation of Electricity

and District Heating. Available online at: https://ens.dk/en/our-services/

projections-and-models/technology-data/technology-data-generation-

electricity-and (accessed November 30, 2020).

Danish Energy Agency (2020c). Technology Data for Renewable Fuels. Available

online at: https://ens.dk/en/our-services/projections-and-models/technology-

data/technology-data-renewable-fuels (accessed November 30, 2020).

Dongsha, Z., Ning, S., Jun, L., Li, L., and Yinghua, Z. (2017). Comparative Research

on LNG Receiving Terminals and FSRU. Available online at: https://www.

jtsi.wa.gov.au/docs/default-source/LNG-2017-Graduation-Presentations/

comparative-research-on-lng-receiving-terminals-and-fsru.pdf?sfvrsn=

9266d1c_8 (accessed February 22, 2021).

Economic Research Institute for ASEAN and East Asia (ERIA) (2018). Investment

in LNG Supply Chain Infrastructure Estimation. Available online at: https://

www.eria.org/research/formulating-policy-options-for-promoting-natural-

gas-utilization-in-the-east-asia-summit-region-volume-ii-supply-side-

analysis/ (accessed February 22, 2021).

EIA (2018). Assessing HVDC Transmission for Impacts of Non-Dispatchable

Generation. Available online at: https://www.eia.gov/analysis/studies/

electricity/hvdctransmission/pdf/transmission.pdf (accessed February 22,

2021).

European Center for Medium Range Weather Forecasts (ECMWF) (2020). ERA5

Database. Available online at: https://www.ecmwf.int/en/forecasts/datasets/

reanalysis-datasets/era5 (accessed February 22, 2021).

Eveloy, V., Romeo, L. M., Parra, D., and Qadrdan, M. (2021). Editorial: advances

in power-to-X: processes, systems, and deployment. Front. Energy Res. 9:63.

doi: 10.3389/fenrg.2021.650510

Fasihi, M., Bogdanov, D., and Breyer, C. (2015). “Economics of global LNG

trading based on hybrid PV-wind power plants,” in Proceedings of the

31st European Photovoltaic Solar Energy Conference (Hamburg), 3051–3067.

doi: 10.4229/EUPVSEC20152015-7DO.15.6

Fasihi, M., Bogdanov, D., and Breyer, C. (2017). Long-term hydrocarbon trade

options for the Maghreb region and Europe: renewable energy based synthetic

fuels for a net zero emissions world. Sustainability 9. doi: 10.3390/su9020306

Gallo, G., Longo, G., Pallottino, S., and Nguyen, S. (1993). Directed

hypergraphs and applications. Discrete Appl. Math. 42, 177–201.

doi: 10.1016/0166-218X(93)90045-P

Garcia, D. J., and You, F. (2015). Supply chain design and optimization:

challenges and opportunities. Comput. Chem. Eng. 81, 153–170.

doi: 10.1016/j.compchemeng.2015.03.015

Ghavam, S., Vahdati, M., Wilson, I. A. G., and Styring, P. (2021).

Sustainable ammonia production processes. Front. Energy Res. 9:34.

doi: 10.3389/fenrg.2021.580808

Gorre, J., Ruoss, F., Karjunen, H., Schaffert, J., and Tynjala, T. (2020).

Cost benefits of optimizing hydrogen storage and methanation capacities

for power-to-gas plants in dynamic operation. Appl. Energy 257:113967.

doi: 10.1016/j.apenergy.2019.113967

Gotz, M., Lefebvre, J., Mors, F., Koch, A. M., Graf, F., Bajohr, S., et al. (2016).

Renewable power-to-gas: a technological and economic review. Renew. Energy

85, 1371–1390. doi: 10.1016/j.renene.2015.07.066

Grossmann, I. E. (2002). Review of nonlinear mixed-integer and

disjunctive programming techniques. Optimizat. Eng. 3, 227–252.

doi: 10.1023/A:1021039126272

Haas, S., Schachler, B., and Krien, U. (2019). Windpowerlib - A Python

Library to Model Wind Power Plants (v0.2.0). Available online at: https://

windpowerlib.readthedocs.io/en/stable/index.html (accessed February

22, 2021).

Hashimoto, K., Yamasaki, M., Fujimura, K., Matsui, T., Izumiya, K., Komori,

M., et al. (1999). Global CO2 recycling: novel materials and prospect for

prevention of global warming and abundant energy supply. Mater. Sci. Eng. A

267, 200–206. doi: 10.1016/S0921-5093(99)00092-1

Frontiers in Energy Research | www.frontiersin.org 20 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

Heuser, P.-M., Ryberg, D. S., Grube, T., Robinius, M., and Stolten, D. (2019).

Techno-economic analysis of a potential energy trading link between Patagonia

and Japan based on CO2-free hydrogen. Int. J. Hydrogen Energy 44,

12733–12747. doi: 10.1016/j.ijhydene.2018.12.156

HOMER (2020). HOMER Solar PV Power Output Calculator. Available online at:

https://www.homerenergy.com/products/pro/docs/latest/index.html (accessed

February 22, 2021).

Howard Rogers (2018). The LNG Shipping Forecast: Costs Rebounding, Outlook

Uncertain. Available online at: https://www.oxfordenergy.org/publications/

lng-shipping-forecast-costs-rebounding-outlook-uncertain/ (accessed

February 22, 2021).

IEA ETSAP (2014). Electricity Transmission and Distribution. Available online at:

https://iea-etsap.org/E-TechDS/PDF/E12_el-t&d_KV_Apr2014_GSOK.pdf

(accessed February 22, 2021).

Interior Gas Utility (2013). LNG Storage Tank Cost Analysis. Available

online at: https://www.interiorgas.com/wpdm-package/lng-storage-tank-cost-

analysis/ (accessed February 22, 2021).

International Energy Agency (IEA) (2019). The Future of Hydrogen. Available

online at: https://www.iea.org/reports/the-future-of-hydrogen (accessed

February 22, 2021).

International Renewable Energy Agency (IRENA) (2012). Water Desalination

using Renewable Energy: Technology Brief. Available online at: https://www.

irena.org/publications/2012/Mar/Water-Desalination-Using-Renewable-

Energy (accessed February 22, 2021).

Jalving, J., Cao, Y., and Zavala, V. M. (2019). Graph-based modeling

and simulation of complex systems. Comp. Chem. Eng. 125, 134–154.

doi: 10.1016/j.compchemeng.2019.03.009

JRC (2003). Hydrogen Storage: State-of-the-Art and Future Perspective.

Available online at: https://ec.europa.eu/jrc/en/publication/eur-scientific-

and-technical-research-reports/hydrogen-storage-state-art-and-future-

perspective (accessed February 22, 2021).

Kallrath, J. (ed.). (2012). Algebraic Modeling Systems. Heidelberg: Springer-Verlag

Berlin Heidelberg. doi: 10.1007/978-3-642-23592-4

Keith, D. W., Holmes, G., Angelo, D. S., and Heidel, K. (2018). A

process for capturing CO2 from the atmosphere. Joule 2, 1573–1594.

doi: 10.1016/j.joule.2018.05.006

Kiani, A., Jiang, K., and Feron, P. (2020). Techno-economic assessment for CO2

capture from air using a conventional liquid-based absorption process. Front.

Energy Res. 8:92. doi: 10.3389/fenrg.2020.00092

Kokossis, A., Tsakalova, M., and Pyrgakis, K. (2015). “Design of integrated

biorefineries. computers and chemical engineering,” in Special Issue:

Selected papers from the 8th International Symposium on the Foundations

of Computer-Aided Process Design (FOCAPD 2014) (Cle Elum, WA).

doi: 10.1016/j.compchemeng.2015.05.021

Liesche, G., Schack, D., and Sundmacher, K. (2019). The FluxMax approach for

simultaneous process synthesis and heat integration: production of hydrogen

cyanide. AIChE J. 65:e16554. doi: 10.1002/aic.16554

Limpens, G., Jeanmart, H., and Marechal, F. (2020). Belgian energy transition: what

are the options? Energies 13. doi: 10.3390/en13010261

MacKay, D. (2008). Sustainable Energy - Without the Hot Air. Available online at:

https://www.withouthotair.com/

Maggi, A., Wenzel, M., and Sundmacher, K. (2020). Mixed-integer linear

programming (MILP) approach for the synthesis of efficient power-to-syngas

processes. Front. Energy Res. 8:161. doi: 10.3389/fenrg.2020.00161

Miftari, B., Berger, M., Bolland, A., Djelassi, H., and Ernst, D. (2021). GBOML

Repository: Graph-Based Optimization Modeling Language. Available online

at: https://gitlab.uliege.be/smart_grids/public/gboml (accessed February 22,

2021).

Mills, G. A., and Steffgen, F. W. (1974). Catalytic methanation. Catal. Rev. 8,

159–210. doi: 10.1080/01614947408071860

Mitsubishi Heavy Industries (2004). Ship Transport of CO2 (Report PH4/30.

Available online at: https://ieaghg.org/publications/technical-reports (accessed

February 22, 2021).

Montastruc, L., Belletante, S., Pagot, A., Negny, S., and Raynal, L. (2019).

From conceptual design to process design optimization: a review on

flowsheet synthesis. Oil Gas Sci. Technol. Rev. 74:80. doi: 10.2516/ogst/20

19048

Poncelet, K., Delarue, E., Six, D., Duerinck, J., and D’haeseleer, W. (2016).

Impact of the level of temporal and operational detail in energy-system

planning models. Appl. Energy 162, 631–643. doi: 10.1016/j.apenergy.2015.

10.100

Pospisil, J., Charvat, P., Arsenyeva, O., Klimes, L., Spilacek, M., and Klemes, J. J.

(2019). Energy demand of liquefaction and regasification of natural gas and the

potential of LNG for operative thermal energy storage. Renew. Sustain. Energy

Rev. 99, 1–15. doi: 10.1016/j.rser.2018.09.027

Radu, D., Berger, M., Fonteneau, R., Hardy, S., Fettweis, X., Le Du, M., et al. (2019).

Complementarity assessment of south Greenland katabatic flows and west

Europe wind regimes. Energy 175, 393–401. doi: 10.1016/j.energy.2019.03.048

Roensch, S., Schneider, J., Matthischke, S., Schlaeter, M., Gaetz, M., Lefebvre, J.,

et al. (2016). Review on methanation: from fundamentals to current projects.

Fuel 166, 276–296. doi: 10.1016/j.fuel.2015.10.111

Ryberg, D. S., Robinius, M., and Stolten, D. (2018). Evaluating land

eligibility constraints of renewable energy sources in Europe. Energies 11.

doi: 10.3390/en11051246

Sahinidis, N. V. (2004). Optimization under uncertainty: state-of-

the-art and opportunities. Comput. Chem. Eng. 28, 971–983.

doi: 10.1016/j.compchemeng.2003.09.017

Schack, D., Rihko-Struckmann, L., and Sundmacher, K. (2016). “Structure

optimization of power-to-chemicals (P2C) networks by linear programming

for the economic utilization of renewable surplus energy,” in 26th European

Symposium on Computer Aided Process Engineering, eds Z. Kravanja and M.

Bogataj (Elsevier), 1551–1556. doi: 10.1016/B978-0-444-63428-3.50263-0

Schichl, H. (2004). Theoretical Concepts and Design of Modeling

Languages for Mathematical Optimization. Boston, MA: Springer US.

doi: 10.1007/978-1-4613-0215-5_4

Schlereth, D., and Hinrichsen, O. (2014). A Fixed-bed reactor modeling

study on the methanation of CO2. Chem. Eng. Res. Design 92, 702–712.

doi: 10.1016/j.cherd.2013.11.014

Schmidt, O., Gambhir, A., Staffell, I., Hawkes, A., Nelson, J., and Few, S. (2017).

Future cost and performance of water electrolysis: an expert elicitation study.

Int. J. Hydrogen Energy 42, 30470–30492. doi: 10.1016/j.ijhydene.2017.10.045

Segreto, M., Principe, L., Desormeaux, A., Torre, M., Tomassetti, L., Tratzi,

P., et al. (2020). Trends in social acceptance of renewable energy

across Europe–A literature review. Int. J. Environ. Res. Public Health 17.

doi: 10.3390/ijerph17249161

Theurich, S. (2019). Unsteady-state operation of a fixed-bed recycle reactor for the

methanation of carbon dioxide (Ph.D. thesis). Ulm University, Ulm, Germany.

Timera Energy (2018). Deconstructing LNG Shipping Costs. Available online

at: https://timera-energy.com/deconstructing-lng-shipping-costs/ (accessed

February 22, 2021).

TrinaSolar (2017). TrinaSolar AllMax M Plus Monocrystalline Module Data Sheet.

Available online at: https://www.trinasolar.com/en-apac/resources/downloads

(accessed February 22, 2021).

Wulf, C., Zapp, P., and Schreiber, A. (2020). Review of power-to-x demonstration

projects in Europe. Front. Energy Res. 8:191. doi: 10.3389/fenrg.2020.00191

Wurzbacher, J. A., Gebald, C., and Steinfeld, A. (2011). Separation of CO2 from air

by temperature-vacuum swing adsorption using diamine-functionalized silica

gel. Energy Environ. Sci. 4, 3584–3592. doi: 10.1039/c1ee01681d

Xiang, X., Merlin, M. M. C., and Green, T. C. (2016). “Cost analysis and

comparison of HVAC, LFAC and HVDC for offshore wind power connection,”

in 12th IET International Conference on AC and DC Power Transmission (ACDC

2016), 1–6. doi: 10.1049/cp.2016.0386

Zeman, F. S., and Keith, D. W. (2008). Carbon neutral hydrocarbons. Philos. Trans.

R. Soc. A Math. Phys. Eng. Sci. 366, 3901–3918. doi: 10.1098/rsta.2008.0143

Conflict of Interest: GD and TD were employed by company Fluxys SA.

The remaining authors declare that the research was conducted in the absence of

any commercial or financial relationships that could be construed as a potential

conflict of interest.

Copyright © 2021 Berger, Radu, Detienne, Deschuyteneer, Richel and Ernst. This

is an open-access article distributed under the terms of the Creative Commons

Attribution License (CC BY). The use, distribution or reproduction in other forums

is permitted, provided the original author(s) and the copyright owner(s) are credited

and that the original publication in this journal is cited, in accordance with accepted

academic practice. No use, distribution or reproduction is permitted which does not

comply with these terms.

Frontiers in Energy Research | www.frontiersin.org 21 June 2021 | Volume 9 | Article 671279

Berger et al. Renewable Hubs for Fuel Production

NOMENCLATURE

Sets and Indices

E, e set of hyperedges and hyperedge index

eT, eH tail and head of hyperedge e ∈ E

G hypergraph with node set N and hyperedge set E

In, i set of external variables at node n, and variable index

N, n set of nodes and node index

T, t set of time periods and time index

Parameters

ν ∈ N number of years spanned by optimisation horizon

πn t ∈ [0, 1] (operational) availability of conversion node n at time t

κn ∈ R+ existing flow capacity of conversion or storage node n

κ̄n ∈ R+ maximum flow capacity of conversion or storage node n

µn ∈ [0, 1] minimum operating level of conversion node n (fraction of capacity)

δ ∈ R+ duration of each time period

1n i,+

∈ [0, 1] maximum ramp-up rate for commodity i and conversion node n (frac. of capacity per unit time)

1n i,−

∈ [0, 1] maximum ramp-down rate for commodity i and conversion node n (frac. of capacity per unit time)

φn i ∈ R+ conversion factor between flows of reference commodity r

and commodity i for conversion or storage node n

τn i ∈ N conversion time delay for commodity i of conversion node n

ηn S ∈ [0, 1] self-discharge rate of storage node n

ηn+ ∈ [0, 1] charge efficiency of storage node n

ηn− ∈ [0, 1] discharge efficiency of storage node n

σ n ∈ [0, 1] minimum inventory level of storage node n (fraction of capacity)

ǭn ∈ R+ maximum inventory capacity of storage node n

ǫn ∈ R+ existing inventory capacity of storage node n

ρn ∈ R+ maximum discharge-to-charge ratio of storage node n

λe t ∈ R withdrawal/injection at time t and conservation hyperedge e

ζ n ∈ R+ annualised CAPEX of node n (flow component)

θn f ∈ R+ FOM cost of node n (flow component)

θn t,v

∈ R+ VOM cost of node n (flow component)

ςn ∈ R+ annualised CAPEX of storage node n (stock component)

ϑn f ∈ R+ FOM cost of storage node n (stock component)

ϑn t,v

∈ R+ VOM cost of storage node n (stock component)

θn t,L

∈ R+ cost of unserved demand at conservation node n

Variables

qn it ∈ R+ flow of commodity i at node n and time t

Kn ∈ R+ new flow capacity of node n

en t ∈ R+ inventory level of storage node n at time t

En ∈ R+ new stock capacity of storage node n

Frontiers in Energy Research | www.frontiersin.org 22 June 2021 | Volume 9 | Article 671279

  • Remote Renewable Hubs for Carbon-Neutral Synthetic Fuel Production
    • 1. Introduction
    • 2. Literature Review
    • 3. Methodology
      • 3.1. Graph-Based Optimisation Modelling Framework
      • 3.2. Application to Energy Supply Chains
        • 3.2.1. Modelling Assumptions
        • 3.2.2. Nodes
        • 3.2.3. Hyperedges
      • 3.3. Implementation
    • 4. Case Study
      • 4.1. System Configuration
        • 4.1.1. Conversion Nodes
        • 4.1.2. Storage Nodes
        • 4.1.3. Conservation Hyperedges
      • 4.2. Scenarios
    • 5. Results
      • 5.1. Reference Scenario
      • 5.2. Sensitivity Analysis
      • 5.3. Discussion
    • 6. Conclusion and Future Work
    • Data Availability Statement
    • Author Contributions
    • Acknowledgments
    • References
    • Nomenclature

Impact of Renewable Energy.PDF

79

The Impact of Renewable Energy Forecasts on Intraday Electricity Prices

Sergei Kulakova and Florian Zielb

abstract

In this paper we study the impact of errors in wind and solar power forecasts on intraday electricity prices. We develop a novel econometric model which is based on day-ahead wholesale auction curves data and errors in wind and solar power forecasts. The model shifts day-ahead supply curves to calculate intraday prices. We apply our model to the German EPEX SPOT data. Our model outperforms both linear and non-linear benchmarks. Our study allows us to conclude that errors in renewable energy forecasts exert a non-linear impact on intraday prices. We demonstrate that additional wind and solar power capacities induce non-linear changes in the intraday price volatility. Finally, we comment on economical and policy implications of our findings.

Keywords: Energy economics, Energy forecasting, Energy Policy, Forecasting and Prediction Methods, Renewable Resources

https://doi.org/10.5547/2160-5890.10.1.skul

f 1. INTRODUCTION g

1.1 Literature Review

In an effort to curb climate change, contemporary government policies actively promote, support and even force the increased use of clean power. As a result, structural changes to energy systems are occurring. Moreover, wind and solar power is main driver of the energy transition. Multiple indicators predict their booming future due to their continuously falling costs, widespread availability and low global warming potential. Yet, energy harnessed by wind turbines or photovoltaic panels is intermittent. This signifies the importance of load and price forecasting.

The variability of the sun and wind energy is critical in the German EPEX SPOT. A sim- plified temporal trading scheme of this energy exchange is depicted in Figure 1. Two markets are of particular interest to us: day-ahead and continuous intraday. They differ in their tempo- ral proximity to point t of physical electricity delivery and in their microstructures. The former market is a non-continuous limit order book auction conducted at 12:00 a day prior to t. The latter one is a continuous trading market which closes 30 minutes prior to t.

Note that prices in both markets are announced before the physical delivery of electricity occurs. Therefore, market prices are based on wind and solar power supply forecasts. Further-

a Corresponding author. Chair of Environmental Economics and Renewable Energies, House of Energy Markets and Finance, University of Dusiburg-Essen. Send correspondence to House of Energy Markets, University of Duisburg-Essen, Universitätssträße 2, 45141, Essen, Germany. E-mail: [email protected]. Phone: +49 171 780 22 14. b House of Energy Markets and Finance, University of Duisburg-Essen.

Economics of Energy & Environmental Policy, Vol. 10, No. 1. Copyright  2021 by the IAEE. All rights reserved.

80 Economics of Energy & Environmental Policy

Copyright © 2021 by the IAEE. All rights reserved.

more, forecasts are usually more precise in the intraday market. Hence, prices in the intraday market are closer to the actual fundamental equilibrium if market participants actively trade in both markets (see e.g. Weber (2010) or Pape (2017)). As a consequence, the influence of forecast errors on intraday prices drops the closer the trading occurs to the point of actual electricity delivery. In fact, the decrease of forecast errors can be non-linear as e.g. the work in Kühnert (2016), Larson et al. (2016) or Ahlstrom et al. (2013) suggest.1 Moreover, note that the impact of forecast errors on intraday prices may depend on the size of the error. Therefore, this paper attempts to prove that the impact of forecast errors on intraday prices is non-linear.

FIGURE 1 A simplified trading scheme of the German EPEX SPOT.

The influence of forecast errors on electricity market prices has already been given a thor- ough attention in the academic literature. The work in Von Roon and Wagner (2009) demon- strated the importance of errors in wind forecasts and attempted to measure the impact of these errors on intraday prices. In a more recent study, Kiesel and Paraschiv (2017) show that intraday prices are indeed affected by errors in renewable energy forecasts and are even sensible to the signs of forecast errors. The work in Pape (2017) shows that errors in wind and solar power forecasts affect not only intraday prices, but also controllable electricity producers. The work in Pape (2018) demonstrates that balancing day-ahead forecast errors by trading in the intraday market is preferred to using imbalancing mechanisms. Garnier and Madlener (2015) take the perspective of an operator who tries to compensate forecast errors in a continuous intraday market. In doing so, the authors of the paper attempt to determine optimal timing and volume decisions of the operator. The study of Gürtler and Paulsen (2018) uses panel data analysis and supports the conclusions drawn by Kiesel and Paraschiv (2017). The key findings of the latter paper regarding asymmetries are disputed by Ziel (2017) who agrees that forecast errors influence intraday market prices but shows that the asymmetric effects are insignificant.

Non-linearities in the impact of forecast errors have also been analyzed in the academic literature. By investigating the Nord Pool data, Jónsson et al. (2010) show that electricity prices react on adjustments in wind power predictions more strongly during the day than during the night. Furthermore, the study claims that the price reaction becomes smaller when the level of wind penetration rises. Hagemann (2013) relies on the German electricity data and demon- strates that the impact of wind forecasts on electricity prices is more significant from midnight to 8 a.m. due to office hours and forecast updates arriving. Moreover, many fundamental mod-

1. Please note that in this paper we always mean forecast errors in wind and solar power forecasts when we refer to forecast errors.

The Impact of Renewable Energy Forecasts on Intraday Electricity Prices 81

Copyright © 2021 by the IAEE. All rights reserved.

els which were applied to study the influence of forecast errors (e.g. Goodarzi et al. (2019) or Zareipour et al. (2009) besides the above mentioned) are non-linear by nature.

Nevertheless, the non-linear impact of errors in wind and solar power forecasts particularly on intraday electricity prices has not yet been studied. The present paper attempts to solve this problem and demonstrates that the impact of forecast errors on intraday prices depends on e.g. the sector of a merit-order curve in which an intraday price is realized. More importantly, one of key innovations of the present paper is an auction-curves-based non-linear econometric model which we develop. The core of the model is built around manipulations with empirical supply and demand curves (also known as sale and purchase curves) recorded in a day-ahead wholesale electricity market. Furthermore, we show that forecast errors influence the volatility of intraday prices in a non-linear manner.

The paper is organized as follows. In the second part of the present section we comment on our idea and provide an intuitive description of our auction-curves-based model. The first part of section 2 is devoted to the data description. The second part of section 2 comments on a method to transform wholesale supply and demand curves into an equilibrium with a per- fectly inelastic demand curve. Section 3 is dedicated to the description of our models. Section 4 presents the obtained results. More specifically, we discuss the results and features of our model, show the out-of-sample evaluation of the models and construct a numerical example to demonstrate the impact of forecast errors on the volatility of intraday prices. Furthermore, subsection 4.4 elaborates on economic and policy implications of our research. Section 5 con- cludes the paper.

1.2 Main Idea

To elaborate on the main intuition behind our idea, we assume a toy example of an imag- inary electricity market. This example is depicted in Figure 2. We suppose that the blue curves denote supply curves recorded in a wholesale day-ahead market. The green curves are the cor- responding hypothetical intraday supply curves. For the matter of simplicity we assume that the distance between the blue and the green curves depends only on a forecast error of 2500 MW. Note that the blue curves are located to the right of the green curves. It follows that the actual amount of electricity was overestimated in the day-ahead market. Naturally, the blue curves would be shifted towards the green ones and the day-ahead price would be closer to the intraday price if the forecast in the day-ahead market would be more precise.

Note that the two curves, their shapes and the distances between them are identical on both sides of Figure 2. The only difference between the two sides of Figure 2 is the realized demand size. The intraday price is more different to the day-ahead price in Figure 2a (low demand case) than in Figure 2b (high demand case), even though the size of the forecast error is the same. Figure 2 thus demonstrates that forecast errors may influence intraday prices in a non-linear manner. As the Figure suggests, the sector of the supply curve in which the price is realized or the non-linear shape of the merit-order curve are factors which may determine the impact of forecast errors on intraday prices.

Figure 2 can also be used to discuss the functioning of our novel auction-curves-based econometric model. The model is optimization-based and is not built on the analysis of the day-ahead and intraday price time series. Instead, the model is based on manipulations with empirical wholesale supply and demand curves. More specifically, our model shifts the whole- sale day-ahead supply curves to approximate the corresponding intraday supply curves. The magnitudes and the directions of the shifts are determined by (a) errors in wind and solar

82 Economics of Energy & Environmental Policy

Copyright © 2021 by the IAEE. All rights reserved.

power forecasts and (b) absolute amounts of wind and solar power generated at the moment of delivery.2 To optimally adjust the shift size, a non-linear optimization technique is applied. In other words, we add or subtract the adjusted forecast errors from the day-ahead supply. As a result, the day-ahead wholesale supply curves are shifted horizontally. The shifted day-ahead supply curves are thus our approximations of the intraday supply curves. Naturally, the in- tersections of the shifted day-ahead supply curves with the demand curves coincide with the intraday prices.3 Therefore, our auction-curves-based model provides us with an innovative modeling approach of electricity prices. As opposed to conventional quadratic models, our model allows us to interpret the impact of each of the model’s parameters. Hence, to draw further conclusions, we will compare our auction-curves-based model and its modifications with similarly parametrized linear, quadratic and combined benchmarks.

2. We do not have the data regarding the wind and solar output at time points shortly before the delivery. Thus, to carry out our study, we focus on the actually realized data. Hence, we can not use our model for intraday price forecasting. However, our model still allows us to trace the non-linear impact of forecast errors on intraday prices.

3. The fact that intersections of the auction curves yield equilibrium prices is the core of fundamental models (also known as structural models) elaborated by e.g. Howison and Coulon (2009) or Carmona et al. (2013). Moreover, the work in Ziel and Steinert (2016) provided a lengthy analysis of the intersections of the day-ahead wholesale auction curves. They showed that in 64 % of the cases the intersections between the auction curves are identical to the reported prices, in 89 % the error is less than 0.1 EUR and in 99.8 % the error is less than 1 EUR. The reason for the errors is e.g. the presence of block or other complex orders which are neglected in the auction curves.

FIGURE 2 A toy example of an electricity market with the distance between day-ahead and intraday supply curves being dependent only on a forecast error of 2500 MW.

(a) Low demand scenario with demand Dlow,t = 5000 MW

(b) High demand scenario with demand Dhigh,t = 23000 MW

The Impact of Renewable Energy Forecasts on Intraday Electricity Prices 83

Copyright © 2021 by the IAEE. All rights reserved.

f 2. INPUTS OF THE MODELS g

2.1 Data Description

Following the introductory section of the present paper, we focus on the German-Austrian EPEX SPOT market and study the period between 01.01.2016 and 31.12.2017 (see EPEX (2019a) and EPEX (2019b)). We denote day-ahead prices by DAtP and hourly weighted average intraday prices (usually referred to as VWAP by the EPEX SPOT) by PtID.4 As the regulation of the exchange suggests, prices in the day-ahead market are bound with max = 3000P EUR from above and with −min = 500P EUR from below. In turn, the price range in the intraday market comprises –9.999 EUR to 9.999 EUR. Moreover, besides the price data, from the EPEX SPOT we also obtain the data regarding the day-ahead wholesale supply and demand curves.

Furthermore, from ENTSO-E Transparency (see ENTSO-E (2019)) we have the data regarding the forecasted and actual wind and solar power supply. We index day-ahead forecasts of wind and solar generation by F and the corresponding realized values by A. Hence, there are two pairs of parameters: W F and W A, S F and S A, where W and S stand for wind and solar power, respectively. We compute forecast errors as W Δ = W A – W F and S Δ = S A – S F.5

Figures 3 and 4 were constructed to present the data. The former Figure demonstrates an example of the wholesale supply and demand curves recorded in the German day-ahead mar- ket. The latter Figure shows the dynamics of day-ahead and hourly weighted average intraday prices, the total amount of wind and solar power supply and the forecast errors in wind and solar power output. From Figure 4 it can be seen that day-ahead prices may deviate from in- tra-day prices if the amount of wind or solar power was wrongly predicted. More specifically, the segment bounded by two red lines demonstrates that the intra-day prices can be smaller than the day-ahead prices (upper plot) when the actual amount of wind power (lower plot) was underestimated in the day-ahead market.

2.2 Transformation of Empirical Supply and Demand Curves

The toy example illustrated in Figure 2 plots a market equilibrium with a perfectly inelastic demand curve. Such setting allows us to shift the supply curve back and forward because mar- ket participants remain insensitive to the equilibrium price. However, Figure 3 demonstrates that the actual wholesale supply and demand curves are elastic. Shifting elastic curves may lead to ambiguous results because market participants may act differently depending on the equilibrium price.

To avoid this problem, we will use a method developed by Coulon et al. (2014). This method allows us to transform actual wholesale auction curves into a market equilibrium with perfectly inelastic demand. The equilibrium prices remain unchanged before and after the transformation, while the corresponding volume sizes increase.

The economic reasoning behind the transformation was elaborated at length in the orig- inal paper by Coulon et al. (2014) or in e.g. Kulakov and Ziel (2019). Moreover, the paper

4. Our choice of the weighted average intraday prices is motivated by the findings of von Luckner et al. (2017). Following the conclusions of this paper, the majority of orders in the continuous intraday market arrive shortly before the gate closure.

5. Please note that the data as to the renewable energy supply was provided in the quarter-hourly resolution. We used simple arithmetic averages to adjust the granularity of the data to the hourly resolution. We did not extrapolate missing data and omitted the time points in which a price-volume observation was not available in at least one of the datasets. We neglected daylight saving times within the current research and did not make any clock-change adjustments. Furthermore, we rounded the prices to two decimal places to spare computational time.

84 Economics of Energy & Environmental Policy

Copyright © 2021 by the IAEE. All rights reserved.

by Knaut and Paulus (2016) can help understand the reasoning because this paper describes trading strategies of market participants in a wholesale energy market. Generally speaking , the idea is to transfer all elasticities from the demand to the supply side. Following Coulon et al. (2014), there exists two bilateral markets for electricity trading: an OTC market and a whole- sale market. Since prices in the latter market can be lower than costs of electricity generation, arbitrage opportunities exist in the wholesale market. Therefore, market participants can try to buy electricity in the wholesale market instead of generating it. As a result, some orders in the wholesale demand curve can be of arbitrage nature. Therefore, even though being based on a somewhat relaxed assumption, transferring all demand elasticities to the supply side allows us to obtain a more stable and predictable market equilibrium with a perfectly inelastic demand curve.

Following the original paper, the expression for the inelastic demand curve can be written as

−1 min= ( )

inelastic t tDem WSDem P (1)

where the demand curve in the wholesale market at time point t is denoted by WSDem and −min = 500P as follows from the regulation of the EPEX SPOT. In turn, the equation for the

transformed inverse supply curve can be written as

− − − −+ −  

1 1 1 1 min

Contribution of the wholesale supply curve Contribution of the wholesale demand cruve

( ) = ( ) ( ) ( )t t t tSup z WSSup z WSDem P WSDem z (2)

FIGURE 3 Wholesale supply and demand curves in the German electricity market on 2017-04-02

at 09:00:00. The Figure shows the entire auction curves (left) and the same two curves with a particular focus on their intersection (right).

The Impact of Renewable Energy Forecasts on Intraday Electricity Prices 85

Copyright © 2021 by the IAEE. All rights reserved.

where the supply curve in the wholesale market a time point t is denoted by WSSup. More specifically, the transformed version of the supply curve Supt at time point t consists of two parts: the initial wholesale supply curve −1( )tWSSup z and an adjustment in the form of the horizontally mirrored wholesale demand curve − −−1 1min( ) ( )t tWSDem P WSDem z . Therefore, fol- lowing the equation and the above described intuition, the transformed supply curve Supt incorporates all demand elasticities additionally to the initial supply volumes. Furthermore, the supply and demand curves as defined in equations 1 and 2 allow the equilibrium price to remain unchanged before and after the transformation. Furthermore, please note that equation 2 defines ( )tSup z automatically since the function is monotonic. Finally, an example of the transformed wholesale auction curves is illustrated in Figure 5.

FIGURE 4 The dynamics of day-ahead and intra-day prices (upper plot), total generation of wind and solar

energy (middle plot) and differences between actual and predicted wind and solar generation loads (lower plot) for a one-week sample from August, 28 to September, 03, 2017.

86 Economics of Energy & Environmental Policy

Copyright © 2021 by the IAEE. All rights reserved.

f 3. METHODOLOGY g

3.1 Benchmark Models

The benchmark models we use in the present study are state-of-the-art approaches in the field of intraday electricity price forecasting (see e.g. Narajewski and Ziel (2019) or Uniejewski et al. (2019)). However, we neglect autoregressive components in our models to measure the impact of forecast errors on intraday prices.

The first benchmark model is a typical naive model P Pt naive

t DA

t= ,�� where Pt DA stands for

the day-ahead price at time point t and εt is an error term. The other models in our paper are based on a component denoted by Zt. This component is written in the following vector form and includes 6 elements

( )∆− ∆ ∆− ∆= , , , , ,A At t t t t t tW W S S W SZ (3) where Wt

∆ and St ∆ are the errors in wind and solar supply forecasts at t, respectively;

∆ ∆− −= max( ,0)t tW W and ∆ ∆− −= max( ,0)t tS S stand for the negative errors in wind and solar

supply forecasts at t, respectively; Wt A and St

A denote the absolute volumes of the harvested wind and solar energy at t, respectively. Note that we model negative forecast errors separately analogously to what was done in e.g. the papers by Soysal et al. (2017), Kiesel and Paraschiv (2017) or Ziel (2017). Vector Zt will thus be incorporated into several models and differ- ent estimation techniques will be used to determine the corresponding vectors of parameters

β β ′β 0= ( ,..., )i . The first linear benchmark model lm1 can be characterized as follows

β ε′+ + +1 0 1:6= . lm DA

t t t tP Pβ Z (4)

FIGURE 5 A wholesale market equilibrium on 2017-04-02 08-00-00 CET (left)

vs. its manipulated form with the inelastic demand curve (right).

The Impact of Renewable Energy Forecasts on Intraday Electricity Prices 87

Copyright © 2021 by the IAEE. All rights reserved.

The second linear model lm2 is analogous to the first one, save for the fact that term DA

tP is assigned with its own β coefficient.

β β ε′+ + +2 0 1:6 7= . lm DA

t t t tP Pβ Z (5)

It follows that the model in equation 4 is a special case of the model in equation 5 with β7 = 1. Moreover, the former model assumes a perfect cointegration of intraday and day-ahead prices. In fact the use of two similar models is justified because there is no clear academic consensus about which of them usually performs better (see e.g. Soysal et al. (2017) or Narajewski and Ziel (2019)).

The last benchmark model is an extension of model lm2 with quadratic terms. We will denote this model by qlm and make the following statement

( )β β β ε′ ′+ + + + + 20 1:6 7 8:13 14= ( ) ,qlm DA DAt t t t t t tP P Pβ βZ Z Z (6)

where  denotes the Hadamard or entrywise product.

3.2 Auction-Curves-Based Models

Following the intuition elaborated in section 1.2, our auction-curves-based models do not focus on the day-ahead and intraday price time series. Instead, our models are based on manipulations with the wholesale auction curves. From this perspective, our models follow a novel approach, but extend the family of econometric models developed in e.g. Ziel and Stein- ert (2016), Dillig et al. (2016), Shah and Lisi (2018) or Kulakov and Ziel (2019). Moreover, we keep the same parameter specification in auction-curves-based and benchmark models to allow for the comparison of the models.

We will denote the first auction-curves-based model by nlm and define this model as fol- lows

β ε′− − +15:21 15 16:21( ) = ( )) nlm inelastic

t t t t tP Sup Demβ β Z (7)

where the price at t is an intersection of the shifted day-ahead supply curve with the inelastic demand curve. Furthermore, to estimate the optimal vector of coefficients β15:21, we solve a non-linear least squares problem in the form 

∈ −

β β β 27 15:21= ( ( ))

ID nlm nlm t targ min P P . In doing

so, we use built-in R optimizer optim with default settings. Our second model lnlm incorporates linear model lm2 and auction-curves-based model

nlm. The price equation of the model can be written as follows

β β β β ε −

′ ′+ + + − − +  

0:7,15:22 0 1:6 7 22 15 16:21 linear component non linear component

( ) = ( )lnlm DA inelastict t t t t t tP P Sup Demβ β βZ Z (8)

where the linear component coincides with the price produced by linear model lm2 and the non-linear component is the price nlmP . Writing the corresponding non-linear least squares problem yields 

β∈ − 216 0:7,15:22= ( ( )) .

ID lnlm lnlm t targ min P Pβ β

88 Economics of Energy & Environmental Policy

Copyright © 2021 by the IAEE. All rights reserved.

3.3 Combined Models

Following e.g. Ziel andWeron (2018), simply combining price models may yield further improvements of their performance. Therefore, additionally to the benchmark and auction- curves-based models, we also consider some of their equally weighted linear combinations. Please note that we report only on those models which were better compared to the non-com- bined models. First, model clq is a combination of linear model lm2 and quadratic model qlm with +2= 0.5 0.5lmclq qlmt t tP P P . We use model lm2 and not lm1 since the former one shows better performance with respect to its MAE and RMSE value. Second model cnq has equation P P Pt

cnq t nlm

t qlm= 0.5 0.5+ , third model clnq can be represented as P P Pt

clnq t lnlm

t qlm= 0.5 0.5 .+

f 4. RESULTS g

4.1 Model Analysis

The obtained β-coefficients for the year 2017 are summarized in Table 2. Please note that β-coefficients of linear models lm1 and lm2 and quadratic model qlm are primarily negative. These findings are consistent with those of e.g. Ziel (2017), Kiesel and Paraschiv (2017), Clò et al. (2015), Ketterer (2014) or Gürtler and Paulsen (2018) where the signs of the coefficients are similar to the ones we obtained. Hence, when the sizes of e.g. negative forecast errors grow, market participants expect a lower electricity supply. As a result, market prices rise and thus β-coefficients are negative.

Yet, the coefficients of the auction-curves-based models are primarily positive. The higher the magnitude of e.g. a positive forecast error is, the more will the merit-order be shifted to the right, and thus the lower the prices are (see e.g. Neubarth et al. (2006), Cludius et al. (2014), Ketterer (2014), Kiesel and Paraschiv (2017), Fürsch et al. (2012) or Roldan-Fernandez et al. (2016)). From this perspective, the intuition behind the functioning of both linear and auc- tion-curves-based models is the same.

The application of model nlm to the real data is illustrated in Figure 6. The areas high- lighted in various colors demonstrate the shift sizes induced by the components of the model. The presence of the shaded areas to the left of the intraday supply curve indicates that the shift was partially negative. Furthermore, the intersection of the shifted supply curve with the inelastic demand curve yields price nlmtP .

Moreover, we constructed Figure 7 as an example to show the differences in the impacts of positive and negative forecast errors on intraday prices. Both sides of the Figure plot the day-ahead supply curve (blue) recorded on 2017-04-19 at 10:00:00 and simulated shifts of this curve. The green curves show the shifts induced by negative forecast errors, the purple curves by positive. We used the framework of model nlm to build up the Figure. Both sides of the Figure incorporate a low demand scenario with Dlow,t = 37440 MW and a high demand scenario with Dhigh,t = 56740 MW. We assume that the absolute amounts of wind and solar power output were equal to = 15000AtW MW and = 15000

A tS MW. Figure 7(a) shows a sce-

nario with forecast errors only in wind forecasts, i.e. Wt �� �= 5000 MW, Wt

∆ = 5000 MW and St ∆ = 0 MW. On the contrary, Figure 7(b) shows a scenario with St

�� �= 5000 MW, St ∆ = 5000

MW and Wt ∆ = 0 MW.

From Figure 7 it appears clear that the horizontal distance between the green and blue curves is greater than the corresponding distance between the blue and purple curves. Follow- ing Table 2, coefficients Wt

�� and St �� are positive. Therefore, the impact of negative forecast

The Impact of Renewable Energy Forecasts on Intraday Electricity Prices 89

Copyright © 2021 by the IAEE. All rights reserved.

errors on the shift of the day-ahead curves is greater than the impact of positive errors. As a result, the green curves are moved further from the initial day-ahead supply curves. From the economic standpoint, we can argue that negative forecast errors can lead to the use of positive reserve power capacities . Running additional power plants is costly, which is why negative errors can exert a greater impact on intraday prices.

Moreover, it appears clear that the horizontal distances between the green, blue and purple curves are greater in Figure 7(b) than in Figure 7(a). In other words, an error in a solar power forecast causes a potentially greater impact on intraday prices than an equally sized impact in a wind power forecast due to solar occurring on hours with usually higher load (i.e. on peak hours). However, large forecast errors (as e.g. of size S ∆ = 5000 MW or S �� �= 5000 MW in our example) are relatively rare compared to similarly sized errors in wind forecasts due to a smaller amount of the generated solar output. For example, in our sample (for both years 2016 and 2017) the absolute mean error (MAE) in wind forecasts was about 1000 MW, for solar 330 MW. Thus, the magnitude of an absolute average error in a solar forecast was roughly a third of that of a wind forecast. However, the corresponding mean absolute percentage error (MAPE) values were about 10% for the wind power and about 8% for the solar power.

Furthermore, we also analyzed asymmetries in wind and solar power forecasts. Figure 8 plots the corresponding β-coefficients for the negative parts of the errors. Please note that the coefficients of linear model lm2 (left side of the Figure) are negative, the coefficients of auction- curves-based model nlm (right side of the Figure) are positive. More importantly, both sides of

FIGURE 6 Price nlmtP on 2017-04-19 at 10:00:00 as the result of shifting the

transformed day-ahead supply curve to the left.

90 Economics of Energy & Environmental Policy

Copyright © 2021 by the IAEE. All rights reserved.

Figure 8 show that the influence of negative forecast errors tends to drop over the year 2017. Similar findings are in e.g. Gürtler and Paulsen (2018).

4.2 Out-of-Sample Evaluation

To evaluate the performance of the models, we first used the MAE and RMSE mea- sures with the following specifications

−∑∑ 24

,, =1 =

1 MAE = | |

24

D IDID d hd h

d d h

P P D

and

( )−∑∑ 224

,, =1 =1

1 RMSE =

24

D IDID d hd h

d h

P P D

where = 364D and h is a hour index. Hence, we used a rolling window study with 365 in-sam- ple observations (year 2016) and 364 out-of-sample observations (year 2017). The window size was equal to 24 hours.

The results of the MAE and RMSE tests are summarized in Table 1. The Table allows us to see that model lm2 has lower MAE and RMSE values than model lm1. Model lm2 was thus used in model lnlm. Furthermore, model nlm does not surpass the linear models and quadratic

FIGURE 7 The day-ahead curve (blue) recorded on 2017-04-19 10:00:00 and its simulated shifts as results

of positive and negative forecast errors for a low demand case with Dlow t, = 37440 MW and Pt

DA = 6.48− and a high demand case with Dhigh t, = 56740 MW and = 51.25 DA

tP .

(a) WtΔ– = –5000 MW (green curve), WtΔ = 5000 MW (purple curve) with StΔ = 0 MW

(b) StΔ– = –5000 MW (green curve), StΔ = 5000 MW (purple curve) with WtΔ = 0 MW

The Impact of Renewable Energy Forecasts on Intraday Electricity Prices 91

Copyright © 2021 by the IAEE. All rights reserved.

model qlm. Our model lnlm performs better than model qlm with respect to the RMSE mea- sure and has a very similar MAE value. In turn, combined model clq fails to beat model qlm. Moreover, the MAE value of combined model cnq and the RMSE value of model cnlq are best compared to those of the other models .

TABLE 1 MAE and RMSE Values of the Models

Naive lm lm qlm nlm lnlm clq cnq clnq

MAE 4.884 4.294 4.277 4.242 4.336 4.247 4.244 4.198 4.185 RMSE 8.048 7.352 7.286 7.103 7.123 7.097 7.171 6.984 7.010

To determine best model among the ones considered, we used two types of the DM-tests. The first one is a multivariate version of the DM-test that provides one statistics on the grounds of 24-dimensional vector of errors. This approach was described in Ziel et al. (2016). The sec- ond one is the standard univariate DM-test that evaluates models in each of the 24 hours as defined in Diebold (2015). To specify the parameters of the multivariate DM-test, we define the loss differential between models  and B on day d as ϕ ϕ ϕδ − B  B, , , ,=d d dL L where

ω ϕ, dL is the

loss function of model ω = ,B on day d with ϕ = 1,2 to compare the models with respect to ⋅ 1-norm (ϕ = 1) and Euclidian norm ⋅ 2 (ϕ = 2). For the univariate DM-test, we define the

loss differential between models  and B at hour h on day d as ϕ ϕ ϕδ − B  B, , , ,, , ,=d h d h d hL L where ω ϕ,

,d hL

FIGURE 8 β-coefficients for the negative parts of forecast errors for models lm2 and nlm.

92 Economics of Energy & Environmental Policy

Copyright © 2021 by the IAEE. All rights reserved.

is the loss function of model ω = ,B with ϕ = 1,2 to compare the models with respect to ab- solute errors (ϕ = 1) and quadratic errors (ϕ = 2). The loss functions of models  and B in the multivariate and in the univariate settings can be written respectively as

ϕ ω ϕ ω ϕ ω ϕ ω ϕε ε

      ∑

1/24 , ,

, , , =1

ˆ ˆ= | | and =| | ,d d h d h d h h

L L ϕ

where ω = ,B and ϕ = 1,2. The test statistics of the multivariate and univariate DM-tests are respectively given by

ϕ ϕ δ δ

δ δ σ σ

 B  B  B  B

 B  B

, , , , , ,

, , , , = and = hh

h

t t

where δ δ∑ B  B, ,1 =1= D

D dd and δ δ∑ B  B, ,1 ,=1=

D Dh d hd

. Moreover, δσ  B, , and δσ  B, ,h denote sample standard deviations of δ  B,d and δ

 B, ,d h , respectively.

The comparison of all models with each other according to the multivariate DM-test is illustrated in Figure 9. More specifically, Figure 9(a) shows the comparison with ϕ = 1, Figure 9(b) with ϕ = 2. As Figure 9 suggests, linear model lm2 performs better than model lm1. Fur- thermore, neither quadratic benchmark qlm nor models nml and lnlm are significantly better than model lm2 when ϕ = 1. However, model qlm shows better performance than the linear models and than model nlm when ϕ = 2. In turn, the combined models tend to show overall best performance. More specifically, model clq outperforms models lm2 and lnlm. Models cnq and clnq outperform the other models in the comparison, however, none of the two can sig- nificantly outperform another. Hence, Figure 9 allows us to conclude that the combinations of linear and non-linear models show overall best performance among our models.

FIGURE 9 Results of the multivariate DM-test with ϕ = 1 (left) and with ϕ = 2 (right).

(a) Multivariate DM-test with ϕ = 1 (b) Multivariate DM-test with ϕ = 2

Figure 10 illustrates the hourly DM-test comparisons of models clnq (Figure 10(a)) and lnlm (Figure 10(b)) against our best benchmark model qlm. Given the 5% confidence inter- val, we see that model clnq outperforms model qlm during several hours of the day. More importantly, as Figure 9 shows, model clnq shows overall better performance than model qlm.

The Impact of Renewable Energy Forecasts on Intraday Electricity Prices 93

Copyright © 2021 by the IAEE. All rights reserved.

Therefore, the overall performance improvement was achieved due to the fact that model clnq performs better during several hours of the day and is not significantly worse during the other hours. Hence, we show that combining our auction-curves-based model lnlm with the qua- dratic benchmark allows us to achieve a significantly better result. Moreover, Figure 10(b) demonstrates that model lnlm beats model qlm during two peak hours of the day when ϕ = 1. However, model qlm is almost better than model lnlm at hour 5 when ϕ = 1 and is better at hour 21 when ϕ = 2. Hence, given the conclusion drawn from Figure 9, model lnlm is not significantly better than model qlm.

The above findings have the following implications. First, our model clnq can be applied successfully to model intraday prices. Despite being unconventional, the model yields similar (and even superior) performance relative to quadratic model qlm. Furthermore, our model allows for a straightforward interpretation of results . As opposed to quadratic model qlm, we can easily interpret the influence of each of the considered parameters by studying the contri- butions of each parameter to the shift size. Second, given the fact that the main components of our models are forecast errors in wind and solar power supply and absolute amounts of wind and solar power, we can conclude that the impact of forecast errors on intraday prices is non-linear. This holds because model qlm outperforms the linear models and because our combined auction-curves-based models cnq and clnq shows a better performance even com- pared to quadratic model qlm. Hence, as was mentioned earlier, the non-linear shape of the merit order curve and the sector of this curve in which the equilibrium price is realized are possible reasons for the non-linear impact of forecast errors on intraday electricity prices. In fact, the exact shape of the actual merit order curves remains unknown when both day-ahead and intraday prices are established. Therefore, wind and solar forecast errors cause a non-linear impact on intraday market prices.6

4.3 Forecast Errors and Volatility of Intraday Prices

Following e.g. Clò et al. (2015), additional wind and solar power capacities not only in- duced a merit-order effect in Italy, but also increased the volatility of electricity prices. These findings are consistent with the work in Woo et al. (2011) where a similar study is conducted for the electricity market in Texas. The authors of the latter paper show that the rising wind generation induces an increase in the variance of 15-minutes electricity spot prices. The cor- responding analysis of the German electricity market is provided in Ketterer (2014). The con- clusions of this paper show that a rise in the volatility of electricity prices may stem from the growing penetration of renewable resources.

In the present section, we will develop a numerical example to show that the rising amount of wind or solar power capacities in fact increases the volatility of intraday prices. More im- portantly, our example demonstrates that the growth in the price volatility is driven by rising forecast errors and is non-linear.

We assume an operating onshore wind power plant and suppose that this power plant is extended with additional capacities. Let Wt be the amount of energy currently harvested by the plant,  tW an incremental wind supply from additional wind power capacities and γ ≥ 0 a scale

6. Following e.g. Kyle (1985) or de Frutos and Manzano (2014), higher market liquidity implies lower risk premia. Therefore, the impact of a higher forecast error on intraday prices may be lower in a deeper market. Yet, deriving a scientific proof of the statement is a subject of another study.

94 Economics of Energy & Environmental Policy

Copyright © 2021 by the IAEE. All rights reserved.

factor. As the conventional portfolio theory suggests (see e.g. [Berk and DeMarzo, 2007]), the generation variance of the extended wind power plant can be computed as follows

 Var Var Var Var Varγ γρ γ+ + + 2 ,

[ ] = [ ] 2 ( ) ( ) [ ]t t tt t tW W tt W W W W W W

where ρ denotes the correlation coefficient between the outputs of the old and new capacities. Assuming that variances Var[ ]tW and Var[ ]tW are equal allows the above expression to be represented as follows

SD SD ρ γ

γ γρ γ −

+ + + 

2 ,

greater than 1 if > /2 ,

[ ] = [ ] 1 2tt t W W tt W W tt

W W W (9)

Hence, the standard deviation of the electricity output of the combined power plant increases when ρ γ−> / 2 (which is especially the case for ρ > 0). Moreover, the unpredictability of the volatile energy output implies that forecast errors, too, are high. In fact, Weber (2010) suggests that the magnitude and amount of forecast errors rises together with expanding wind and solar power capacities.

To show that growing forecast errors may induce a non-linear increase in the volatility of electricity prices, we consider a numerical example. As Weber (2010) suggests, we can assume that forecast errors are proportional to the standard deviation of the power generation. Given this assumption, we can modify component Zt to test the sensitivity of our models to changes in the amounts of wind and solar power capacities. In line with equation 9, we suppose that our models are applied to a market with both old and additional wind and solar power ca- pacities. We assume that modified component Zt is denoted by Zt,γ and can be represented as follows

FIGURE 10 Results of the DM-test comparison of models clnq vs. qlm (left) and lnlm vs. qlm (right)

for each hour of the day for the out-of-sample year 2017. Circles denote p-values with ϕ = 1, squares denote p-values with ϕ = 2.

(a) Models clnq vs. qlm (b) Models lnlm vs. qlm

The Impact of Renewable Energy Forecasts on Intraday Electricity Prices 95

Copyright © 2021 by the IAEE. All rights reserved.

 

∆ ∆ γ γ ρ γ γ ρ γ

− + + + +2 2, , ,= ( 1 2 , 1 2 ,t t W W t W WW W W Wt tt t W WZ

 

∆ ∆γ ρ γ γ ρ γ γ γ− + + + + + +2 2 , ,

1 2 , 1 2 , (1 ) , (1 ))A At S S t S S t W t SS S S St tt t S S W S (10)

To test our models under the new assumption, component Zt was replaced with Zt,γ in two of our models. We decided to compare our best linear model lm2 with auction-curves-based model nlm to demonstrate the differences between linear and non-linear settings. We have chosen γ = 0.1,1,5 for the amounts of additional capacities and ρ = 0,0.5,0.8 as possible cor- relation coefficients to conduct the study. Table 3 summarizes the standard deviations of prices P lm2 and P nlm relative to the respective baseline models with γ = 0 and ρ = 0. Table 4 shows rela- tive differences between the true and the modeled intraday prices for the 0.1% quantile relative to the respective baseline models with γ = 0 and ρ = 0. Table 5 shows the same as Table 4 but for the 99.9% quantile. Hence, selecting different values of scaling coefficients γ and ρ allows us to test and re-evaluate the model presented in this section under various assumptions. As a result, we show that e.g. a linear increase in the values of γ (i.e. a linear increase in additional wind and solar power capacities) induces a non-linear increase in the standard deviations of electricity prices. Therefore, given that the amount of forecast errors and the sizes of wind and solar power capacities are proportional, the model elaborated in this section is yet another confirmation of the fact that the impact of forecast errors in wind and solar power forecasts on electricity prices is non-linear.

Note that the values in all three tables tend to increase the greater the additional power capacities are (i.e. the bigger γ is) or the stronger the correlation between the old and the new capacities is (i.e. the higher ρ is). More importantly, the values grow much quicker in model nlm. This indicates that the increase in the values is non-linear. For example, when considering Table 3, non-linear effects can be seen by comparing the difference in the values produced by models lm2 and nlm for γ = 1, ρ = 0.8, and γ = 5, ρ = 0.8, especially for W+S case. Therefore, given the assumption that forecast errors are proportional to the standard deviations, Table 3 shows that the volatility of intraday prices increases when forecast errors grow. Hence, the numerical example allows us to conclude that forecast errors not only have a non-linear impact on intraday prices, but also influence the intraday price volatility in a non-linear manner.

Furthermore, as the work in e.g. Ziel (2017) and Monforti et al. (2014) shows, wind and solar forecast errors are nearly uncorrelated and the power output from wind and solar resources is negatively correlated. The model described in this section allows us to account for the effect of these correlations. More specifically, we achieve this effect in the 0.5(W+S) case, where, following equation 9, a part of the variation induced by additional wind and solar capacities is eliminated. Of course, this effect is not present in the other cases and thus the volatility is highest in the W+S case. As a result, Tables 3–5 show that simultaneous build up of wind and solar capacities in the 0.5(W+S) case induces a relatively small change (or even decreases) the volatility of intraday prices, especially at the 0.1% and 99.9% quantiles.

Tables 4 and 5 offer another interesting conclusion. Table 4 shows that both in linear and non-linear models in the case of solar power the values can drop when γ increases from to 1. This observation means that a moderate increase in the solar power capacities lowers the vol- atility of intraday prices at the 0.1% quantile. The drop in the volatility happens because e.g. negative price spikes occur less often due to the increased solar output during the peak hours. A similar effect can be seen in Table 5 which focuses on the 99.9% quantile. More specifically, the values in Table 5 can be lower than 1 in all four cases. However, the observed effect is

96 Economics of Energy & Environmental Policy

Copyright © 2021 by the IAEE. All rights reserved.

stronger in linear model lm2. It follows that moderate increase in wind and solar capacities can be beneficial for lowering the influence of positive price spikes.

The model elaborated in section 4.3 shows that locations of new wind and solar power plants must be selected to minimize the correlations between the outputs of the old and new capacities. This issue has already been investigated on the grounds of empirical data in e.g. Palmintier et al. (2008), Jónsson et al. (2010) and Grams et al. (2017). Moreover, our numer- ical example supports the findings of these papers. In fact, there are several factors which may influence the strength of the correlation. For example, a new wind power plant can be built spatially close to the old one, or a new plant can be built in a place with similar weather condi- tions to the old one. As a result, fluctuations in wind output will influence the power supply of both old and new power plants simultaneously. Therefore, e.g. an unexpectedly small amount of wind in the system will almost equally affect the power supply of both old and new plants. Hence, the volatility of the power supply (and thus of electricity prices) increases. On the other hand, if the outputs of the old and new plants are negatively correlated, a drop in the output of one power plant will be offset by an increase of the output of another. The impact of extreme events on intraday prices and their volatility, too, is lower the lower the correlations between the old and new plants are. As a result, the overall volatility of electricity supply remains lower when the correlations are minimized. Hence, basic conclusions of conventional portfolio the- ory regarding portfolio diversification (see e.g. Cochrane (2009)) hold in this context too.

4.4 Policy Implications

The above described models demonstrate the need to decrease the non-linear impact of forecast errors on electricity prices and their volatility. The problem seems especially important given that the shares of clean power in the worldwide energy mix are expected to grow steadily. Given the setting of our models, the policy implications that we draw in this subsection are more applicable to the case of Europe, especially to the countries participating in the XBID (cross-border intraday initiative). Thus, the following can be done to achieve the reduction of the impact of forecast errors on electricity prices:

a. Regional deployment The model elaborated in section 4.3 and Tables 3–5 show that locations of new wind and solar power plants must be selected to minimize the correlations between the out- puts of the old and new capacities. This holds because lower variability of renewable energy supply will lead to lower forecast errors and thus lower variability of electricity prices. Therefore, our findings are consistent with academic research. More specifically, e.g. Engeland et al. (2017) argue that renewable energies should not be concentrated in several parts of a country and should be spread more equally across the country’s area be- cause spatial diversification of renewable resources on regional and local levels decreases the variability of energy output. Furthermore, as the work in e.g. Drake and Hubacek (2007), Hasche (2010), Handschy et al. (2017), Novacheck and Johnson (2017) or Eis- ing et al. (2020) shows, the variability of wind supply drops when wind farms are spread geographically. Similar conclusions for the case of solar power are drawn in e.g. Mills (2010), Perez et al. (2012) or Perez et al. (2016). Therefore, policy makers can employ a more intensive and directed subsidization of renewable resources in particular regions, or direct the support only on certain regions to trigger regional deployment.

The Impact of Renewable Energy Forecasts on Intraday Electricity Prices 97

Copyright © 2021 by the IAEE. All rights reserved.

b. Diversity of renewable power supply As was mentioned earlier, the model developed in section 4.3 allows us to account for the effect of negative correlation between outputs of wind and solar power. Therefore, we can conclude that maintaining a balance between additional wind and solar ca- pacities is important for offsetting the impact of forecast errors and keeping the prices more stable. Similar conclusions regarding the benefits of simultaneous integration of negatively correlated wind and solar power (as well as diverse wind and solar technolo- gies) can be found in e.g. Widén (2011), Sioshansi and Denholm (2013), Zheng et al. (2019) or Eising et al. (2020) . Besides, expanding the energy mix with further renew- able energy sources (e.g. geothermal or tidal) will induce positive portfolio smoothing effects. These effects lead to a further decrease in the volatility of energy output as described on the grounds of empirical data in e.g. Awerbuch (2006), Engeland et al. (2017), Monforti et al. (2014) or Zipf and Möst (2013).

c. Cross-border interconnections and intraday trading The work in e.g. Association (2009), Scholz (2012), Paternò et al. (2016) or Brown et al. (2016) shows that building the cross-border transmission infrastructure to avoid country-specific bottlenecks is a financially efficient decision. In fact, Schlachtberger et al. (2017) demonstrates that restricting continental transmission expansion in Europe leads to a non-linear cost increase of up to 30%. The work in Grams et al. (2017) sug- gests that collaboration on the European level will decrease the variance of wind power output, whereas the absence of collaboration will imply a further increase in overall wind output variability. Therefore, a greater support (especially a greater centralized support on e.g. the EU-level) should be given to developing the cross-border electricity infrastructure (see e.g. Jacottet (2012) or Puka and Szulecki (2014)). In addition to the cross-border interconnections and for the case of Europe, it is also important to include more countries into the cross-border intraday trading system (XBID). As e.g. Kath (2019) or TGE (2019) argue, expanding the XBID system will allow for an increased European market market liquidity and thus improve overall market and transmission efficiency.

d. Better quality of renewable energy forecasts Given that the impact of forecast errors on intraday prices is non-linear, the quality of renewable energy forecasts should be improved to minimize forecast errors. Of course, general improvements of methodology and computational capacities (see e.g. Chang et al. (2014), Antonanzas et al. (2016) or Voyant et al. (2017)) should be considered in this context. Moreover, a measure suggested in Joos and Staffell (2018) or Pinson (2016) is to establish a centralized platform for information sharing between partic- ipants of energy markets, especially between power plant operators and transmission system operators. Furthermore, an improvement of meteorological models can be ben- eficial for increasing the overall performance of electricity forecasting models (see e.g. Andrade and Bessa (2017) or De Giorgi et al. (2011)).

e. Flexible generation and demand side management Technically speaking and following e.g. Pape et al. (2016), flexibility measures allow the merit-order to stay wider and more elastic for longer periods of time. Thus, as the intuition behind our auction-curves-based models (section 3.2) suggests, the impact of shifting the supply curve on intraday prices will be lower because a greater part of the shift will take place inside of more elastic segments of the merit order. In fact,

98 Economics of Energy & Environmental Policy

Copyright © 2021 by the IAEE. All rights reserved.

measures on the supply side include greater investments into flexible power plants and energy storage capacities (see e.g. Child et al. (2019), Eyer and Corey (2010) or Lund et al. (2015)). Demand side management are discussed at length in a review paper by Palensky and Dietrich (2011) and include e.g. reducing demand during peak hours or load shifting. Examples of the latter are e.g. the use of excess electricity for domestic or industrial purposes, i.e. by replacing CHP units with boilers and electrical heating or by increasing the reliance on electric vehicles.

f. Efficient and transparent renewable energy curtailment Management The practice of electricity curtailment suggests that electricity supply from renewable resources, especially from wind and solar power plants, should be cut off if further elec- tricity infeed from these power plants threatens the stability of the energy system (a review of international practices is presented in Bird et al. (2016)). More specifically for the case of Germany, guidelines for the curtailment are described in a regulation called EinsMan (Einspeisemanagement, see Jacobsen and Schröder (2012) or Bundesnetza- gentur (2018)).However, researchers (see e.g. Ostermann et al. (2019), BDEW (2017), Ketterer (2014) or Joos and Staffell (2018)) argue that centralized public disclosure of the extent and duration of the curtailment (e.g. on the grounds of a prognosis) will fur- ther improve the effectiveness of the curtailment. In fact, information should be made public as soon as possible, especially prior to moments of important decision making (e.g. for prior to the European coupling of the region day-ahead auction at 12:00 CE(S) T the case of Europe), because the announcement will allow traders to adjust their strat- egies for a potential curtailment. As a result, a timely announcement of the information about the curtailment will further decrease the impact of forecast errors on intraday market prices.

f 5. CONCLUSION g

In this paper we studied the impact of errors in wind and solar power forecasts on wholesale intraday electricity prices. To derive our conclusions, we elaborated a novel econo- metric model. Our model is based on manipulations with empirical supply and demand curves recorded in a wholesale electricity market. To compute the intraday price at a given time point, we horizontally shift the corresponding day-ahead supply curve. The magnitude and the di- rection of the shift depend on errors in wind and solar power forecasts and absolute amounts of wind and solar power. The shifted day-ahead supply curve is our approximation of intraday supply curve. The intersection of the approximated intraday supply curve with the demand curve coincides with the intraday price. Given that we can see the contribution of each of the model’s parameters to the shift size, the main advantage of our model is the ease of interpre- tation of the results. Our results indicated that our auction-curves-based model outperforms other models in the study during several hours of the day. The quadratic benchmark performs better than the linear benchmarks and, since the parameters in all of our models included only forecast errors and absolute amounts of wind and solar power, we could conclude that that the impact of forecast errors on intraday prices is non-linear. Based on a numerical example, we show that the impact of forecast errors on the volatility of intraday prices is non-linear too.

Given the results of our study, we argue that it can be efficient to build additional wind and solar power capacities such that the correlations between the old and new plants is minimized. This can be achieved by improving regional deployment and cross-border in-

The Impact of Renewable Energy Forecasts on Intraday Electricity Prices 99

Copyright © 2021 by the IAEE. All rights reserved.

terconnections. Moreover, we argue that it is efficient to minimize forecast errors themselves, introduce better flexibility measures (including flexible generation and storage capacities) and improve the demand side management. Finally, a central and timely announcement of poten- tial curtailment measures (e.g. prior to day-ahead trading.) can further benefit the stability of energy system.

Further improvements of our auction-curves-based model are possible. For example, we only employed the shifts of the day-ahead supply curves to estimate the intraday supply curves. Additionally, we could also shift the demand curve. Moreover, using intraday data shortly before the delivery (and not the actually realized data) will allow the model to be used for intraday price forecasting. Improving the model described in Subsection 4.3 is possible by implementing a more sophisticated correlation structure between the generations of wind and solar power plants.

f APPENDIX g

TABLE 2 The obtained β-coefficients, significance levels are: ● =10% *=5%, **=1%, ***=0.1% with respect to

zero, 

=10%,  =5%,  =1%,  =0.1% with respect to one.

Multiplier lm1 lm2 qlm nlm lnlm

β0 1 –0.19777 1.24052*** 2.46489*** — 0.10064 β1 WtΔ– –0.00039* –0.00040* 0.00129*** — 0.00000 β2 WtΔ –0.00214*** –0.00209*** –0.00410*** — –0.00002 β3 StΔ– –0.00043 –0.00015 –0.00173* — 0.00000 β4 StΔ –0.00258*** –0.00273*** –0.00267*** — –0.00002 β5 WtA –0.00009*** 0.00005*** 0.00014*** — –0.00000 β6 StA 0.00000 –0.00002 –0.00010* — –0.00000 β7 PtDA — 0.97019*** 0.86481*** — 0.39731* β8 (WtΔ–)2 — — 0.00000*** — — β9 (WtΔ)2 — — 0.00000*** — — β10 (StΔ–)2 — — 0.00000*** — — β11 (StΔ)2 — — 0.00000 — — β12 (WtA)2 — — 0.00000*** — — β13 (StA)2 — — 0.00000* — — β14 (PtDA)2 — — 0.00125*** — — β15 1 — — — 0.00004 –0.00061 β16 WtΔ– — — — 0.33663*** 0.90624 β17 WtΔ — — — 0.39478*** 0.24175*** β18 StΔ– — — — 0.86325*** 1.48092*** β19 StΔ — — — 0.37144*** 0.25544*** β20 WtA — — — –0.02659*** –0.06149 β21 StA — — — –0.02590*** –0.01337 β22 Ptnlm — — — — 0.55152***

100 Economics of Energy & Environmental Policy

Copyright © 2021 by the IAEE. All rights reserved.

TABLE 3 Standard deviations of prices P lm2 and P nlm relative to the respective baseline models

with γ = 0 and ρ = 0. W abbreviates Wind, S abbreviates Solar.

Model lm2 Model nlm

γ ρ W S W+S 0.5(W+S) W S W+S 0.5(W+S) 0.1 0.000 1.000 1.000 1.000 1.000 1.001 1.000 1.001 1.000 0.1 0.500 1.000 1.000 1.000 1.000 1.001 1.000 1.001 1.001 0.1 0.800 1.000 1.000 1.001 1.000 1.002 1.001 1.002 1.001 1 0.000 1.016 1.004 1.020 1.005 1.026 1.006 1.029 1.009 1 0.500 1.036 1.012 1.048 1.012 1.062 1.025 1.081 1.023 1 0.800 1.051 1.018 1.069 1.018 1.089 1.034 1.123 1.033 5 0.000 1.799 1.330 2.000 1.323 3.797 4.455 11.245 1.850 5 0.500 1.931 1.397 2.161 1.385 4.245 5.031 13.273 2.193 5 0.800 2.008 1.437 2.254 1.421 4.775 5.424 14.100 2.310

0 0 4.980 4.974

TABLE 4 Relative differences between true and modeled intraday prices at the 0.1% quantile relative to the respective baseline models with γ = 0 and ρ = 0. W abbreviates Wind,

S abbreviates Solar.

Model lm2 Model nlm

γ ρ W S W+S 0.5(W+S) W S W+S 0.5(W+S) 0.1 0.000 1.005 1.000 1.005 1.002 1.047 1.000 1.047 1.023 0.1 0.500 1.000 0.997 0.997 0.999 1.042 0.992 1.042 1.009 0.1 0.800 0.997 0.996 0.993 0.996 1.036 0.986 1.036 1.009 1 0.000 1.009 0.981 0.990 0.995 1.124 0.986 1.124 1.095 1 0.500 0.977 0.963 0.940 0.970 1.059 0.986 1.068 1.094 1 0.800 0.960 0.951 0.915 0.957 1.060 0.986 1.127 1.040 5 0.000 1.076 1.043 1.232 0.965 1.851 2.166 3.158 2.105 5 0.500 1.101 1.103 1.340 0.939 1.956 2.195 3.705 2.106 5 0.800 1.129 1.136 1.401 0.945 2.438 2.200 4.009 2.107

0 0 –39.190 –29.605

TABLE 5 Relative differences between true and modeled intraday prices at the 99.9% quantile relative to the respective baseline models with γ = 0 and ρ = 0. W abbreviates Wind,

S abbreviates Solar.

Model lm2 Model nlm

γ ρ W S W+S 0.5(W+S) W S W+S 0.5(W+S) 0.1 0.000 0.993 1.000 0.993 0.997 0.962 0.973 0.961 0.963 0.1 0.500 0.991 1.000 0.991 0.996 0.991 0.963 0.991 0.990 0.1 0.800 0.990 1.000 0.990 0.995 1.022 0.962 1.022 0.991 1 0.000 0.966 1.000 0.966 0.979 1.060 1.046 1.061 1.025 1 0.500 0.962 1.000 0.962 0.977 1.335 1.095 1.605 1.105 1 0.800 0.967 1.000 0.967 0.976 1.468 1.095 1.851 1.109 5 0.000 1.680 1.435 1.911 1.178 7.273 19.345 19.572 3.349 5 0.500 1.870 1.558 2.133 1.188 13.708 19.587 19.723 3.648 5 0.800 1.963 1.607 2.265 1.242 19.242 19.739 19.947 3.985

0 0 25.504 26.135

The Impact of Renewable Energy Forecasts on Intraday Electricity Prices 101

Copyright © 2021 by the IAEE. All rights reserved.

References

Ahlstrom, M., Bartlett, D., Collier, C., Duchesne, J., Edelson, D., Gesino, A., Keyser, M., Maggio, D., Milligan, M., Möhrlen, C., et al. 2013. “Knowledge is power: Efficiently integrating wind energy and wind forecasts.” IEEE Power and Energy Magazine, 11(6):45–52. https://doi.org/10.1109/MPE.2013.2277999.

Andrade, J. R. and Bessa, R. J. 2017. “Improving renewable energy forecasting with a grid of numerical weather predictions.” IEEE Transactions on Sustainable Energy, 8(4):1571–1580. https://doi.org/10.1109/ TSTE.2017.2694340.

Antonanzas, J., Osorio, N., Escobar, R., Urraca, R., Martinez-de Pison, F. J., and Antonanzas-Torres, F. 2016. “Review of photovoltaic power forecasting.” Solar Energy, 136:78–111. https://doi.org/10.1016/j.solen- er.2016.06.069.

Association, E. W. E. 2009. Oceans of opportunity: harnessing Europe’s largest domestic energy resource. EWEA. Awerbuch, S. 2006. “Portfolio-based electricity generation planning: policy implications for renewables and en-

ergy security.” Mitigation and adaptation strategies for Global Change, 11(3):693–710. https://doi.org/10.1007/ s11027-006-4754-4.

BDEW 2017. Energie-info—leitfaden zur informationsbereitstellung bei einspeisemanagement-maßnahmen. Bundesverband der Energie-und Wasserwirtschaft e.V, Technical Report.

Berk, J. B. and DeMarzo, P. M. 2007. Corporate finance. Pearson Education. Bird, L., Lew, D., Milligan, M., Carlini, E. M., Estanqueiro, A., Flynn, D., Gomez-Lazaro, E., Holttinen, H., Me-

nemenlis, N., Orths, A., et al. 2016. “Wind and solar energy curtailment: A review of international experience.” Renewable and Sustainable Energy Reviews, 65:577–586. https://doi.org/10.1016/j.rser.2016.06.082.

Brown, T., Schierhorn, P.-P., Tröster, E., and Ackermann, T. 2016. “Optimising the european transmission system for 77% renewable electricity by 2030.” IET Renewable Power Generation, 10(1):3–9. https://doi.org/10.1049/ iet-rpg.2015.0135.

Bundesnetzagentur. 2018. Leitfaden zum einspeisemanagement, version 3.0. Techical report. Carmona, R., Coulon, M., and Schwarz, D. 2013. “Electricity price modeling and asset valuation: a multi-fuel

structural approach.” Mathematics and Financial Economics, 7(2):167–202. https://doi.org/10.1007/s11579- 012-0091-4.

Chang, W.-Y. et al. 2014. “A literature review of wind forecasting methods.” Journal of Power and Energy Engineer- ing, 2(04):161. https://doi.org/10.4236/jpee.2014.24023.

Child, M., Kemfert, C., Bogdanov, D., and Breyer, C. 2019. “Flexible electricity generation, grid exchange and storage for the transition to a 100% renewable energy system in Europe.” Renewable energy, 139:80–101. https:// doi.org/10.1016/j.renene.2019.02.077.

Clò, S., Cataldi, A., and Zoppoli, P. 2015. “The merit-order effect in the italian power market: The impact of solar and wind generation on national wholesale electricity prices.” Energy Policy, 77:79–88. https://doi.org/10.1016/j. enpol.2014.11.038.

Cludius, J., Hermann, H., Matthes, F. C., and Graichen, V. 2014. “The merit order effect of wind and photovoltaic electricity generation in germany 2008–2016: Estimation and distributional implications.” Energy Economics, 44:302–313. https://doi.org/10.1016/j.eneco.2014.04.020.

Cochrane, J. H. 2009. Asset pricing: Revised edition. Princeton University Press. Coulon, M., Jacobsson, C., and Ströjby, J. 2014. Hourly resolution forward curves for power: Statistical modeling

meets market fundamentals. https://doi.org/10.1007/978-1-137-37027-3_6. de Frutos, M. Á. and Manzano, C. 2014. “Market transparency, market quality, and sunshine trading.” Journal of

Financial Markets, 17:174–198. https://doi.org/10.1016/j.finmar.2013.06.001. De Giorgi, M. G., Ficarella, A., and Tarantino, M. 2011. “Assessment of the benefits of numerical weather pre-

dictions in wind power forecasting based on statistical methods.” Energy, 36(7):3968–3978. https://doi. org/10.1016/j.energy.2011.05.006.

Diebold, F. X. 2015. “Comparing predictive accuracy, twenty years later: A personal perspective on the use and abuse of diebold-mariano tests.” Journal of Business & Economic Statistics, 33(1):1–8. https://doi.org/10.1080/0 7350015.2014.983236.

Dillig, M., Jung, M., and Karl, J. 2016. “The impact of renewables on electricity prices in germany-an estimation based on historic spot prices in the years 2011–2013.” Renewable and Sustainable Energy Reviews, 57:7–15. https://doi.org/10.1016/j.rser.2015.12.003.

Drake, B. and Hubacek, K. 2007. “What to expect from a greater geographic dispersion of wind farms?-a risk port- folio approach.” Energy Policy, 35(8):3999–4008. https://doi.org/10.1016/j.enpol.2007.01.026.

102 Economics of Energy & Environmental Policy

Copyright © 2021 by the IAEE. All rights reserved.

Eising, M., Hobbie, H., and Möst, D. 2020. “Future wind and solar power market values in germany-evidence of spatial and technological dependencies?” Energy Economics, page 104638. https://doi.org/10.1016/j.ene- co.2019.104638.

Engeland, K., Borga, M., Creutin, J.-D., François, B., Ramos, M.-H., and Vidal, J.-P. 2017. “Space-time variability of climate variables and intermittent renewable electricity production-a review.” Renewable and Sustainable Ener- gy Reviews, 79:600–617. https://doi.org/10.1016/j.rser.2017.05.046.

ENTSO-E. 2019. Wind and solar power actual and forecasted values. Available via: https://transparency.entsoe. eu/, accessed on 30.06.2019.

EPEX. 2019a. Day ahead auction curves data. Available via: https://www.epexspot.com/de/marktdaten/daya- headauktion/preiskurve/auction-aggregated-curve/2019-07-30/DE/00, accessed on 30.06.2019.

EPEX. 2019b. Day ahead prices and volumes data. Available via: https://www.epexspot.com/de/marktdaten/daya- headauktion, accessed on 30.06.2019.

Eyer, J. and Corey, G. 2010. “Energy storage for the electricity grid: Benefits and market potential assessment guide.” Sandia National Laboratories, 20(10):5. https://doi.org/10.2172/1031895.

Fürsch, M., Malischek, R., and Lindenberger, D. 2012. Der Merit-Order-Effekt der Erneuerbaren Energien-Anal- yse der kurzen und langen Frist. Technical report, EWI working paper.

Garnier, E. and Madlener, R. 2015. “Balancing forecast errors in continuous-trade intraday markets.” Energy Sys- tems, 6(3):361–388. https://doi.org/10.1007/s12667-015-0143-y.

Goodarzi, S., Perera, H. N., and Bunn, D. 2019. “The impact of renewable energy forecast errors on imbalance volumes and electricity spot prices.” Energy Policy, 134:110827. https://doi.org/10.1016/j.enpol.2019.06.035.

Grams, C. M., Beerli, R., Pfenninger, S., Staffell, I., and Wernli, H. 2017. “Balancing europe’s wind-power out- put through spatial deployment informed by weather regimes.” Nature climate change, 7(8):557. https://doi. org/10.1038/nclimate3338.

Gürtler, M. and Paulsen, T. 2018. “The effect of wind and solar power forecasts on day-ahead and intraday electric- ity prices in Germany.” Energy Economics, 75:150–162. https://doi.org/10.1016/j.eneco.2018.07.006.

Hagemann, S. 2013. Price determinants in the German intraday market for electricity: an empirical analysis. https://doi.org/10.2139/ssrn.2352854.

Handschy, M. A., Rose, S., and Apt, J. 2017. “Is it always windy somewhere? Occurrence of low-wind-power events over large areas.” Renewable Energy, 101:1124–1130. https://doi.org/10.1016/j.renene.2016.10.004.

Hasche, B. 2010. “General statistics of geographically dispersed wind power.” Wind Energy, 13(8):773–784. https:// doi.org/10.1002/we.397.

Howison, S. and Coulon, M. 2009. “Stochastic behaviour of the electricity bid stack: from fundamental drivers to power prices.” The Journal of Energy Markets, 2:29–69. https://doi.org/10.21314/JEM.2009.032.

Jacobsen, H. K. and Schröder, S. T. 2012. “Curtailment of renewable generation: Economic optimality and incen- tives.” Energy Policy, 49:663–675. https://doi.org/10.1016/j.enpol.2012.07.004.

Jacottet, A. 2012. Cross-border electricity interconnections for a well functioning eu internal electricity market. Jónsson, T., Pinson, P., and Madsen, H. 2010. “On the market impact of wind energy forecasts.” Energy Economics,

32(2):313–320. https://doi.org/10.1016/j.eneco.2009.10.018. Joos, M. and Staffell, I. 2018. “Short-term integration costs of variable renewable energy: Wind curtailment

and balancing in Britain and Germany.” Renewable and Sustainable Energy Reviews, 86:45–65. https://doi. org/10.1016/j.rser.2018.01.009.

Kath, C. 2019. “Modeling intraday markets under the new advances of the cross-border intraday project (xbid): Evidence from the German intraday market.” Energies, 12(22):4339. https://doi.org/10.3390/en12224339.

Ketterer, J. C. 2014. “The impact of wind power generation on the electricity price in Germany.” Energy Economics, 44:270–280. https://doi.org/10.1016/j.eneco.2014.04.003.

Kiesel, R. and Paraschiv, F. 2017. “Econometric analysis of 15-minute intraday electricity prices.” Energy Economics, 64:77–90. https://doi.org/10.1016/j.eneco.2017.03.002.

Knaut, A. and Paulus, S. 2016. “Hourly price elasticity pattern of electricity demand in the German day-ahead market.” Technical report, EWI Working Paper.

Kühnert, J. 2016. Development of a photovoltaic power prediction system for forecast horizons of several hours. PhD thesis, Universität Oldenburg.

Kulakov, S. and Ziel, F. 2019. Determining the fundamental supply and demand curves in a wholesale electricity market. arXiv preprint arXiv:1903.11383.

Kyle, A. S. 1985. “Continuous auctions and insider trading.” Econometrica: Journal of the Econometric Society, 1315–1335. https://doi.org/10.2307/1913210.

The Impact of Renewable Energy Forecasts on Intraday Electricity Prices 103

Copyright © 2021 by the IAEE. All rights reserved.

Larson, D. P., Nonnenmacher, L., and Coimbra, C. F. 2016. “Day-ahead forecasting of solar power output from photovoltaic plants in the american southwest.” Renewable Energy, 91:11–20. https://doi.org/10.1016/j. renene.2016.01.039.

Lund, P. D., Lindgren, J., Mikkola, J., and Salpakari, J. 2015. “Review of energy system flexibility measures to en- able high levels of variable renewable electricity.” Renewable and Sustainable Energy Reviews, 45:785–807. https:// doi.org/10.1016/j.rser.2015.01.057.

Mills, A. 2010. Implications of wide-area geographic diversity for short-term variability of solar power. https://doi. org/10.2172/986925.

Monforti, F., Huld, T., Bódis, K., Vitali, L., D’isidoro, M., and Lacal-Arántegui, R. 2014. “Assessing complemen- tarity of wind and solar resources for energy production in Italy. A monte carlo approach.” Renewable Energy, 63:576–586. https://doi.org/10.1016/j.renene.2013.10.028.

Narajewski, M. and Ziel, F. 2019. “Econometric modelling and forecasting of intraday electricity prices.” Journal of Commodity Markets, page 100107. https://doi.org/10.1016/j.jcomm.2019.100107.

Neubarth, J., Woll, O., Weber, C., and Gerecht, M. 2006. Impact of wind power generation on spot market prices; beeinflussung der spotmarktpreise durch windstromerzeugung.

Novacheck, J. and Johnson, J. X. 2017. “Diversifying wind power in real power systems.” Renewable Energy, 106:177–185. https://doi.org/10.1016/j.renene.2016.12.100.

Ostermann, A., Köppl, S., and Estermann, T. 2019. Analysen zum Einspeisemanagement: Regionalisierter Flexibil- itätsbedarf und Auswirkung auf den Strommarkt. Technical Report.

Palensky, P. and Dietrich, D. 2011. “Demand side management: Demand response, intelligent energy sys- tems, and smart loads.” IEEE transactions on industrial informatics, 7(3):381–388. https://doi.org/10.1109/ TII.2011.2158841.

Palmintier, B., Hansen, L., and Levine, J. 2008. “Spatial and temporal interactions of solar and wind resources in the next generation utility.” In SOLAR 2008 Conference and Exhibition. San Diego, CA, 3–8.

Pape, C. 2017. The impact of intraday markets on the market value of flexibility-decomposing effects on profile and the imbalance costs. https://doi.org/10.2139/ssrn.3090889.

Pape, C. 2018. “The impact of intraday markets on the market value of flexibility-decomposing effects on profile and the imbalance costs.” Energy Economics, 76:186–201. https://doi.org/10.1016/j.eneco.2018.10.004.

Pape, C., Hagemann, S., and Weber, C. 2016. “Are fundamentals enough? explaining price variations in the Ger- man day-ahead and intraday power market.” Energy Economics, 54:376–387. https://doi.org/10.1016/j.ene- co.2015.12.013.

Paternò, G., Madonia, A., Ippolito, M. G., Massaro, F., Favuzza, S., and Cassaro, C. 2016. “Analysis of the new submarine interconnection system between Italy and Malta: simulation of transmission network operation.” In 2016 IEEE 16th International Conference on Environment and Electrical Engineering (EEEIC), pages 1–6. IEEE. https://doi.org/10.1109/EEEIC.2016.7555467.

Perez, R., David, M., Hoff, T. E., Jamaly, M., Kivalov, S., Kleissl, J., Lauret, P., Perez, M., et al. 2016. “Spatial and temporal variability of solar energy.” Foundations and Trends in Renewable Energy, 1(1):1–44. https://doi. org/10.1561/2700000006.

Perez, R., Kivalov, S., Schlemmer, J., Hemker Jr, K., and Hoff, T. E. 2012. “Short-term irradiance variability: Pre- liminary estimation of station pair correlation as a function of distance.” Solar Energy, 86(8):2170–2176. https:// doi.org/10.1016/j.solener.2012.02.027.

Pinson, P. 2016. “Introducing distributed learning approaches in wind power forecasting.” In 2016 International Conference on Probabilistic Methods Applied to Power Systems (PMAPS), pages 1–6. IEEE. https://doi.org/10.1109/ PMAPS.2016.7764224.

Puka, L. and Szulecki, K. 2014. “The politics and economics of cross-border electricity infrastructure: A framework for analysis.” Energy Research & Social Science, 4:124–134. https://doi.org/10.1016/j.erss.2014.10.003.

Roldan-Fernandez, J.-M., Burgos-Payan, M., Riquelme-Santos, J.-M., and Trigo-Garcia, A.-L. 2016. “The merit-or- der effect of energy efficiency.” Energy Procedia, 106:175–184. https://doi.org/10.1016/j.egypro.2016.12.114.

Schlachtberger, D. P., Brown, T., Schramm, S., and Greiner, M. 2017. “The benefits of cooperation in a highly renewable European electricity network.” Energy, 134:469–481. https://doi.org/10.1016/j.energy.2017.06.004.

Scholz, Y. 2012. Renewable energy based electricity supply at low costs: development of the remix model and ap- plication for Europe. PhD Thesis.

Shah, I. and Lisi, F. 2018. Forecasting of electricity price through a functional prediction of supply and demand curves. Technical report, Working paper.

104 Economics of Energy & Environmental Policy

Copyright © 2021 by the IAEE. All rights reserved.

Sioshansi, R. and Denholm, P. 2013. “Benefits of colocating concentrating solar power and wind.” IEEE Transac- tions on Sustainable Energy, 4(4):877–885. https://doi.org/10.1109/TSTE.2013.2253619.

Soysal, E. R., Olsen, O. J., Skytte, K., and Sekamane, J. K. 2017. “Intraday market asymmetries-a nordic example.” In European Energy Market (EEM), 2017 14th International Conference on the, pages 1–6. IEEE. https://doi. org/10.1109/EEM.2017.7981920.

TGE. 2019. Single intraday coupling (sidc). https://tge.pl/pub/TGE/files/MC/SIDC_XBID_2nd_Wave_Pre_ Launch.2019.pdf. Accessed on 01 March. 2020.

Uniejewski, B., Marcjasz, G., and Weron, R. 2019. “Understanding intraday electricity markets: Variable selection and very short-term price forecasting using lasso.” International Journal of Forecasting. https://doi.org/10.1016/j. ijforecast.2019.02.001.

von Luckner, N. G., Cartea, Á., Jaimungal, S., and Kiesel, R. 2017. Optimal market maker pricing in the German intraday power market.

Von Roon, S. and Wagner, U. 2009. “The interaction of conventional power production and renewable power under the aspect of balancing forecast errors.” In Proceedings of the 10th IAEE European Conference on Energy, Policies and Technologies for Sustainable Economies.

Voyant, C., Notton, G., Kalogirou, S., Nivet, M.-L., Paoli, C., Motte, F., and Fouilloy, A. 2017. “Machine learning methods for solar radiation forecasting: A review.” Renewable Energy, 105:569–582. https://doi.org/10.1016/j. renene.2016.12.095.

Weber, C. 2010. “Adequate intraday market design to enable the integration of wind energy into the european power systems.” Energy Policy, 38(7):3155–3163. https://doi.org/10.1016/j.enpol.2009.07.040.

Widén, J. 2011. “Correlations between large-scale solar and wind power in a future scenario for sweden.” IEEE transactions on sustainable energy, 2(2):177–184. https://doi.org/10.1109/TSTE.2010.2101620.

Woo, C.-K., Horowitz, I., Moore, J., and Pacheco, A. 2011. “The impact of wind generation on the electrici- ty spot-market price level and variance: The Texas experience.” Energy Policy, 39(7):3939–3944. https://doi. org/10.1016/j.enpol.2011.03.084.

Zareipour, H., Canizares, C. A., and Bhattacharya, K. 2009. “Economic impact of electricity market price fore- casting errors: A demand-side analysis.” IEEE Transactions on Power Systems, 25(1):254–262. https://doi. org/10.1109/TPWRS.2009.2030380.

Zheng, J., Yanni, K., Mengshi, L., and Qinghua, W. 2019. “Stochastic optimization of cost-risk for integrated energy system considering wind and solar power correlated.” Journal of Modern Power Systems and Clean Energy, 7(6):1472–1483. https://doi.org/10.1007/s40565-019-0519-4.

Ziel, F. 2017. “Modeling the impact of wind and solar power forecasting errors on intraday electricity prices.” In European Energy Market (EEM), 2017 14th International Conference on the, pages 1–5. IEEE. https://doi. org/10.1109/EEM.2017.7981900.

Ziel, F. and Steinert, R. 2016. “Electricity price forecasting using sale and purchase curves: The x-model.” Energy Economics, 59:435–454. https://doi.org/10.1016/j.eneco.2016.08.008.

Ziel, F. and Weron, R. 2018. “Day-ahead electricity price forecasting with high-dimensional structures: Univar- iate vs. multivariate modeling frameworks.” Energy Economics, 70:396–420. https://doi.org/10.1016/j.ene- co.2017.12.016.

Ziel, F., Weron, R., et al. 2016. Day-ahead electricity price forecasting with high-dimensional structures: Univariate vs. multivariate models. Technical report, Hugo Steinhaus Center, Wroclaw University of Technology.

Zipf, M. and Möst, D. 2013. “Impacts of volatile and uncertain renewable energy sources on the German electricity system.” In 2013 10th International Conference on the European Energy Market (EEM), pages 1–8. IEEE. https:// doi.org/10.1109/EEM.2013.6607397.

Natural Gas 1.PDF

Natural Gas as a Key Alternative Energy Source in Sustainable Renewable Energy Transition: A Mini Review Norsyahida Mohammad*, Waznatol Widad Mohamad Ishak, Siti Indati Mustapa and Bamidele Victor Ayodele

Institute of Energy Policy and Research, Universiti Tenaga Nasional, Kajang, Malaysia

Energy security and sustainability are undeniably the main concerns in combatting climate change. While an immediate call for all-green and renewable energy seems to be impossible due to huge financial implications and inadequate supporting energy structure, an alternative to fossil fuels needs to be established. Natural gas, a naturally occurring fossil gas is a cleaner energy source option compared to other fossil fuels such as coal, bitumen, and diesel. Natural gas makes the best fit for a sustainable renewable energy transition in any country around the globe due to its competitiveness towards other fossil fuels such as coal and its ability to aid the integration of renewables. This review highlights the technological pathways of utilizing natural gas in a transition to sustainable renewable energy systems, with a focus on the natural gas components and resources point of view for ASEAN member states (AMS). Policies that support the development of natural gas as a key alternative energy source in sustainable renewable energy transition would also be reviewed. This review aims to provide a thorough guide to researchers, stakeholders, and policymakers to construct and support efficient, reliable, affordable, sustainable, and environmentally friendly energy systems utilizing the abundant inexpensive natural gas.

Keywords: ASEAN, natural gas, energy transition, sustainable energy, alternative energy

INTRODUCTION

A sustainable energy transition is defined as a shift toward a high-efficiency energy system that is well-managed to balance environmental and social costs, risks, and benefits such that the shift is deemed to be sustainable (Chen et al., 2019). The transition also includes the shift from utilizing fossil fuel to exploiting renewable energy sources in energy generation and the evolution from centralized to decentralized energy systems (Guidolin and Alpcan, 2019). In Southeast Asia, coal-fired power plants are increasing due to the lower price of coal in the region, compared to oil and gas (Steckel et al., 2015). The use of coal for power generation contributes toward substantial greenhouse gas (GHG) emissions and deaths related to air pollution which include fine particulate matter emissions and toxic air contaminants (Kittner et al., 2018). However, immediate implementation of clean and renewable energy seems to be far-fetched due to the high capital costs and inadequate infrastructure in most ASEAN member states (AMS). A cleaner and more efficient fossil fuel such as natural gas

Edited by: Kalpit V. Shah,

RMIT University, Australia

Reviewed by: Fehmi Akgun,

TUBITAK Marmara Research Centre Energy Institute, Turkey S. M. Ashrafur Rahman,

Queensland University of Technology, Australia

*Correspondence: Norsyahida Mohammad

[email protected]

Specialty section: This article was submitted to

Advanced Clean Fuel Technologies, a section of the journal

Frontiers in Energy Research

Received: 02 November 2020 Accepted: 06 May 2021 Published: 24 May 2021

Citation: Mohammad N, Mohamad Ishak WW, Mustapa SI and Ayodele BV (2021)

Natural Gas as a Key Alternative Energy Source in Sustainable

Renewable Energy Transition: A Mini Review.

Front. Energy Res. 9:625023. doi: 10.3389/fenrg.2021.625023

Frontiers in Energy Research | www.frontiersin.org May 2021 | Volume 9 | Article 6250231

MINI REVIEW published: 24 May 2021

doi: 10.3389/fenrg.2021.625023

could be utilized as an alternative energy source in bridging toward a sustainable renewable energy system in the ASEAN.

Natural gas consists of mainly methane gas, which emits the least amount of carbon dioxide (CO2) and other harmful substances such as nitrogen oxides, sulfur dioxide, and particulates when combusted, compared to oil and coal. Direct replacement of coal with natural gas for power generation has proven to reduce GHG emissions tremendously (Mac Kinnon et al., 2018). The integration of natural gas and renewable energy resources in power generation ensures energy security and sustainability, while contributing toward a significant reduction of GHG toward a sustainable energy system, in line with the Paris Agreement adopted in 2015 (Denchak, 2018). The multi-lateral agreement aims to limit the global average temperature rise to below 2°C, by reducing GHG emissions.

NATURAL GAS RESERVES, PRODUCTION, AND DEMAND IN ASEAN

As the ASEAN region is bestowed with an abundance of natural gas deposits, which put the region as a net natural gas exporter, it makes sense that natural gas is the key energy source for the sustainable renewable energy transition in the region. Malaysia, Indonesia, and Brunei are AMS that has been known as liquefied natural gas (LNG) exporter for the last four decades. Malaysia poses 42 trillion cubic feet (tcf) of proved natural gas reserves in 2016, which is the fifth-largest natural gas reserve holder in the Asia-Pacific region behind China, Indonesia, Australia, and India. According to Energy Information Administration (EIA), Malaysia is the third-largest exporter of LNG in the world after Qatar and Australia in 2016, and the second-largest oil and natural gas producer in Southeast Asia, behind Indonesia

(EIA, 2017). Indonesia, the most populous country in Southeast Asia, is the fifth-largest exporter of LNG in the world. Indonesia possessed 102 trillion cubic feet (tcf) of proved natural gas reserves in 2016. The country’s proved natural gas reserves are the second largest in the Asia-Pacific region, after China. Brunei, the smallest country in ASEAN, produced 410 billion cubic feet of dry natural gas in 2016 and has been a long-term LNG exporter to Japan and Korea. The comparison of the reserves, production, and consumption of natural gas in ASEAN countries in 2016, based on data from EIA is depicted in Figure 1.

According to the ASEAN Centre for Energy (ACE), natural gas makes up to 24% of the ASEAN energy mix in 2016 (Silitonga and Anugrah, 2015). Natural gas is mainly utilized for power generation with more than one-third of the total installed power generation capacity in AMS (Silitonga and Anugrah, 2015). According to a report by IRENA in 2018, natural gas contributed the largest share (41%) to the power generation mix for ASEAN in 2015, followed by coal (33%) and hydropower (16%) (IRENA, 2018). Three AMS that are Brunei, Singapore, and Thailand generate more than 70% of electricity from natural gas. However, under a business-as- usual scenario, coal is expected to replace natural gas as the dominant source of power generation in the region by 2040 (IRENA, 2018).

Apart from utilizing natural gas as an export commodity and power generation, a smaller portion of natural gas is also utilized domestically for transportation and feedstock in industrial processes producing methanol, fertilizers, and pharmaceutical products. Compressed natural gas (CNG) is used in natural gas vehicles (NGV) due to its lower price compared to petrol or diesel, and growing concern for the environment (Yusaf et al., 2001). Although most developed countries such as the United States, Canada, Europe, Australia, and New Zealand

FIGURE 1 | The reserves, production, and consumption of natural gas in ASEAN countries in 2016.

Frontiers in Energy Research | www.frontiersin.org May 2021 | Volume 9 | Article 6250232

Mohammad et al. Natural Gas Key Alternative Energy

have been encouraging the use of NGV, in AMS however, the development and use of NGVs is relatively slow.

AN OVERVIEW OF SYNTHETIC NATURAL GAS PRODUCTION METHODS

There are three main pathways to produce synthetic natural gas (SNG) that are the biochemical, thermochemical, and electrochemical pathways (Fendt et al., 2016). The most common method of producing SNG would be the thermochemical method, where conventional non-renewable source, coal undergo the multi-step process to produce natural gas starting with gasification to produce mainly CO and H2, also known as the producer gas. The gas produced would undergo a scrubbing step to remove unwanted waste products, producing a pure mixture of H2, CO, and CO2 termed as the synthesis gas. The synthesis gas then undergoes a water-gas shift reaction, followed by the methanation process to produce SNG. The methanation process requires catalysts from a range of transition metals dispersed on metal oxide supports such as Ni/Al2O3 (Li et al., 2013), Ni/TiO2 (Shinde and Madras, 2014), Ni/α-Al2O3 (Gao et al., 2013), and NiO-CeO2 (Atzori et al., 2018). A full review on methanation catalysts advancements for SNG production could be found in (Gao et al., 2015). The most common reactor type used in the methanation process is the fixed bed reactor due to its simple and effective design, although several other designs such as the fluidized bed reactor, honeycomb reactor, and microchannel reactor were also considered due to better catalyst exposure (Bolt et al., 2020).

Renewable sources such as dry biomass or solid waste obtained from municipal and agricultural wastes could be converted to SNG using the thermochemical pathway like coal. In the biochemical pathway, wet biomass undergoes fermentation to produce biogas, which is high in methane and CO2. After CO2 separation, the resulting gas could be fed directly into the natural gas grid. Another pathway for SNG production is the electrochemical pathway where the electricity generated from renewables was used to produce H2 via electrolysis, followed by methanation to produce SNG. Comprehensive reviews on SNG production methods could be found elsewhere (Kopyscinski et al., 2010; Fendt et al., 2016; Bolt et al., 2020).

ROLE OF NATURAL GAS IN THE SUSTAINABLE RENEWABLE ENERGY TRANSITION

Natural gas plays a major role in the short- to mid-term transition toward sustainable energy systems. Besides being a cleaner and more efficient fossil fuel than coal, natural gas is highly flexible such that natural gas peaking combustion turbines has a dynamic ramping ability that can increase or decrease electricity generation within less than an hour, allowing it to respond rapidly to fluctuations on the demand side and adjust to fluctuating power produced from inconsistencies of renewable energy resources such as solar and wind (Huang et al., 2019). In

Thailand, heavy reliance on natural gas for electricity generation enables the country to integrate utility-scale solar onto its flexible power grid. Thailand owns the highest share of utility-scale solar compared with other AMS (Tongsopit et al., 2015). Similarly, heavy reliance on natural gas for electricity generation in Brunei and Singapore enables future investments in offshore wind power generation. This boosts the potential of Brunei and Singapore in becoming flexible trading hubs for electricity and gas infrastructure (Huang et al., 2019). Besides, the existing natural gas supporting infrastructure which includes storage, transportation, and distribution can facilitate the integration of gaseous types of renewable fuels such as biogases. This enables the transition to fully utilize renewable energy for power generation in the future.

AN OVERVIEW OF NATURAL GAS UTILIZATION FOR POWER GENERATION

One of the main environmental concerns of natural gas power generation is that it produces significant amounts of CO2. However, the CO2 emission could be reduced by substituting conventional natural gas combustion turbines with state-of-the- art advanced natural gas power generation technologies such as natural gas combined cycle plants, fuel cells, micro-turbines, and hybrid fuel cell/heat engine plants (Mac Kinnon et al., 2018). The advanced natural gas conversion technologies, in combination with low carbon and renewable energy in power generation, as well as the technology for carbon capture and storage (CCS) will be able to further reduce CO2 emission. The main principles of the CCS technologies to reduce CO2 emissions include pre- combustion capture, post-combustion capture, oxyfuel O2/CO2 recycle combustion capture, and chemical looping. However, the implementation of CCS technologies is limited due to unclear social and political acceptability (Bui et al., 2018).

Other technologies to capture CO2 from natural gas before input to power generation include chemical looping and high- temperature membranes. Besides, natural gas electricity generation could be integrated with renewable energy sources such as solar and wind, to further reduce CO2 emissions. Hydrogen could be produced by natural gas steam reforming and used to produce electricity. A system integrating hydrogen produced from natural gas combined with CCS has been proposed (Khan et al., 2020). The pathways for sustainable natural gas utilization in electricity generation are presented in Figure 2.

LIFE CYCLE ANALYSIS ON NATURAL GAS USE FOR A SUSTAINABLE RENEWABLE ENERGY TRANSITION There are several life cycle analyses (LCAs) conducted on electricity generation from natural gas and its renewable energy alternatives including potential renewables for ASEAN which are solar, biomass, and wind. Although most LCA conducted used GHG emission as the main indicator for the

Frontiers in Energy Research | www.frontiersin.org May 2021 | Volume 9 | Article 6250233

Mohammad et al. Natural Gas Key Alternative Energy

environmental performance of a system, there are inevitably more factors that could be compared such as energy pay-back time, lifetime, and land use. Three LCA phases that need to be considered are fuel provision, plant operation, and infrastructure, although intricate details such as energy recovery efficiency, electricity mix utilized during construction, and quality of feedstocks contribute toward the variability of LCAs conducted (Turconi et al., 2013). It was found that the highest percentage of emissions came from plant operation for fossil fuel- based electricity generation such as power generation from natural gas, whereas in biomass-based power systems fuel provision produced the highest percentage of emissions (Turconi et al., 2013). GHG emissions in biomass-only electricity generation are highest when using wastes as feedstock, compared to using agriculture residues, dedicated crops, forestry, and industrial wastes (Kadiyala, 2016). It was also found that emissions from infrastructures remained the largest percentage in other renewables-based power generation systems such as systems driven by hydropower, solar, and wind (Turconi et al., 2013).

The GHG emission of a natural gas power plant ranges from 380 to 1,000 CO2-eq, compared to its renewable alternatives such as solar (13–190 CO2-eq), biomass (8.5–130 CO2-eq), and wind (3–41 CO2-eq), based on the electricity outputs (Turconi et al., 2013). The LCA serves as a guideline on the cost-competitiveness and emissions of electricity generation using natural gas and its alternatives. However, the LCA conducted for the various electricity generation technologies provides differing

judgements on environmental impacts due to different bases and indicators used in the assessments.

NATURAL GAS INFRASTRUCTURE AND RENEWABLE ENERGY POLICIES IN ASEAN

The Trans-ASEAN Gas Pipeline (TAGP) project and the ASEAN Power Grid (APG) have been established to enhance regional cooperation between the AMS and other countries outside the region, in meeting natural gas demands and ensuring energy security and sustainability within the region. The TAGP project aims to enhance connectivity between AMS to ensure energy security, flexibility, and accessibility via pipelines and regasification terminals producing approximately 17.8 million tonnes per annum (MTPA) natural gas (ACE, 2015). The TAGP project does not only provide natural gas for electricity generation but also provides natural gas for feedstock in heating and cooling systems for industries and municipal users. The APG enables the incorporation of a larger and more diverse pool of renewable energy resources from AMS, reducing dependence on the region’s natural gas reserves. The electricity output from the multiple-resources integrated power system is also smoother as the individual renewable energy generating plants are accumulated over larger geographic areas (IEA, 2019). The establishment of the transnational supergrid would enhance economic growth and development within the region, meeting the growing electricity demand, and producing a resilient electricity infrastructure within the region.

FIGURE 2 | Pathways for sustainable natural gas utilization for electricity generation.

Frontiers in Energy Research | www.frontiersin.org May 2021 | Volume 9 | Article 6250234

Mohammad et al. Natural Gas Key Alternative Energy

The price-competitiveness of natural gas produced by AMS is indeed a big determinant for ensuring sustainable utilization of natural gas in this region. Increasing reliance on imports makes natural gas less price-competitive (IEA, 2019). Economic challenges in sustainable utilization of natural gas include capital intensity, financing, and returns on investment of the natural gas pipelines and LNG projects within the region (Sovacool, 2009). It is important to seek cost-optimal pathways for sustainable natural gas utilization, which includes electricity generation, transmission, and storage to maintain robust economic growth within the region.

The existing policy frameworks need to be expanded to support natural gas as a mediator for the integration of renewable energy into the grid. Most AMS have put forward policies and targets of integrating renewable energy into their energy mix. Brunei, an oil-rich AMS has put forward plans to generate 10% of power from renewables by 2035, while Indonesia has set a target of 31% of renewable energy in its energy mix by 2050 (IRENA, 2018). Moreover, energy efficiency targets, renewable energy policies, and institutional frameworks on renewable energy resources have been put in place to enhance the integration of renewable energy in the energy mix of the AMS. A detailed discussion on energy and renewable energy policies in ASEAN could be found elsewhere (Lidula et al., 2007). Various instruments on renewable energy such as feed-in-tariff, capital subsidies, permits, tax incentives, and renewable portfolio standards have been introduced in AMS to achieve its targeted 23% renewable energy share at the regional level by 2025 (Erdiwansyah, 2019). However, the implementation of the existing policies is yet to achieve a satisfactory level of effectiveness due to barriers such as high capital of renewable energy installation, inadequate investments, and lack of technology and knowledge transfer (Chang, 2015).

CONCLUSION AND FUTURE RECOMMENDATIONS

The abundance of natural gas in the ASEAN region indicates the suitability of natural gas as the key energy source for the transition toward short- and medium-term sustainable

renewable energy transition. Besides, natural gas could be produced via various methods utilizing renewable energy resources. The flexibility of power systems utilizing natural gas enables the integration of intermittent renewable energy such as solar and wind. The CO2 emissions resulted from natural gas power generation could be reduced via advanced natural gas conversion technologies, integration with renewables, and CCS technologies. The establishment of an ASEAN-integrated natural gas pipeline and power system can ensure sustainable utilization of natural gas. A comprehensive techno-economic analysis on advanced natural gas power generation integrated with CCS and renewables is crucial to assist transition toward sustainable renewable energy transition. Natural gas as a transition fuel in attaining sustainable renewable energy transition should be further evaluated based on four pillars namely availability, applicability, acceptability, and affordability. Renewable energy targets, natural gas policies, and related frameworks should be enhanced to support the utilization and development of natural gas as the alternative energy source in transitioning toward a sustainable energy system in AMS.

AUTHOR CONTRIBUTIONS

Conceptualization, SM and BA; literature review and resources, NM and WM, writing, NM, review and editing, BA, funding acquisition, SM. All authors have read and agree to the published version of the manuscript.

FUNDING

This research was supported by the Innovation and Research Management Centre (iRMC), Universiti Tenaga Nasional.

ACKNOWLEDGMENTS

The authors acknowledge the financial support of iRMC, Universiti Tenaga Nasional.

REFERENCES

ACE (2015). “ASEAN Member States. Indonesia,” in Asean Plan of Action for Energy Cooperation (APAEC) 2016-2025 Phase I: 2016-2020. Editor Christopher G. Zamora (Jakarta, Indonesia: ASEAN Centre for Energy).

Atzori, L., Cutrufello, M. G., Meloni, D., Cannas, C., Gazzoli, D., Monaci, R., et al. (2018). Highly Active NiO-CeO2 Catalysts for Synthetic Natural Gas Production by CO2 Methanation. Catal. Today 299, 183–192. doi:10.1016/j.cattod.2017.05.065

Bolt, A., Dincer, I., and Agelin-Chaab, M. (2020). A Critical Review of Synthetic Natural Gas Production Techniques and Technologies. J. Nat. Gas Sci. Eng. 84, 103670. doi:10.1016/j.jngse.2020.103670

Bui, M., Adjiman, C. S., Bardow, A., Anthony, E. J., Boston, A., Brown, S., et al. (2018). Andy Boston, Solomon Brown, Paul S Fennell, Sabine Fuss, Amparo Galindo, and Leigh A Hackett.Carbon Capture and Storage (CCS): the Way Forward. Energy Environ. Sci. 11 (5), 1062–1176. doi:10.1039/c7ee02342a

Chang, Y., and Li, L. (2015). Renewable energy and policy options in an integrated ASEAN electricity market: Quantitative assessments and policy implications. Energy Policy 85, 39–49. doi:10.1029/2008GM000822

Chen, B., Xiong, R., Li, H., Sun, Q., and Yang, J. (2019). Pathways for Sustainable Energy Transition. J. Clean. Prod. 228, 1564–1571. doi:10.1016/j.jclepro.2019.04.372

Denchak, M. (2018). Paris Climate Agreement: Everything You Need to Know. (New York: Natural Resources Defense Council). https://www.nrdc.org/stories/ paris-climate-agreement-everything-you-need-know.

EIA (2017). Country Analysis Brief: Malaysia. U.S. Energy Information Administration. https://www.eia.gov/international/analysis/country/MYS.

Erdiwansyah, M., Mamat, R., Sani, M. S. M., Khoerunnisa, F., and Kadarohman, A. (2019). Target and demand for renewable energy across 10 ASEAN countries by 2040. The Electricity Journal 32 (10), 106670. doi:10.1016/j.tej.2019.106670

Fendt, S., Buttler, A., Gaderer, M., and Spliethoff, H. (2016). Comparison of Synthetic Natural Gas Production Pathways for the Storage of Renewable Energy. Wires Energ. Environ. 5 (3), 327–350. doi:10.1002/wene.189

Frontiers in Energy Research | www.frontiersin.org May 2021 | Volume 9 | Article 6250235

Mohammad et al. Natural Gas Key Alternative Energy

Gao, J., Jia, C., Zhang, M., Gu, F., Xu, G., and Su, F. (2013). Effect of Nickel Nanoparticle Size in Ni/α-Al2O3 on CO Methanation Reaction for the Production of Synthetic Natural Gas. Catal. Sci. Technol. 3 (8), 2009–2015. doi:10.1039/c3cy00139c

Gao, J., Liu, Q., Gu, F., Liu, B., Zhong, Z., and Su, F. (2015). Recent Advances in Methanation Catalysts for the Production of Synthetic Natural Gas. RSC Adv. 5 (29), 22759–22776. doi:10.1039/C4RA16114A

Guidolin, M., and Alpcan, T. (2019). Transition to Sustainable Energy Generation in Australia: Interplay between Coal, Gas and Renewables. Renew. Energ. 139, 359–367. doi:10.1016/j.renene.2019.02.045

Huang, Y. W., Kittner, N., and Kammen, D. M. (2019). ASEAN Grid Flexibility: Preparedness for Grid Integration of Renewable Energy. Energy Policy 128, 711–726. doi:10.1016/j.enpol.2019.01.025

IEA (2019). Southeast Asia Energy Outlook 2019. Paris: International Energy Agencyhttps://www.iea.org/reports/southeast-asia-energy-outlook-2019.

IRENA (2018). Renewable Energy Market Analysis: Southeast Asia. Abu Dhabi: International Renewable Energy Agency

Kadiyala, A., Kommalapati, R., and Huque, Z. (2016). Evaluation of the Life Cycle Greenhouse Gas Emissions from Different Biomass Feedstock Electricity Generation Systems. Sustainability 8 (11), 1181

Khan, M. N., Cloete, S., and Amini, S. (2020). Efficient Production of Clean Power and Hydrogen through Synergistic Integration of Chemical Looping Combustion and Reforming. Energies 13 (13), 3443. doi:10.3390/ en13133443

Kittner, N., Fadadu, R. P., Buckley, H. L., Schwarzman, M. R., and Kammen, D. M. (2018). Trace Metal Content of Coal Exacerbates Air-Pollution-Related Health Risks: The Case of Lignite Coal in Kosovo. Environ. Sci. Technol. 52 (4), 2359–2367. doi:10.1021/acs.est.7b04254

Kopyscinski, J., Schildhauer, T. J., and Biollaz, S. M. A. (2010). Production of Synthetic Natural Gas (SNG) from Coal and Dry Biomass - A Technology Review from 1950 to 2009. Fuel 89 (8), 1763–1783. doi:10.1016/j.fuel.2010. 01.027

Li, J., Zhou, L., Li, P., Zhu, Q., Gao, J., Gu, F., et al. (2013). Enhanced Fluidized Bed Methanation over a Ni/Al2O3 Catalyst for Production of Synthetic Natural Gas. Chem. Eng. J. 219, 183–189. doi:10.1016/j.cej.2013.01.005

Lidula, N. W. A., Mithulananthan, N., Ongsakul, W., Widjaya, C., and Henson, R. (2007). ASEAN towards Clean and Sustainable Energy: Potentials, Utilization

and Barriers. Renew. Energ. 32 (9), 1441–1452. doi:10.1016/j.renene.2006. 07.007

Mac Kinnon, M. A., Brouwer, J., and Samuelsen, S. (2018). The Role of Natural Gas and its Infrastructure in Mitigating Greenhouse Gas Emissions, Improving Regional Air Quality, and Renewable Resource Integration. Prog. Energ. Combustion Sci. 64, 62–92. doi:10.1016/j.pecs.2017.10.002

Shinde, V. M., and Madras, G. (2014). CO Methanation toward the Production of Synthetic Natural Gas over Highly Active Ni/TiO2catalyst. Aiche J. 60 (3), 1027–1035. doi:10.1002/aic.14304

Silitonga, R. J. P., and Anugrah, P. (2015). The Role of Natural Gas in ASEAN Energy Security. https://aseanenergy.org/the-role-of-natural-gas-in-asean- energy-security/. (Jakarta, Indonesia: ASEAN Centre for Energy (ACE)).

Sovacool, B. K. (2009). Energy Policy and Cooperation in Southeast Asia: The History, Challenges, and Implications of the Trans-ASEAN Gas Pipeline (TAGP) Network. Energy Policy 37 (6):2356–2367. doi:10.1016/j.enpol.2009.02.014

Steckel, J. C., Edenhofer, O., and Jakob, M. (2015). Drivers for the Renaissance of Coal. Proc. Natl. Acad. Sci. USA 112, E3775–E3781. doi:10.1073/pnas.1422722112

Tongsopit, S., Chaitusaney, S., Limmanee, A., Kittner, N., and Hoontrakul, P. (2015). Scaling up Solar PV: A Roadmap for Thailand. Energy Research Institute, Chulalongkorn University

Turconi, R., Boldrin, A., and Astrup, T. (2013). Life Cycle Assessment (LCA) of Electricity Generation Technologies: Overview, Comparability and Limitations. Renew. Sustainable Energ. Rev. 28, 555–565. doi:10.1016/j.rser.2013.08.013

Yusaf, T. F., Salleh, H., Lai, T. Y., and Ramesh, S. (2001). Economic Aspect of Natural Gas Vehicle in Malaysia. J. Ind. Technol. SIRIM, Malaysia 10, 25–36.

Conflict of Interest: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2021 Mohammad, Mohamad Ishak, Mustapa and Ayodele. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

Frontiers in Energy Research | www.frontiersin.org May 2021 | Volume 9 | Article 6250236

Mohammad et al. Natural Gas Key Alternative Energy

  • Natural Gas as a Key Alternative Energy Source in Sustainable Renewable Energy Transition: A Mini Review
    • Introduction
    • Natural Gas Reserves, Production, and Demand in ASEAN
    • An Overview of Synthetic Natural Gas Production Methods
    • Role of Natural Gas in the Sustainable Renewable Energy Transition
    • An Overview of Natural Gas Utilization for Power Generation
    • Life Cycle Analysis on Natural Gas Use for a Sustainable Renewable Energy Transition
    • Natural Gas Infrastructure and Renewable Energy Policies in ASEAN
    • Conclusion and Future Recommendations
    • Author Contributions
    • Funding
    • Acknowledgments
    • References

Natural gas 2.PDF

113

All the DUCs in a Row: Natural Gas Production in U.S.

Douglas Mugabe,a Levan Elbakidze,b and Tim Carrc

abstract

Using data from seven shale gas regions in the United States, we examine natural gas production in terms of drilling rig activity and well completion rates. Our objective is to examine the determinants of well completion decisions in the U.S. natural gas production. We observe that in recent years, the explanatory power of drilling rig count has declined. On the other hand, the number of producing wells remains a significant factor for explaining the variation in gas production. We find that an increase in the number of drilled but uncompleted wells (DUCs) plays a significant role in natural gas supply. The number of DUCs depends on drilling rig activity and futures prices of oil and natural gas. Also, our results indicate that well completion decisions and the duration of DUC status depend on oil and gas prices, pipeline capacity, producing well type and well depth. Keywords: Natural Gas Production, Rigs, Drilling, Completion, Pipelines, Prices.

https://doi.org/10.5547/01956574.42.3.dmug

1. INTRODUCTION

Understanding the determinants of natural gas supply is important because of its signifi- cance for the U.S. power sector (Peters and Hertel, 2017; Stephens, 2018) and U.S. economic activ- ity in general (Arora and Lieskovsky, 2014; Melick, 2014; Weber, 2012; Joskow, 2013).1 Previous academic literature relied on drilling rig activity (the count of actively drilling rigs) as the primary determinant of oil and gas production because of the simplicity, availability, and global applicability of drilling rig count as an indicator (Apergis, Ewing and Payne, 2016; Melek, 2015). The oil and gas industry also has been relying on the rig count as a measure of oil and gas production activity.2 However, as Figure 1 illustrates, natural gas production in the U.S. increased even though drilling activity has declined in recent years (EIA, 2019a).

With the growth in the use of hydraulic fracturing and horizontal drilling technologies, market analysts, researchers and government agencies have noted the increase in the inventory of drilled but uncompleted wells (DUCs) in the U.S. (Hegarty, 2017; EIA, 2013; EIA, 2019b; Dunning, 2016; Srinivasan, Krishnamurthy and Kaufman, 2019; IHS, 2016; Piotrowski, 2016). However, little or no systematic information is available on the growth of DUC inventory and the implications

1. Large number of studies document the relationship between energy in general and economic growth (See Hamilton, 2013).

2. Baker Hughes has been reporting rig count since 1944 (Baker Hughes, 2019).

a Corresponding author. Division of Resource Economics and Management; Regional Research Institute, West Virginia University, 333 Evansdale Drive, Morgantown, WV 26506-6108. E-mail: [email protected].

b Division of Resource Economics and Management; Center for Innovation in Gas Research and Utilization, West Virginia University.

c Department of Geology and Geography, West Virginia University. The Energy Journal, Vol. 42, No. 3. Copyright © 2021 by the IAEE. All rights reserved.

114 / The Energy Journal

All rights reserved. Copyright © 2021 by the IAEE.

for natural gas production. This paper examines the determinants of DUC inventories and the im- pacts of drilling rig activity and well completion on natural gas output in the U.S.

Technological developments in unconventional oil and gas (UOG) production have trans- formed the U.S. gas industry. According to the U.S. EIA, domestic production of gas from the UOG industry grew by more than 100% from 2000 to 2010. Data from the EIA (2016a) also indicate that the daily production of U.S. dry shale increased from 2.5 in 2002 to 43 billion cubic feet in 2016, with most of the new production coming from the Northern Appalachian basin (Marcellus and Utica shale units). Substantial gains in productivity continue through advances such as super pads (which can include up to 20 wells), extended horizontal laterals (reaching up to 20 thousand feet3) and im- proved drilling and fracturing technologies. The share of horizontally drilled wells increased from 3% in 2008 to 12% in 2017 (EIA, 2018). As a result, although the number of drilling rigs fell since 2014, natural gas production continued to grow (Figure 1) (EIA, 2019a).

In general, UOG production involves two stages. The first stage involves drilling, casing the well with multiple strings of steel pipe, and cementing the pipe. In the second stage (comple- tion), the steel casing is perforated, and the well is stimulated via hydraulic fracturing to initiate gas flow from fractured formations. Completion, which can be significantly more expensive and time consuming than the first stage activities, can be delayed indefinitely. However, interrupting the flow from a producing well can be prohibitively costly in terms of foregone income (Kleinberg et al., 2018). Hence, production timing decisions take the form of drilling and completion decisions corre- sponding to stages one and two, respectively (Mason and Roberts, 2018). Wells drilled (stage one), but not hydraulically fractured or completed are labeled as drilled but uncompleted wells (DUCs).

Figure 2 shows that the aggregate number of DUCs has increased since 2007 across all regions. From November 2016 to the end of 2017, the number of DUCs rose 37.4% to 7,493 (DI, 2016). EIA’s (2019a) drilling productivity report shows more than 8,700 DUCs as of November 2018. Growth in DUCs varies by region, with the largest increase observed in the Permian Basin. The reasons for the delays in well completion, and consequent growth in the DUC numbers, may

3. See Eclipse Purple Hayes well at 20,803 feet. https://www.hartenergy.com/exclusives/super-laterals-going-really- really-long-appalachia-31209.

Fig 1: U.S. Rig Activity and Natural Gas Production

All the DUCs in a Row: Natural Gas Production in U.S. / 115

Copyright © 2021 by the IAEE. All rights reserved.

include: shortage of hydraulic fracturing equipment and teams, contractual lease obligations that require active well development in stage one, pipeline capacity bottlenecks, and operators’ timing decisions to take advantage of favorable prices (EIA, 2019b; Kleinberg et al., 2018).

One implication of the increase in DUCs is that aggregate natural gas production depends less on drilling rig activity and more on well completion rates. As a result of growth in unconven- tional production, and associated two-stage production technology use, drilling rig counts no longer directly correspond to the number of producing wells. Hence, the number of completed wells may be increasingly important for modeling natural gas production. Though the drilling rig count re-

Figure 2: Rig Count, Drilled and Un-Completed Well Count, and Natural Gas Production Trends from 2007–2018

116 / The Energy Journal

All rights reserved. Copyright © 2021 by the IAEE.

mains an important factor in natural gas production, supply growth is achieved with fewer drilling rigs given improvements in drilling technology and despite the backlog of DUCs (EIA, 2019a).

Importantly, for the production to grow, the productivity of new wells must offset declines in productivity of legacy wells (Boyce and Nøstbakken, 2011). Therefore, this paper considers both the number of producing wells and the number of newly completed wells as drivers of natural gas output. The objectives of this study are threefold. First, we examine the role of well completion rates in explaining natural gas production. Second, we examine the determinants of DUC numbers, which represent the gap between drilled and completed wells. Third, we identify the factors that influence the length of time that operators take to complete the unconventional wells.

The literature on the determinants of natural gas production is limited. Iledare (1995) uses a supply model for natural gas reserve additions in West Virginia to study the responsiveness of drilling effort and gross reserve additions to changes in the expected wellhead price, taxes, resource depletion and reserve life index. He concludes that drilling activity shifts across geological forma- tions in response to varying geologic conditions and economic incentives. Boyce and Nøstbakken (2011) show a positive correlation between output prices and drilled wells, considering a significant decrease in the cost of drilling. Chen and Linn (2017) examine the effects of oil and gas futures prices on drilling activity in the U.S. and the rest of the world. They show that drilling activities re- spond to futures prices more than spot prices. This is consistent with the industry practice of hedging gas production. Gülen et al. (2013) also document the sensitivity of drilling new wells to changes in natural gas prices. Similar results with a positive association between oil rig activity and crude oil prices have been documented by Ringlund, Rosendahl and Skjerpen (2008), Apergis et al. (2016), Anderson, Kellogg, and Salant (2018) and Khalifa, Caporin and Hammoudeh (2017).

Mason and Roberts (2018) examine the sensitivity of well level natural gas production in Wyoming to geologic and economic factors. They show that geologic factors affect intra-well production variation (well productivity) while prices affect inter-well production changes (number of producing wells) via producer drilling decisions. They conclude that after a well has started producing, prices have limited effect on well-level production. Instead, geologic and engineering factors determine well productivity. However, prices have a significant effect on aggregate supply due to the elasticity of producers’ drilling decisions. The authors show that at lower prices, only the most productive wells are drilled, while higher prices enable drilling of less productive wells. They also observe that the elasticity of drilling decisions in Wyoming increased following the growth in the use of horizontal drilling and hydraulic fracturing technologies. Ikonnikova and Gulen (2015) also examine the effect of prices on drilling activities in Barnett, Haynesville, and Fayetteville shale units. They show that at lower prices, producers in some locations may find it more profitable to rely on low-cost infill4 wells to minimize capital costs as opposed to drilling relatively more productive but costlier wells in new locations.

None of the previous studies examine growth in the DUC numbers and the relationships between gas production, drilling rig activity and well completion across shale regions in the U.S. We disentangle these variables, which allows us to present a more nuanced account of production activities given the recent growth in the number of drilled but uncompleted wells. Our results doc- ument greater explanatory power of the number of producing wells relative to the count of active rigs for modeling natural gas production.5 We also show that changes in oil and gas futures prices

4. Infill wells are drilled and completed next to the existing wells as opposed to new locations. Infill wells are less pro- ductive but require lower upfront capital costs by taking advantage of existing infrastructure and existing lease arrangements.

5. Although the principles addressed in this study are applicable to both oil and gas production, we focus the analysis on unconventional gas production and reserve the analysis of oil production to future studies. We acknowledge that in some cases

All the DUCs in a Row: Natural Gas Production in U.S. / 117

Copyright © 2021 by the IAEE. All rights reserved.

and drilling rig activity affect DUC numbers and the length of time that operators take to complete individual wells.

2. DATA

Unconventional shale gas production makes up more than 50% of all-natural gas produced in the U.S and its contribution continues to increase with most of the production coming from seven major shale regions (EIA, 2017). This study is based on the data from Anadarko, Appalachia (Mar- cellus and Utica), Bakken, Eagle Ford, Haynesville, Niobrara and Permian regions.

We use monthly regional data from January 2007 to July 2018 to examine cumulative nat- ural gas production and DUC counts, and daily well level data from 2000 to 2018 to estimate hazard ratios.6 The data summary is presented in Table 1. Rig count and natural gas production7 (million cubic feet - mmcf) data are obtained from the EIA. Well completion data obtained from DrillingInfo (now Enverus) include a monthly cumulative number of producing wells. Rig count data (disregard- ing the differences in rig requirements across regions due to geological characteristics) are provided by Baker Hughes.

Rig activity in this study reflects only the number of actively8 drilling rigs. Figure 2 presents data trends for drilling rig counts, DUCs, and gas production. Since 2007, natural gas production has been increasing significantly in most regions, except in Niobrara and Haynesville. Substantial increase in production can be attributed to significant gains in productivity enabled by recent tech- nological improvements. Haynesville lies deeper than the shale reservoirs in other regions making supply sensitive to price variation. Drilling rig activity in this region went down significantly and in 2016 drilling rig count dropped to 20.

To examine the growth in the number of DUCs, we use estimated monthly count of DUCs from January 2007 to July 2018.9 Figure 2 shows that the numbers of DUCs have been increasing since 2007, with greater increases observed in Permian, Niobrara, and Anadarko regions. How- ever, in Appalachia and Eagle Ford regions, the numbers of DUCs have decreased since 2014. The declines in the numbers of DUCs in Appalachia and Eagle Ford imply that completion has been outpacing drilling of new wells.

To explain the variation across regions and over time, we control for pipeline capacity, drilling rig count, and futures prices of natural gas (measured in dollars per thousand cubic feet)

oil and gas production is joint. For example, unconventional production in the Permian basin is primarily aimed at oil with associated gas production.

6. Natural gas production and DUC count analysis covers 2007 to 2018 because EIA data on monthly rig count and production per region are available only starting from January 2007. We used an expanded sample time frame in the survival analysis from January 2000 to July 2018 based on DUC duration data availability.

7. EIA estimates natural gas production using data reported by various industry sources. In this study, we use up to date natural gas production numbers as reported by the EIA.

8. The rig is active if it is drilling at least 15 days during the month. This measure excludes rigs involved in non-drilling activities like workovers and production testing. This definition is consistent with EIA (2019a) and Baker Hughes (2019).

9. Estimates of DUC numbers can vary depending on methodologies, assumptions, and availability of data. EIA counts a drilled well to be uncompleted after 20 days’ post spudding (EIA, 2016b). EIA started providing DUC count as of December 2013. To increase the sample size, we estimate DUCs using DrillingInfo well level data from January 2000 to July 2018 fol- lowing the EIA methodology. Comparison of EIA DUC data and our estimated DUC numbers after 2013 reveals insignificant mean difference at 5% significance level in most regions except Appalachia and Permian. In these regions, the difference is insignificant at 1% level. The comparisons are available upon request. The minor difference in some regions can be due to the estimation method. Our computations account for DUCs drilled since 2000. On the other hand, EIA excludes wells drilled prior to December 2013.

118 / The Energy Journal

All rights reserved. Copyright © 2021 by the IAEE.

Table 1: Descriptive Statistics Region Variable N. Mean Std. Dev. Min. Max. All Regions N.Gas Futures price ($/mcf)** 6,787 4.91 2.29 1.35 14.74 Oil Futures price ($/b)** 6,787 62.41 27.20 14.06 145.90 Anadarko Pipeline Capacity (103 mmcfd)** 139 16.3 2.7 13.3 19.3 N.Gas Production (103mmcf)/month* 139 5.1 0.8 3.9 7.0 Rig Count/month* 139 155 55 55 247 DUC Count 103/month* 139 1.1 0.4 0.4 1.7 Producing Well Count 103/month* 139 16.2 2.7 9.8 19.7

DUC Durationa (Days)** 14,840 88 121 1 1,778 UOG Well Measured Depthb (103Ft)** 14,781 13.4 3.8 0.1 38.9

Appalachia Pipeline Capacity (103 mmcfd)** 139 40.1 11.1 29.7 59.8 N.Gas Production (103mmcf)/month* 139 11.2 9.1 1.3 28.7 Rig Count/month* 139 90 33 36 154 DUC Count 103/month* 139 1.1 0.7 0.01 2.0 Producing Well Count 103/month* 139 47.6 7.2 20.5 54.9

DUC Duration (Days)** 14,649 317 249 1 1821 UOG Well Measured Depth (103Ft)** 14,529 12.6 3.8 0.04 40.0

Bakken Pipeline Capacity (103mmcfd)** 139 7.7 0.4 7.2 8.1 N.Gas Production (103mmcf)/month* 139 0.9 0.7 0.2 2.4 Rig Count/month* 139 105 64 24 218 DUC Count 103/month* 139 0.9 0.5 0.2 1.7 Producing Well Count 103/month* 139 6.2 4.7 0.5 13.2

DUC Duration (Days)** 15,738 142 153 1 1656 UOG Well Measured Depth (103Ft)** 15,705 18.8 3.4 1.9 27.2

Eagle Ford Pipeline Capacity (103 mmcfd)** 139 5.5 1.5 3.7 7.8 N.Gas Production (103mmcf)/month* 139 4.2 2.2 1.5 7.4 Rig Count/month* 139 134 87 30 279 DUC Count 103/month* 139 1.3 0.6 0.3 2.4 Producing Well Count 103/month* 139 7.4 2.9 3.0 11.6

DUC Duration (Days)** 25,230 149 204 1 1,821 UOG Well Measured Depth (103Ft)** 25,225 14.9 3.2 0.4 39.4

Haynesville Pipeline Capacity (103 mmcfd)** 139 38.4 6.0 30.4 46.8 N.Gas Production (103mmcf)/month* 139 6.7 1.9 3.6 10.6 Rig Count/month* 139 104 73 16 244 DUC Count 103/month* 139 0.6 0.2 0.4 1.0 Producing Well Count 103/month* 139 15.7 3.0 8.1 18.7

DUC Duration (Days)** 7,965 116 137 1 1,728 UOG Well Measured Depth (103Ft)** 7,949 14.9 3.6 1.1 39.9

Niobrara Pipeline Capacity (103 mmcfd)** 139 21.2 6.4 10.6 27.3 N.Gas Production (103mmcf)/month* 139 4.5 0.3 3.4 5.1 Rig Count/month* 139 72 30 16 127 DUC Count 103/month* 139 0.5 0.2 0.2 0.9 Producing Well Count 103/month* 139 2.5 0.7 1.2 3.3

DUC Duration (Days)** 11,872 133 150 1 1,821 UOG Well Measured Depth (103Ft)** 11,373 11.5 3.6 0.4 40.0

Permian Pipeline Capacity (103 mmcfd)** 139 16.6 2.5 13.3 20.2 N.Gas Production (103mmcf)/month* 139 5.7 1.7 3.8 1.0 Rig Count/month* 139 335 132 92 565 DUC Count 103/month* 139 2.5 0.8 0.4 4.3 Producing Well Count 103/month* 139 109.3 21.8 30.4 136.9

DUC Duration (Days)** 36,513 132 182 1 1,822 UOG Well Measured Depth (103Ft)** 36,486 13.1 4.1 0.4 40.0

Note: **Data is from 2000- 2018 and *Data is from 2007–2018. a DUC duration variable measures the length of time in days from end of drilling (spud date plus 20 days) to well comple- tion or to first production for only completed unconventional wells. Minimum DUC duration of 1 day indicates that every region has at least 1 well which was completed in 21 days after spudding. Maximum DUC duration reflects maximum duration before the DUC is treated as “dead”. For the purpose of this study, outlier wells (wells drilled and not completed within the period of 5 years) are treated as “dead” DUCs and they constitute about 0.3% of our data. b Well measured depth is the borehole and horizontal length of unconventional wells (horizontal and directional). We do not have access to the information on lateral length, number of fracking stages and proppant intensity. Such data likely would have improved the accuracy of our DUC duration analysis.

All the DUCs in a Row: Natural Gas Production in U.S. / 119

Copyright © 2021 by the IAEE. All rights reserved.

and oil (measured in dollars per barrel). Following Chen and Linn (2017), we compute average fu- tures prices of natural gas and oil using all available, ,m futures contracts from the trading floor of the New York Mercantile Exchange (NYMEX). We define futures price (Ft) at time t as a function

of the contract prices such that , 1

1 ( )

n

t t m m

F C n =

= ∑ , where Ct,m denotes the price of the m-th contract at time t. Contract prices and natural gas pipeline capacity data are obtained from the EIA. Pipeline capacity measures outflow volume of pipeline infrastructure expressed in million cubic feet per day (mmcf/d). Table 1 indicates that pipeline capacity increased the most in the Appalachian region with more than 30,000 mmcf added between 2007 to 2018. On the other hand, Bakken experienced the least expansion in pipeline capacity with less than 1000mmcf added over the same period. In Eagle Ford, Niobrara, Haynesville, Anadarko and Permian capacities increased by 4,141mmcf, 16,708mmcf, 16,386mmcf, 5,968mmcf and 6,920mmcf respectively.

In the time-to-event (survival) analysis of DUC duration status, we use individual well level data from January 2000 to July 2018. DUC duration status for an individual unconventional well is the number of days between the end of stage one10 and completion. Completion date in our analysis is the earliest of the reported well completion date or the date of first reported production.11 Summary statistics of DUC duration are presented in Table 1.

3. EMPIRICAL STRATEGY

Our empirical strategy includes: a) the analysis of natural gas production in terms of drill- ing rig counts and producing wells using linear fixed effects and vector autoregressive models, b) the analysis of DUC Counts within and across regions using linear fixed effects regressions, and c) the analysis of individual DUC duration status using survival analysis technique.

Natural gas production

We first use a linear regression model in double log form to explore the effect of (lagged) rig count (RC) and producing wells (PW) on natural gas production (NGP) individually and in combination. Next, we estimate autoregressive models as a robustness check. We test for unit roots using Phillips-Perron (Phillips and Perron, 1988), Augmented Dickey-Fuller (Dickey and Fuller, 1981) and panel Levin-Lin-Chu (Levin, et al., 2002) statistics. Subsequently, we conduct a panel cointegration analysis to determine the long-run relationship between natural gas production, rig count, and the number of producing wells.12 The Pedroni’s heterogeneous panel cointegration test is used to test for the group and bivariate cointegration relationships. We compute four panel and three group statistics following Neal (2014) based on the ‘within’ and the ‘between’ dimensions respectively (Pedroni 1999, 2004). We also test for cointegration within each region using the Jo- hansen test (Johansen, 1988, 1995a, b). We proceed with estimating a panel VAR with generalized methods of moments (Abrigo and Love, 2016). Next, we estimate each region’s VEC (vector error correction) model to account for cointegration within regions (Engle and Granger, 1987). The VEC model is specified as follows:

10. Following EIA methodology, we assume that stage one takes 20 days on average. 11. Many wells are completed/fractured more than once, and the data do not indicate whether a specific completion date

corresponds to first completion or a recompletion. Therefore, we use the earlier of the first production or completion dates to avoid re-completion entries.

12. Following Liew (2004), Hannan-Quinn criterion (HQC) (Hannan and Quinn 1979) is used to determine the appro- priate lag length for each series in each region.

120 / The Energy Journal

All rights reserved. Copyright © 2021 by the IAEE.

1 1 1 1 1

p p

kt k kt ki kt i ki kt i i i

Y Y Xα δη β− − − = =

∆ = + + ∆ + Φ ∆ +…∑ ∑

1

... p

jki jkt i kt i

X ε− =

+ Φ ∆ +∑ (1)

where Δ is the first difference operator; Ykt is natural gas production in log form in region k and period t; Xj is the j-th explanatory variable in log form; η are the residuals from the cointegration vector; p is the optimal lag length; kα is the intercept, i is the lag length, and tε is the error term. ,δ β and Φ are the parameters.

DUC counts

To examine the growth in the number of DUCs, we use regional fixed effects regression models in log-log form with and without time fixed effects,13 with first differences, and standardized variables. The independent variables include pipeline capacity, drilling rig count, natural gas and oil futures prices. Futures prices (FP) rather than spot prices are used following Chen and Linn (2017) who showed that futures prices have a more significant effect on natural gas production than spot prices. The futures prices (FP) are lagged to account for the time that it takes the operators to initiate production in response to price movements (Osmundsen et al., 2015). We use standardized variables obtained by subtracting the mean (across regions and within regions) and pipeline ca- pacity expressed in first difference to estimate the fixed effects regression model.14 Standardization approach reduces the scale of variables but preserves the interpretation of the regression coefficients to represent the mean change in the DUC given a unit change in the independent variable.

DUC Duration

In this analysis, we are interested in examining the factors that influence the length of time that operators take to complete the drilled wells. Time to event (duration/survival) analysis (see Sy and Taylor, 2000; Box-Steffensmeier and Zorn, 2001; Fleming and Harrington, 2011; Hernandez and Dresdner, 2010) is used to analyze DUC duration data. We define a random variable T with a continuous probability distribution function ( )f t to represent DUC duration, or the number of days from the end of drilling to completion. The probability that a drilled well is completed in t days is given by ( ) ( )F t Prob T t= < . Correspondingly, the survival function, or the probability of a drilled well not being completed in t days, is ( ) ( )1S t F t= − . The hazard rate ( ) ( ) ( )( /t f t S tλ = ), is the probability that a drilled well will be completed at time, t, given that it was not completed prior to t. We use semi-parametric15 Cox proportional hazard model (equation 2) (Cox, 1972) to represent the hazard function in the DUC duration analysis (Stogiannis et al. 2011).

( ) ( )0| , (t x t exp Xλ β λ= ′ β) (2)

where β is a vector of unknown parameters of X covariates, ( )0 tλ is the baseline hazard function when ( ) ( )0| 0t x tλ λ= = and can take any form as a function of t. The effects of covariates can be represented in various specifications of the hazard function.

13. Hausman test (Chi2(5) =78.04; Prob>Chi2=0.00) indicted superiority of Fixed Effects regression over a Random Effects model. Joint F test results (F (135, 820) =1.92 Prob>F=0.00) suggest including time fixed effects.

14. After standardization all VIFs were less than 5 with mean of 2.81 (Hair, Anderson, Tatham, and Black, 1995). 15. We also estimate parametric specifications including exponential, Weibull and Gompertz functions. These results are

available on request.

All the DUCs in a Row: Natural Gas Production in U.S. / 121

Copyright © 2021 by the IAEE. All rights reserved.

4. RESULTS

We start with examining the difference in the relationship between natural gas production and rig count (RC) before and after February 2009.16 In addition to the expansion in unconventional gas production, this breakpoint is also close to the economic downturn and to the beginning of the new U.S. administration. Each of these factors could have contributed to the structural break timing. Nevertheless, we believe that our breakpoint adequately reflects the changes in natural gas produc- tion series and enables meaningful comparison of production pre and post 2009.

Table 2: Split Sample Regional Fixed Effects Results for Aggregate NGP

Dependent–NGP Before Feb 2009 After Feb 2009

Rig Countt–1 0.212 (0.04)*** –0.048 (0.03) Constant 13.55 (0.17)*** 15.54 (0.15)***

R-sq 0.17 0.010 Observations n=7, T=25, N=175 n=7, T=113, N=791

Note: Significance values 1%***, 5%**, 10%*; Standard errors in parenthesis.

The results from regional fixed effects regression models with lagged RC are presented in Table 2. These results show a significant change in the explanatory power of lagged RC for natural gas production (NGP). The rig count is positively correlated with natural gas production prior to February 2009. However, after February 2009 RC has a statistically insignificant relationship with natural gas production and a weaker explanatory power. A similar loss of explanatory power of RC is found with heterogeneous break point dates across regions (see Table A1 in the online Appendix).

4.1 Determinants of Natural Gas Production (NGP)

Table 3 shows regression results with region fixed effects and logged NGP as the depen- dent variable. Three model results are presented. R-squared values show that models 2 and 3, which include producing wells (PW) explain more of the variation in NGP than model 1 (with only RC). The marginal contribution of producing wells as an explanatory variable relative to the rig count is significant, as revealed by the difference in R-squared values between models 1 and 2. Comparison of models 2 and 3 illustrates that although rig count is statistically significant and remains to be a meaningful determinant of NGP, its marginal contribution to explaining the variation in natural gas production is smaller relative to the number of producing wells. These results are robust under heterogeneous break period specification across producing regions (see online Appendix Table A2).

To explore the relationship at the regional scale, we estimate models 1, 2 and 3 for each region individually. The results presented in Table 4 are consistent with the results in Table 3, with all but two of the regions showing statistically significant effects of producing wells (PW). In some of the regions, rig count has a negative coefficient as natural gas production increased despite the de- clining number of active rigs. The estimated adjusted R-squared varies among regions and between models. However, in all cases, models 2 and 3, which include the number of producing wells, show

16. We test for the presence of a structural break using Wald-type tests (Vogelsang, 1997; Andrews 1993; Andrews and Ploberger 1994) in the linear regression of natural gas production (NGP) and rig count (RC). We estimate a linear regression model and compute the S-wald test statistic for an unknown break. This method is also used to identify the breaks (Bi) for each region independently. The results show May 2010 for Anadarko, August 2012 for Appalachia, December 2012 for Bakken, February 2013 for Eagle Ford, February 2009 for Haynesville and January 2014 for Niobrara and Permian.

122 / The Energy Journal

All rights reserved. Copyright © 2021 by the IAEE.

better fits compared to model 1. The rig count is a significant indicator for natural gas production in some regions. However, in most regions the rig count is not as informative as the number of produc- ing wells, which accounts for well completions. Similar conclusions are reached in the models with heterogeneous break points across regions (see Table A3 in the online Appendix).

Table 3: Region and Time Fixed Effects Results for Aggregate NGP NGP-Dep Model 1 Model 2 Model 3

Rig Countt–1 0.332 (0.03)*** 0.274 (0.03)*** Producing Wells 0.524 (0.04)*** 0.468 (0.03)*** Feb 2009 1.413 (0.22)*** 1.044 (0.22)*** 1.093 (0.21)*** Constant 12.94 (0.22)*** 9.906 (0.41)*** 9.132 (0.41)***

R-sq 0.58 0.61 0.64

Observations Balanced Panel n=7, T=138, N=966

Note: Significance values 1%***, 5%**, 10%*; Standard Errors in parenthesis; Data from 2007–2018.

Table 4: Regional Regression Results (Determinants of NGP) Region Model 1 Model 2 Model 3

Anadarko Rig Countt–1 –0.003 (0.03) 0.002 (0.02) Producing Wells 0.894 (0.07)*** 0.908 (0.07)*** Feb 2009 0.202 (0.03)*** –0.126 (0.03)*** –0.126 (0.03)*** Adj R-sq 0.230 0.633 0.630

Appalachia Rig Countt–1 –0.663 (0.19)*** –0.521 (0.16)*** Producing Wells 3.978 (0.50)*** 3.874 (0.49)*** Feb 2009 2.222 (0.20)*** 0.596 (0.22)*** 0.915 (0.24)*** Adj R-sq 0.482 0.621 0.642

Bakken Rig Countt–1 –0.264 (0.09)*** –0.083 (0.03)** Producing Wells 0.896 (0.03)*** 0.884 (0.03)*** Feb 2009 1.571 (0.15)*** –0.318 (0.08)*** –0.245 (0.09)*** Adj R-sq 0.436 0.916 0.919

Eagle Ford Rig Countt–1 0.152 (0.06)** 0.153 (0.02)*** Producing Wells 1.525 (0.05)*** 1.530 (0.04)*** Feb 2009 0.838 (0.12)*** –0.235 (0.05)*** –0.337 (0.04)*** Adj R-sq 0.379 0.931 0.937

Haynesville Rig Countt–1 0.077 (0.02)*** 0.160 (0.03)*** Producing Wells 0.177 (0.14) 0.771 (0.17)*** Feb 2009 0.653 (0.05)*** 0.490 (0.08)*** 0.391 (0.07)*** Adj R-sq 0.587 0.571 0.638

Niobrara Rig Countt–1 0.011 (0.01) 0.010 (0.02) Producing Wells –0.023 (0.03) –0.028 (0.03) Feb 2009 0.122 (0.02)*** 0.136 (0.02) 0.138 (0.02)*** Adj R-sq 0.363 0.382 0.203

Permian Rig Countt–1 –0.006 (0.05) –0.121 (0.04)*** Producing Wells 1.255 (0.12)*** 1.390 (0.13)*** Feb 2009 0.172 (0.06)*** –0.387 (0.07)*** –0.408 (0.07)*** Adj R-sq 0.051 0.457 0.497

Note: Significance values: 1%***, 5%**, 10%*; Standard errors in parenthesis; Data from 2007–2018.

All the DUCs in a Row: Natural Gas Production in U.S. / 123

Copyright © 2021 by the IAEE. All rights reserved.

In Table 5 we show the results from the regression where new wells are separated from older (legacy) wells. In this model, new wells represent cumulative number of wells that started producing up to three months ago. New well completions reflect the effect of higher initial produc- tivity of new wells and have a statistically significant effect.17 The new wells contribute to total gas production only after completion, which can be delayed indefinitely after drilling. The delays in completion weaken the correspondence between drilling rates and aggregate gas production. Hence, well completion decisions have a significant impact on aggregate natural gas production following the growth in UOG production. The conclusions are robust with regards to heterogeneous break- points across regions (see Table A4 in the online Appendix).

Table 5: Region and Time Fixed Effects Results (NGP and New Wells)

Dependent–ΔNGP Log-log form

Legacy Wells –0.012 (0.00)*** New Wells 0.009 (0.00)*** Feb 2009 0.042 (0.02)*** Constant 0.070 (0.03)**

R-sq 0.26

Note: Significance values 1%***, 5%**, 10%*; Standard Errors in parenthesis.

Next, we turn to the panel vector autoregressive models. First, we perform several diag- nostic tests. Unit root tests indicate that all variables are non-stationary in levels at the regional and aggregate scales. However, we reject the null hypothesis that the differenced variables contain a unit root at 1% significance level (see Table A5 in the online Appendix for the first-differenced variables, with and without a trend). We use a lag length of four as determined by the HQC test (see Table A6 in the online Appendix). The Pedroni’s heterogeneous panel cointegration test is used to determine the long-run relationships between variables (see Table A7 in the online Appendix), and indicates that panel rho-statistic, panel PP statistic, group rho-statistics and group PP-statistics fail to reject the null hypothesis of no cointegration at the 0.1 significance level.18 However, panel ADF t-statis- tic and group ADF-statistics reject the null hypothesis at the 0.05 significance level. Conversely, the Johansen test for cointegration (Johansen, 1988, 1995a, b) reveals cointegration within regions between some of the variables in our specifications (see online Appendix Table A8). Therefore, we reject the null of zero co-integrating vectors within regions using the trace statistic and conclude that there is at least one co-integrating vector in our specifications, which include natural gas production (NGP), rig count (RC) and producing wells (PW).

The results for the panel vector autoregressive models with regional fixed effects are pre- sented in the appendix (see online Appendix Table A9). The rig count is statistically not significant in the first three models. On the other hand, the lagged number of producing wells is significant. We also estimate the Vector Error Correction model with NGP as a function of RC and PW for each re- gion. Results are presented in the appendix section (see online Appendix Table A10). The rig count has a statistically insignificant effect in three of the seven regions. In Bakken, Eagle Ford, Niobrara

17. We also estimated the model where new wells include those that have started producing longer than three months ago. The results, available upon request, confirm declining productivity after approximately a year.

18. The panel VEC estimation follows two steps. First is the estimation of long run relationship using the following model, , , ,

1

n

k t k k i i k t kt i

Y t Xα δ β ε =

= + + +∑ , to obtain the estimated residuals ktε which form the error correction term in the panel VEC model (see Jiang and Liu, 2014). In the second step, equation 1 is estimated as a panel VAR with the error correction term.

124 / The Energy Journal

All rights reserved. Copyright © 2021 by the IAEE.

and Permian regions, the rig count has a negative and significant coefficient indicating growth in natural gas production despite a declining number of active rigs. This is consistent with our previous results and with the report by the Federal Reserve Bank of Dallas (2019). These results suggest that well productivity and the number of producing wells, which depends on well completion rates, are important determinants of natural gas production.

Overall, the results show that there is a significant relationship between the cumulative number of producing wells and natural gas production. However, the strength of the relationship differs across regions. In comparison, the drilling rig count is statistically weaker in explaining natural gas production. Delays in unconventional well completions, and growth in the number of DUCs, have introduced an additional layer of disparity between drilling rig count and natural gas production. We examine the determinants of the number of DUCs in the next section.

4.2 DUCs Analysis

Region and time fixed effects models are used to examine the number of DUCs as a func- tion of pipeline capacity (Cap), rig count (RC), and natural gas and oil futures prices (FP). The results in Table 6 are consistent with expectations. We observe that futures prices of natural gas and oil have statistically significant and negative effects on the number of DUCs. When futures prices are high, more wells are completed, and DUC numbers decline. This result is consistent with oper- ators selling at favorable prices to cover well completion costs by taking advantage of high initial well production rates. With futures and forward contracts locked in, the operators attract investors to front the money needed for well completion. This result supports the insight that operators defer well completions, leading to high DUC numbers, in anticipation of better oil and natural gas prices (Andrien, 2016; Kleinberg, 2018).

Region fixed effects regression results show that pipeline infrastructure is not a statistically significant factor in explaining DUC numbers.19 Statistical insignificance of pipeline capacity in these models can be due to a lack of variability in pipeline capacity within each region over time. The individual region results in Table 7 confirm the results from the aggregate analysis where rig count and futures prices have positive and negative effects on DUC numbers, respectively. The results also show that, as one would expect, an increase in drilling activity, measured in terms of the number of active drilling rigs, has a statistically significant and positive effect on the number of DUCs. All else constant, greater drilling activities lead to a greater number of DUCs.

4.3 DUC Duration Status Analysis

Next, we examine the length of time that operators take to complete each unconventional well. We use the well level DUC duration status data to examine completion timing. Non-paramet- ric survival functions are presented in Figure 3 using data from 2000 to 2018. The Kaplan-Meier survival curves show the proportion of wells that remain uncompleted over time. Most wells (about 90%) are completed within a year. An insignificant number of outlier DUCs (about 0.3%) remain uncompleted after five years. In this study, such wells are treated as “dead”20 DUCs and are excluded from the regression analyses.

19. We also estimated a regional and time fixed effects regression model using first differences of the explanatory vari- ables. The results show significant negative effect of oil and gas futures prices on DUC growth. However, pipeline capacity and drilling activity are not significant. These results are available on request.

20. Example of such definition can be found in Andrien (2016) where “dead” DUCs are defined as wells which fail to be completed even at better oil and gas prices.

All the DUCs in a Row: Natural Gas Production in U.S. / 125

Copyright © 2021 by the IAEE. All rights reserved.

We use linear regression and time-to-event (survival) models to obtain statistical estimates for the factors that explain the length of time taken to complete drilled wells. The generalized linear model is used to illustrate the general baseline relationship between DUC duration and the ex- planatory variables. However, survival analysis is more appropriate to represent the duration data adequately and to provide a more detailed account using both the survival and hazard functions. The survival function represents the probability that a well remains uncompleted at any given time, while the hazard function gives the probability that a well will be completed in a given period as- suming that it has not yet been completed.

The Results for the generalized linear (column A) and semi-parametric Cox proportional (columns B and C) models with logged days of DUC duration status are presented in Table 8. Cross region variation is captured using dummy variables with Anadarko as the base category. Generalized linear model results show that all variables are statistically significant with expected signs. Pipeline capacity, natural gas and oil futures prices have statistically significant and negative effects on the duration of the DUC status.21 On the other hand, well depth has a positive effect on DUC duration. Interpretation of the coefficients in the Cox proportional survival model (column B) should be oppo-

21. Pipeline capacity limitations have been especially prominent in the Permian basin leading to negative natural gas prices and increase in the number of DUCs (Addison, 2018; Surran, 2019).

Table 6: Drilled and Un-Completed Well Analysis Regional Results Region

Fixed Effects Region and Time

Fixed Effects

DUC –dependent A. Log-log B. Log-log C. Variables standardized across

regions

D. Variables standardized by region

Pipeline Capacity 1.668 (0.86) 0.329 (0.17) 0.232 (0.24) ΔPipeline Capacity — 2.274 (1.73) — — Rig Countt–1 0.448 (0.10) *** 0.585 (0.14) *** 0.485 (0.11) *** 0.412 (0.09) *** NG Futures price –0.213 (0.14) –2.337 (0.69) ** –0.636 (0.16) *** –1.493 (0.73)* Oil Futures price –0.031 (0.15) –4.112 (1.01) *** –0.765 (0.16) *** –1.841 (0.86)* Time 0.007(0.00)** Constant –11.75 (8.71) 24.34 (4.48)*** –0.218 (0.13) –1.008 (0.39)**

Adj R-sq 0.717 0.735 0.792 0.721 Observations 966 966 966 966

Note: Significance values 1%***, 5%**, 10%*; Robust standard errors in parenthesis; Data from 2007–2018.

Table 7: OLS Log-log Results for DUC Well Analysis per Region Dependent- DUC Anadarko Appalachia Bakken Eagle Ford Haynesville Niobrara Permian

ΔPipeline Capacity –0.832 (0.35)

0.767 (2.00)

–0.861 (1.45)

–0.544 (1.47)

0.374 (1.17)

–0.081 (0.26)

–0.781 (1.95)

Rig Countt–1 0.281*** (0.07)

0.462* (0.24)

0.278*** (0.07)

0.510*** (0.05)

0.247*** (0.02)

0.180*** (0.05)

0.354*** (0.03)

NG Futures price –1.089*** (0.08)

–2.949*** (0.31)

–1.363*** (0.12)

–0.934*** (0.10)

–0.666*** (0.06)

–0.262*** (0.07)

–0.391*** (0.06)

Oil Futures price 0.079 (0.10)

0.646 (0.41)

0.158 (0.17)

–0.121 (0.12)

–0.295*** (0.05)

–0.234*** (0.08)

–0.382*** (0.08)

Constant 6.735*** (0.35)

5.747*** (0.89)

6.624*** (0.46)

6.442*** (0.35)

5.035*** (0.18)

8.302*** (0.27)

7.937*** (0.23)

R-sq 0.740 0.747 0.728 0.831 0.634 0.532 0.720 Observations 138 138 138 138 138 138 138

Note: Significance values 1%***, 5%**, 10%*; Standard errors in parenthesis; Data from 2007–2018.

126 / The Energy Journal

All rights reserved. Copyright © 2021 by the IAEE.

site of the estimated signs (see Teachman and Hayward, 1993 for interpretation of hazard models). A positive coefficient indicates a negative effect on the probability that a well remains uncompleted (longer DUC duration). For example, our results show that an increase in natural gas and oil futures prices decreases the probability that a well will remain uncompleted at any given time, which im- plies a decrease in the DUC duration status. On the other hand, the length of the unconventional well has a positive effect on the duration of DUC status. Similarly, we observe that from 2000 to 2018, the probability that an unconventional well remains uncompleted at any given time has increased.

Figure 3: Kaplan-Meier Survival Curves

Table 8: DUC Duration Analysis Results Variables Generalized linear model Semi-Parametric Cox Proportional Model

Dep-DUC Duration A. Coef. B. Coef. C. Hazard Ratio

LL –139760 –1346626 –1401168 LR Chi2(12) 20062 (0.00) 20062 (0.00)

NG Futures price –0.041 (0.01)*** 0.052 (0.01)*** 1.053 (0.01)** Oil Futures price –0.051 (0.01)*** 0.118 (0.01)*** 1.126 (0.01)*** Pipeline capacity –0.502 (0.03)*** 1.013 (0.05)*** 2.753 (0.13)*** Gas Wella 0.232 (0.01)*** –0.269 (0.01)*** 0.764 (0.01)*** Well depth 0.498 (0.01)*** –0.350 (0.01)*** 0.705 (0.00)*** Time 0.0001(0.00)*** –0.0002(0.00)*** 0.999 (0.00)*** Appalachia 1.477 (0.03)*** –1.960 (0.04)*** 0.141 (0.01)*** Bakken –0.320 (0.03)*** 0.651 (0.04)*** 1.917 (0.08)*** Eagle Ford –0.225 (0.03)*** 0.594 (0.05)*** 1.811 (0.09)*** Haynesville 0.544 (0.03)*** –1.061 (0.04)*** 0.346 (0.02)*** Niobrara 0.621 (0.01)*** –0.869 (0.02)*** 0.419 (0.01)*** Permian 0.288 (0.01)*** –0.411 (0.01)*** 0.663 (0.01)***

Observations 126,048 127,627 127,627 Number of Completions 126,048 126,048 126,048

Note: Significance values 1%***, 5%**, 10%*; Standard errors in parenthesis; Data from 2000–2018. a Gas well is a dummy variable (with 1=Natural Gas producing well and 0=Oil producing well) that captures production type as defined by the operator. Wells are classified based on their gas/oil ratio (GOR).

All the DUCs in a Row: Natural Gas Production in U.S. / 127

Copyright © 2021 by the IAEE. All rights reserved.

The estimates for the hazard rates (the probability that a well will be completed at time t given that the well has not been completed prior to t) in column C are consistent with the estimates from the linear regression model results (column A) and prior expectations. A hazard ratio greater (less) than one indicates that a unit increase in the covariate is associated with an increase (decrease) in the probability that a well will be completed at any given time t, given that it is still in DUC status at time t–1. For example, based on the estimates from column C, all else constant, a one dollar in- crease in natural gas price is associated with 5.3% increase in the hazard rate. Similarly, a unit (103 mmcfd) increase in pipeline capacity is associated with 175% increase in hazard rate, on average across regions. This result illustrates the significance of pipeline infrastructure for unconventional well completion decisions. On the other hand, a unit (103 ft) increase in the well depth of an uncon- ventional well is associated with 0.295% (1–0.705) decrease in hazard rate.

The results also show that both survival and hazard rates differ significantly across regions and that gas wells are more likely to have lengthier DUC periods than primarily oil producing wells. This result, in combination with the significance of pipeline capacity, is possibly indicative of more pressing pipeline bottlenecks in natural gas supply than in oil. We also observe that well depth has a negative effect on the probability of completion at any given time. These results, in general, sug- gest that prices, infrastructure, and geologic variables play important roles in operators’ decisions to complete unconventional gas wells. This is consistent with the results in recent literature where prices and geologic factors are reported to be significant determinants of unconventional oil and gas production decisions (Mason and Roberts, 2018; Kleinberg et al., 2018; Ikonnikova and Gülen, 2015).

5. CONCLUSION

The U.S. natural gas production industry has experienced tremendous growth in the recent decade due to the developments in unconventional oil and gas extraction technologies. This growth has affected domestic and international energy markets (Oglend, et al., 2016), electricity generation sector (Peters and Hertel, 2017; Logan et al., 2013), industrial manufacturing sectors (Arora and Lieskovsky, 2014) and labor markets (Agerton, et al., 2017). Therefore, it is important to identify key interdependencies in the natural gas industry for appropriate market analysis and effective pol- icy formulation. The objective of this study is to explain the observed variability in the U.S. natural gas output in terms of the drilling rig count, the number of producing wells, and the completion of drilled unconventional wells. We are particularly interested in the observed growth of the number and duration of DUCs in recent years, given a significant increase in unconventional production.

We find that since the expansion in shale gas production, the explanatory power of rig count has declined, while the effect of the number of producing wells remained statistically signifi- cant. Therefore, new wells and completion of drilled wells are important deteminants of natural gas output. The decline in the significance of rig counts as a determinant is expected given the nature of UOG production technology, where extraction requires hydraulic fracturing as an additional step, which can be delayed indefinitely. Hence, unless delays in well completion are constant across wells, the explanatory power of rig counts is expected to decline. Indeed, we observe heterogeneity in the delay of well completions and an overall increase in the number of DUCs. As a result, the statistical significance of rig counts has diminished as completion decisions have become important determinants of natural gas output.

Our results show that rig count and futures prices have statistically significant effects on the number of DUCs. Aggregate, as well as region-specific results indicate that an increase in the natu-

128 / The Energy Journal

All rights reserved. Copyright © 2021 by the IAEE.

ral gas futures prices decreases the number of DUCs. This suggests that all else constant, increase in natural gas prices motivates operators to complete existing drilled wells sooner. An increase in the futures price of natural gas decreases the probability that a well remains uncompleted and increases the probability that a well will be completed assuming it has not yet been completed. This result is consistent with producers hedging gas production to take advantage of high initial well productivity. Forward contracts and futures markets with favorable prices enable producers to pay off well com- pletion costs faster and attract needed investment to finance well completion.

The duration model also shows that pipeline capacity has a negative effect on the duration of DUC status. This result confirms the effect of pipeline infrastructure bottlenecks in natural gas markets. While the effect of pipeline bottlenecks on natural gas prices has been recognized (Oliver et al., 2014), we show that pipeline capacity has a direct positive effect on the completion of drilled unconventional wells using data from multiple shale regions. Our results are consistent with the observed negative effects of pipeline constrains on completion rates and associated negative impacts on the demand for sand, water, and fracking fleet capacity as reported in industry outlets (Davis, 2018; Andrien, 2016).

It is important to note that this study does not explicitly address the simultaneity of out- put, inventories, and prices. This study is the first to draw attention to the role of well completion in unconventional gas production as a factor in aggregate output. Our objective is to point to the diminished power of rig counts and the increased role of completion decisions. We refrain from also addressing the identification issues and from claiming causal inference involving aggregate natural gas supply and prices. Future studies should examine supply of natural gas considering endogene- ity of prices and inventory to support a proper causal inference for supply. In our analysis of DUC duration, we include prices as one of the factors affecting the timing of well completion decisions. In the well level analysis, causal inferences pertaining to prices and well level completion decisions are not as susceptible to the inconsistency of estimates that may be caused by price endogeneity. For individual well completion modeling, price can be reasonably treated as an exogenous factor.

The results of this study are important for natural gas operators, energy market analysts, government agencies and other stakeholders in the natural gas industry. Investors, operators, market analysts and policy makers rely on natural gas production information to support investment strat- egies, facilitate production decisions, improve market analysis, and formulate regulatory policies. Thus, it is important to have access to the best available information about the primary determinants of natural gas production. EIA produces a monthly report (Drilling Productivity Report) which uses data on drilling rig counts, drilling productivity and production in natural gas wells to develop re- gional forecasts of natural gas production. In this study, we show that the information about drilled but uncompleted wells can be meaningful for improving such projections.

We also show that infrastructure constraints, like pipeline bottlenecks, can have important implications for well completion decisions and natural gas output in the U.S. The implications of such bottlenecks are important for coordinating increasingly interdependent electricity and natu- ral gas markets (Mugabe et al., 2020) considering reliability (Moeller, 2012; U.S. Department of Energy, 2015). Increased availability of shale gas has transformed the U.S. power sector (Mugabe et al., 2020; Kerr, 2010; Rogers, 2011), and future developments in natural gas distribution infra- structure will likely have further implications for U.S. power generation sector (Logan et al., 2013). Future analysis should examine how the U.S. electricity sector will evolve under various natural gas distribution infrastructure bottleneck scenarios.

All the DUCs in a Row: Natural Gas Production in U.S. / 129

Copyright © 2021 by the IAEE. All rights reserved.

ACKNOWLEDGMENTS

The authors acknowledge financial support from West Virginia Higher Education Policy Commission under grant number HEPC.dsr.18.7

REFERENCES

Abrigo, M.R. and I. Love (2016). “Estimation of panel vector auto regression in Stata.” The Stata Journal 16(3): 778–804. https://doi.org/10.1177/1536867X1601600314.

Addison V. (2018). “DUC, Abnormal DUC Counts Rise in Permian.” Accessed 4/18/2019. https://www.hartenergy.com/ex- clusives/duc-abnormal-duc-counts-rise-permian-31323.

Agerton, M., P.R. Hartley., K.B. Medlock III., and T. Temzelides. (2017). “Employment impacts of upstream oil and gas in- vestment in the United States.” Energy Economics, 62: 171–180. https://doi.org/10.1016/j.eneco.2016.12.012.

Anderson, S.T., R. Kellogg, and S.W. Salant. (2018). “Hotelling under pressure.” Journal of Political Economy 126(3): 984–1026. https://doi.org/10.1086/697203.

Andrews, D.W. (1993). “Tests for parameter instability and structural change with unknown change point.” Economet- rica,61(4): 821–856. https://doi.org/10.2307/2951764.

Andrews, D.W. and W. Ploberger. (1994). “Optimal tests when a nuisance parameter is present only under the alternative”. Econometrica 62: 1383–1414. https://doi.org/10.2307/2951753.

Andrien, T. (2016). “A Guide to American DUCs (Drilled Uncompleted Wells).” Accessed 11/11/2018..https://www.forbes. com/sites/drillinginfo/2016/06/27/american-ducs-drilled-uncompleted-wells/#76ba200d729c.

Apergis, N., B.T. Ewing., and J.E. Payne. (2016). “A time series analysis of oil production, rig count and crude oil price: Evi- dence from six US oil producing regions.” Energy 97: 339–349. https://doi.org/10.1016/j.energy.2015.12.028.

Arora, V. and J. Lieskovsky. (2014). “Natural gas and US economic activity.” The Energy Journal 35(3): 167–182. https://doi. org/10.5547/01956574.35.3.8.

Baker Hughes (2019). “Rig Count FAQs.” Accessed 5/12/2019. http://phx.corporate-ir.net/phoenix.zhtml?c=79687&p=i- rol-rigcountsfaqs.

Box-Steffensmeier, J.M., and C.J. Zorn. (2001). “Duration models and proportional hazards in political science.” American Journal of Political Science 45(4): 972–988. https://doi.org/10.2307/2669335.

Boyce, J.R. and L. Nøstbakken. (2011). “Exploration and development of US oil and gas fields, 1955–2002.” Journal of Eco- nomic Dynamics and Control 35(6): 891–908. https://doi.org/10.1016/j.jedc.2010.12.010.

Chen, F. and S.C. Linn. (2017). “Investment and operating choice: Oil and natural gas futures prices and drilling activity.” Energy Economics 66: 54–68. https://doi.org/10.1016/j.eneco.2017.05.012.

Cox, D.R. (1972). “Regression models and life‐tables.” Journal of the Royal Statistical Society: Series B (Methodological) 34(2): 187–202. https://doi.org/10.1111/j.2517-6161.1972.tb00899.x.

Davis C. (2018). “Permian Well Completions Seen Declining, with Fewer Crews and Less Sand Demand.” Accessed 4/7/2019. https://www.naturalgasintel.com/articles/116193-permian-well-completions-seen-declining-with-fewer-crews- and-less-sand-demand.

Dickey, D.A. and W.A. Fuller. (1981). “Likelihood ratio statistics for autoregressive time series with a unit root.” Economet- rica 49: 1057–1072. https://doi.org/10.2307/1912517.

DI (DrillingInfo). (2016). “Production.” Accessed 09/17/2018. https://app.drillinginfo.com/production/#/default. Dunning, P. (2016). “DUC Hunting: Demystifying Drilled and Uncompleted Wells.” Accessed 4/18/2019. https://www.

forbes.com/sites/drillinginfo/2016/04/25/duc-hunting-demystifying-drilled-and-uncompleted-wells/#7fa1580433d6. EIA (2013). “Rethinking rig count as a predictor of natural gas production. Today in Energy.” Accessed 11/06/2018. https://

www.eia.gov/todayinenergy/detail.php?id=13551. EIA (2016a). “Trends in U.S. Oil and Natural Gas Upstream Costs.” Accessed 3/22/2017. https://www.eia.gov/analysis/stud-

ies/drilling/pdf/upstream.pdf/. EIA (2016b). “Drilling Productivity Report. EIA Estimates of Drilled but Uncompleted Wells (DUCs).” Accessed 2/5/2019.

https://www.eia.gov/petroleum/drilling/pdf/duc_supplement.pdf. EIA (2017). “Drilling Productivity Report.” Accessed 2/20/2017. http://www.eia.gov/petroleum/drilling/pdf/dpr-full.pdf. EIA (2018). “The Distribution of U.S. Oil and Natural Gas Wells by Production Rate.” Accessed 1/14/2019. https://www.eia.

gov/petroleum/wells/pdf/full_report.pdf. EIA (2019a). “Drilling Productivity Report.” Accessed 11/01/2019. Available online at: https://www.eia.gov/petroleum/drill-

ing/.

130 / The Energy Journal

All rights reserved. Copyright © 2021 by the IAEE.

EIA (2019b). “The number of drilled but uncompleted wells in the United States continues to climb, Today in Energy.” Ac- cessed 10/23/2019. https://www.eia.gov/todayinenergy/detail.php?id=39332.

Engle, R.F. and C.W. Granger. (1987). “Co-integration and error correction: representation, estimation, and testing.” Econo- metrica 55(2): 251–276. https://doi.org/10.2307/1913236.

Federal Reserve Bank of Dallas (2019). “Permian Basin Economic Indicators.” Accessed 5/07/2019. https://www.dallasfed. org/research/indicators/pb/2018/pb1807.

Fleming, T.R. and D.P. Harrington. (2011). Counting processes and survival analysis (Vol. 169). John Wiley and Sons. Gülen, G., J. Browning., S. Ikonnikova., and S.W. Tinker. (2013). “Well economics across ten tiers in low and high Btu (Brit-

ish thermal unit) areas, Barnett Shale, Texas.” Energy 60: 302–315. https://doi.org/10.1016/j.energy.2013.07.041. Hair, J.F., Jr., R.E. Anderson., R.L. Tatham., and W.C. Black. (1995). Multivariate Data Analysis, 3rd ed, Macmillan Publish-

ing Company, New York. Hamilton, J.D. (2013). Oil prices, exhaustible resources, and economic growth (No. w17759). Edward Elgar Publishing.

https://doi.org/10.3386/w17759 . Hannan, E.J. and B.G. Quinn. (1979). “The determination of the order of an auto regression.” Journal of the Royal Statistical

Society. Series B (Methodological) 41(2): 190–195. https://doi.org/10.1111/j.2517-6161.1979.tb01072.x. Hegarty J. (2017). “The Surge in DUC Wells Begs the Question: How Long Can a DUC Well Hold a Lease?” Accessed

4/18/2019. https://www.wsmtlaw.com/blog/the-surge-in-duc-wells-begs-the-question-how-long-can-a-duc-well-hold-a- lease.html.

Hernandez, A., and J. Dresdner. (2010). “The effect of temporal closures and individual quotas on fishing trip duration: a haz- ard function analysis.” Applied Economics 42(29): 3767–3776. https://doi.org/10.1080/00036840802360096.

IHS (2016). “Oil and Gas Operators with Significant Inventory of Drilled but Uncompleted Wells in Major U.S. Plays to Benefit from Capital Efficiency Gains in 2016.” Accessed 4/18/2019.https://news.ihsmarkit.com/press-release/ducs/ oil-and-gas-operators-significant-inventory-drilled-uncompleted-wells-major-us-pl.

Ikonnikova, S. and G. Gülen. (2015). “Impact of Low Prices on Shale Gas Production Strategies.” The Energy Journal 36. Adelman Special Issue. https://doi.org/10.5547/01956574.36.SI1.siko.

Iledare, O.O. (1995). “Simulating the effect of economic and policy incentives on natural gas drilling and gross reserve addi- tions.” Resource and Energy Economics 17(3): 261–279. https://doi.org/10.1016/0928-7655(95)00003-G.

Johansen, S. (1988). “Statistical analysis of cointegration vectors.” Journal of Economic Dynamics and Control 12(2–3): 231–254. https://doi.org/10.1016/0165-1889(88)90041-3.

Johansen, S. (1995a). “A statistical analysis of cointegration for I (2) variables.” Econometric Theory 11(1): 25–59. https:// doi.org/10.1017/S0266466600009026.

Johansen, S. (1995b). Likelihood-based inference in cointegrated vector autoregressive models. Oxford University Press on Demand. https://doi.org/10.1093/0198774508.001.0001.

Joskow, P.L. (2013). “Natural gas: from shortages to abundance in the United States.” American Economic Review 103(3): 338–43. https://doi.org/10.1257/aer.103.3.338.

Jiang, H. and C. Liu. (2014). “A panel vector error correction approach to forecasting demand in regional construction mar- kets.” Construction Management and Economics 32(12): 1205–1221. https://doi.org/10.1080/01446193.2014.977800.

Kerr, R.A. (2010). “Natural gas from shale bursts onto the scene.” Science 328(5986): 1624–1626. https://doi.org/10.1126/ science.328.5986.1624.

Khalifa, A., M. Caporin., and S. Hammoudeh. (2017). “The relationship between oil prices and rig counts: The importance of lags.” Energy Economics 63: 213–226. https://doi.org/10.1016/j.eneco.2017.01.015.

Kleinberg, R.L., S. Paltsev., C.K.E. Ebinger., D.A. Hobbs, and T. Boersma. (2018). “Tight oil market dynamics: Benchmarks, breakeven points, and in-elasticities.” Energy Economics 70: 70–83. https://doi.org/10.1016/j.eneco.2017.11.018.

Levin, A., C.F. Lin., and C. Chu. (2002). “Unit root tests in panel data: Asymptotic and finite sample properties.” Journal of Econometrics 108(1): 1–24. https://doi.org/10.1016/S0304-4076(01)00098-7.

Liew, V. (2004). “What lag selection criteria should we employ?” Economics Bulletin 33(3): 1–9. Logan, J., A. Lopez., T. Mai., C. Davidson., M. Bazilian., and D. Arent. (2013). “Natural gas scenarios in the US power sec-

tor.” Energy Economics, 40: 183–195. https://doi.org/10.1016/j.eneco.2013.06.008. Mason, C.F. and G. Roberts. (2018). “Price Elasticity of Supply and Productivity: An Analysis of Natural Gas Wells in Wyo-

ming.” The Energy Journal 39(S1): 79–101. https://doi.org/10.5547/01956574.39.SI1.cmas. Melick, W. (2014). “The energy boom and manufacturing in the United States.” Federal Reserve Board (FRB) International

Finance Discussion Paper 1108. Accessed 12/17/2018. https://www.federalreserve.gov/PUBS/ifdp/2014/1108/ifdp1108. pdf. https://doi.org/10.17016/IFDP.2014.1108.

Melek, N.C. (2015). “What could lower prices mean for US oil production?” Federal Reserve Bank of Kansas City. Economic Review 100(1): 51–69.

All the DUCs in a Row: Natural Gas Production in U.S. / 131

Copyright © 2021 by the IAEE. All rights reserved.

Moeller, P.D. (2012). “Request for Comments of Commissioner Moeller on Coordination between the Natural Gas and Elec- tricity Markets, Federal Energy Regulatory Commission.” Accessed 10/19/2019. https://www.ferc.gov/industries/electric/ indus-act/electric-coord/moellergaselectricletter.pdf.

Mugabe, D., L. Elbakidze, and G. Zaynutdinova (2020). “Elasticity of substitution and technical efficiency: evidence from the US electricity generation.” Applied Economics 52(16): 1789–1805. https://doi.org/10.1080/00036846.2019.1678733.

Neal, T. (2014). “Panel cointegration analysis with xtpedroni.” The Stata Journal 14(3): 684–692. https://doi. org/10.1177/1536867X1401400312.

Oliver, M.E., C.F. Mason., and D. Finnoff (2014). “Pipeline congestion and basis differentials.” Journal of Regulatory Eco- nomics, 46(3): 261–291. https://doi.org/10.1007/s11149-014-9256-9.

Oglend, A., M.E. Lindbäck., and P. Osmundsen. (2016). “Shale gas boom affecting the relationship between LPG and oil prices.” The Energy Journal 37(1): 211–233. https://doi.org/10.5547/01956574.37.1.aogl.

Osmundsen, P., K.E. Rosendahl., and T. Skjerpen. (2015). “Understanding rig rate formation in the Gulf of Mexico.” Energy Economics 49: 430–439. https://doi.org/10.1016/j.eneco.2015.03.001.

Pedroni, P. (1999). “Critical values for cointegration tests in heterogeneous panels with multiple regressors.” Oxford Bulletin of Economics and Statistics 61(S1): 653–670. https://doi.org/10.1111/1468-0084.61.s1.14.

Pedroni, P. (2004). “Panel cointegration: asymptotic and finite sample properties of pooled time series test with an application to the PPP hypothesis.” Econometric Theory 20(3): 597–625. https://doi.org/10.1017/S0266466604203073.

Peters, J.C., and T.W. Hertel. (2017). “Achieving the Clean Power Plan 2030 CO 2 Target with the New Normal in Natural Gas Prices.” Energy Journal 38(5): 39–66. https://doi.org/10.5547/01956574.38.5.jpet.

Phillips, P.C. and P. Perron. (1988). “Testing for a unit root in time series regression.” Biometrika 75(2): 335–346. https://doi. org/10.1093/biomet/75.2.335.

Piotrowski, M. (2016). “The DUCs Are Fracking: The Uncertainty Surrounding Drilled but Uncompleted Wells.” Accessed 4/18/2019. http://energyfuse.org/ducs-fracking-uncertainty-surrounding-drilled-uncompleted-wells/.

Ringlund, G.B., K.E. Rosendahl., and T. Skjerpen. (2008). “Does oilrig activity react to oil price changes? An empirical inves- tigation.” Energy Economics 30(2): 371–396. https://doi.org/10.1016/j.eneco.2007.06.002.

Rogers, H. (2011). “Shale gas—the unfolding story.” Oxford Review of Economic Policy 27(1): 117–143. https://doi. org/10.1093/oxrep/grr004.

Stogiannis, D., C. Caroni., C.E. Anagnostopoulos., and I.K. Toumpoulis. (2011). “Comparing first hitting time and propor- tional hazards regression models.” Journal of Applied Statistics 38(7): 1483–1492. https://doi.org/10.1080/02664763.201 0.505954.

Srinivasan, K., J. Krishnamurthy., and P. Kaufman. (2019). “Concerns and Clarifications for Drilled Uncompleted (DUC) Wells in the Williston Basin.” Society of Petroleum Engineers Reservoir Evaluation and Engineering 22(1): 190–202. https://doi.org/10.2118/185753-PA.

Surran, C. (2019). “Permian Basin natural gas prices at negative all-time lows, Seeking Alpha.” Accessed 5/08/2019. https:// seekingalpha.com/news/3448082-permian-basin-natural-gas-prices-negative-time-lows,

Sy, J.P., and J.M. Taylor. (2000). “Estimation in a Cox proportional hazards cure model.” Biometrics 56(1): 227–236. https:// doi.org/10.1111/j.0006-341X.2000.00227.x.

Teachman, J.D. and M.D. Hayward. (1993). “Interpreting hazard rate models.” Sociological Methods and Research 21(3): 340–371. https://doi.org/10.1177/0049124193021003003.

U.S. Department of Energy. (2015). “Natural Gas Infrastructure Implications of Increased Demand from the Electric Power Sector.” Accessed 10/19/2019. https://www.energy.gov/sites/prod/files/2015/02/f19/DOE%20Report%20Natural%20 Gas%20Infrastructure%20V_02-02.pdf.

Vogelsang, T.J. (1997). “Wald-type tests for detecting breaks in the trend function of a dynamic time series.” Econometric Theory 13(6): 818–848. https://doi.org/10.1017/S0266466600006289.

Weber, J.G. (2012). “The effects of a natural gas boom on employment and income in Colorado, Texas, and Wyoming.” En- ergy Economics 34(5): 1580–1588. https://doi.org/10.1016/j.eneco.2011.11.013.

Natural Gas 3.pdf

Disclaimer: This is a machine generated PDF of selected content from our products. This functionality is provided solely for your convenience and is in no way intended to replace original scanned PDF. Neither Cengage Learning nor its licensors make any representations or warranties with respect to the machine generated PDF. The PDF is automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. CENGAGE LEARNING AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGEMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the machine generated PDF is subject to all use restrictions contained in The Cengage Learning Subscription and License Agreement and/or the Gale Academic OneFile Terms and Conditions and by using the machine generated PDF functionality you agree to forgo any and all claims against Cengage Learning or its licensors for your use of the machine generated PDF functionality and any output derived therefrom.

New Findings from University of Tabriz Update Understanding of Renewable Energy (Risk-based Optimal Operation of Coordinated Natural Gas and Reconfigurable Electrical Networks With Integrated Energy Hubs). Date: June 5, 2021 From: Obesity, Fitness & Wellness Week Publisher: NewsRX LLC Document Type: Report Length: 446 words Lexile Measure: 1410L

Full Text: 2021 JUN 5 (NewsRx) -- By a News Reporter-Staff News Editor at Obesity, Fitness & Wellness Week -- Investigators publish new report on Energy - Renewable Energy. According to news reporting originating in Tabriz, Iran, by NewsRx journalists, research stated, "This paper elaborates on optimal scheduling of coordinated power and natural gas (NG) networks in the presence of interconnected energy hubs considering reconfiguration as a flexibility source. With regard to the energy hub system consisting of several generation units, storage and conversion technologies, as well as natural gas-fired units, the high interdependency between gas and electricity carriers should be captured."

Financial supporters for this research include University of Tabriz, Iran National Science Foundation (INSF).

The news reporters obtained a quote from the research from the University of Tabriz, "The hourly reconfiguration capability is developed for the first time in a multi-energy system to enhance the optimal power dispatch and gas consumption pattern. The realistic interdependency of electrical and NG grids is investigated by employing the steady-state Weymouth equation and AC-power flow model for power and gas networks, respectively. Furthermore, to handle the risk associated with strong uncertainty of wind power, load, and real-time power price, the conditional value at risk approach is employed. The proposed model is implemented on the integrated test system and simulation results are presented for different cases. The impact of the risk aversion level on operating cost and optimal scheduling of controllable units is examined."

According to the news reporters, the research concluded: "Numerical results demonstrate that reconfigurable capability reduces the operational cost up to 7.82%."

This research has been peer-reviewed.

For more information on this research see: Risk-based Optimal Operation of Coordinated Natural Gas and Reconfigurable Electrical Networks With Integrated Energy Hubs. IET Renewable Power Generation, 2021. IET Renewable Power Generation can be contacted at: Inst Engineering Technology-iet, Michael Faraday House Six Hills Way Stevenage, Hertford SG1 2AY, England. (The Institution of Engineering and Technology - www.theiet.org/; IET Renewable Power Generation - www.ietdl.org/IET-RPG)

Our news correspondents report that additional information may be obtained by contacting Mohammad Hemmati, University of Tabriz, Faculty of Electrical and Computer Engineering, Tabriz, Iran. Additional authors for this research include Mehdi Abapour, Behnam Mohammadi-Ivatloo and Amjad Anvari-Moghaddam.

Keywords for this news article include: Tabriz, Iran, Renewable Energy, Energy, Natural Gas, Oil & Gas, University of Tabriz.

Our reports deliver fact-based news of research and discoveries from around the world. Copyright 2021, NewsRx LLC

The citation for this news report is: NewsRx. New Findings from University of Tabriz Update Understanding of Renewable Energy (Risk-based Optimal Operation of Coordinated Natural Gas and Reconfigurable Electrical Networks With Integrated Energy Hubs). Obesity, Fitness & Wellness Week. June 5, 2021; p 249.

Copyright: COPYRIGHT 2021 NewsRX LLC http://www.newsrx.com/newsletters/Obesity,-Fitness-and-Wellness-Week.html

Source Citation (MLA 8th Edition) "New Findings from University of Tabriz Update Understanding of Renewable Energy (Risk-based Optimal Operation of Coordinated

Natural Gas and Reconfigurable Electrical Networks With Integrated Energy Hubs)." Obesity, Fitness & Wellness Week, 2021, p. 249. Gale Academic OneFile, link.gale.com/apps/doc/A663572767/AONE?u=lincclin_mdcc&sid=bookmark- AONE&xid=79306bd0. Accessed 22 June 2021.

Gale Document Number: GALE|A663572767

Novel_Power_Smoothing_Techniqu.PDF

Advances in Electrical and Computer Engineering Volume 21, Number 2, 2021

Novel Power Smoothing Technique for a Hybrid AC-DC Microgrid Operating with Multiple

Alternative Energy Sources

Pramod Bhat NEMPU1, Jayalakshmi Narayana SABHAHIT1, Dattatreya Narayan GAONKAR2, Vidya Sudarshan RAO3

1Department of Electrical and Electronics Engineering, Manipal Institute of Technology, Manipal Academy of Higher Education, Manipal, India - 576104

2Department of Electrical and Electronics Engineering, National Institute of Technology, Surathkal, Karnataka, India - 575025

3Department of Instrumentation and Control Engineering, Manipal Institute of Technology, Manipal Academy of Higher Education, Manipal, India – 576104

[email protected]

Abstract—The power produced by renewable sources such

as photovoltaic systems and wind energy conversion systems is highly intermittent due to continuously changing irradiance and wind velocity. When the distributed generation systems employing photovoltaic (PV) array and wind energy conversion system (WECS) operate in grid-tied mode, the power fluctuations affect the power quality of the grid. In a hybrid AC-DC microgrid (HMG), the dynamics of DC and AC subgrids influence each other. This paper proposes a supercapacitor based novel power smoothing methodology for the HMG with PV array, WECS, fuel cell (FC) and electrolyzer (EL) based hydrogen storage system considering the power fluctuations in both subgrids. The power smoothing technique on the DC subgrid aims to facilitate instantaneous power balance. The Kalman filter (KF) based velocity smoothing (KFV) approach is developed for the WECS. The KFV technique is compared with the power smoothing techniques presented in the literature. The KFV method is found to be effective in computing the smooth power reference for the supercapacitor system. By incorporating the proposed power smoothing technique in the HMG, the stress on the interlinking converter (ILC) and utility grid are minimized and the power quality is enhanced.

Index Terms—Kalman filters, microgrids, power smoothing, renewable energy sources, supercapacitors.

I. INTRODUCTION

The concept of microgrids employing non-conventional energy sources has drawn the attention of researchers recently. Sunlight and wind are freely available in nature and are harmless to the environment. The microgrid (MG) operates in grid-integrated mode and stand-alone mode. The HMG configuration is reliable compared to AC MGs and DC MGs. A HMG comprises both AC and DC subgrids [1, 2]. Power management within the subgrids, coordination among the subgrids and power quality issues are the challenges in a grid-tied HMG [3].

The permanent magnet synchronous generator (PMSG) based WECS is efficient due to the absence of gearbox [4]. An irradiance averaging technique for power smoothing of a PV system with BES is proposed in [5]. However, power smoothing is not clearly evaluated in the paper. The experimental investigation of the SC-based smoothing

controller for the renewable energy-based DGs integrated with the low voltage grid is proposed in [6]. By monitoring the state of charge (SOC) of SC, appropriate smooth reference is provided for the SC system and the sliding mode controller is incorporated.

In [7], the power smoothing is investigated for a PV-BES hybrid system with different methods and SOC regulation is also accomplished. Various control approaches for power smoothing of WECS are reviewed in [8]. Energy storage based smoothing is expensive but efficient. A power smoothing controller for a WECS is developed with the combination of MPPT control and frequency deviation control loops. The gain of the supplementary control loop is adjusted for better smoothing performance [9]. This control scheme doesn’t employ a storage system.

A novel power smoothing controller is proposed in [10] for a PV-WECS hybrid system using the battery. Efficient smoothing and SOC regulation of battery are achieved. An adaptive low pass filter (LPF) based power smoothing technique designed with a lithium-ion capacitor for PMSG based WECS in [11]. The comparative study of LPF and moving average methods using superconducting magnetic energy storage is presented in [12]. The LPF is proved to be effective. In [13], the authors proposed the power smoothing of WECS by wavelet transform. The authors inferred that wavelet transform based smoothing is efficient compared to the smoothing achieved with inertial filter.

The SC bank is incorporated for smoothing the output of the PV-WECS hybrid system. The fluctuating output of the PV array and WECS is processed by a rate limiter and then the average is computed to obtain smooth power reference in [14]. LPF is employed to achieve power smoothing in a PV system operating with the weak grid in [15]. The results presented show a reduction in total harmonic distortion (THD).

The analysis of WECS with SC during faults is described in [16]. During normal operation, the SC bank helps to mitigate the fluctuations in the power output of WECS. During the faults, SC absorbs power and helps to ride through the fault. The KF based weighted average approach is described in [17-18]. Various techniques for power

99 1582-7445 © 2021 AECE

Digital Object Identifier 10.4316/AECE.2021.02011

[Downloaded from www.aece.ro on Thursday, June 10, 2021 at 14:03:12 (UTC) by 97.69.214.164. Redistribution subject to AECE license or copyright.]

Advances in Electrical and Computer Engineering Volume 21, Number 2, 2021

smoothing of the PV system and WECS are reviewed in [19]. KF based weighted average approach is proved to be effective.

The research papers addressing power smoothing techniques in the literature have not considered the MGs with DC and AC subgrids. The effect of intermittent power generation from renewable sources on the operation of individual subgrids in HMG architecture is not addressed.

In this work, the PV system is the principal source in the DC subgrid and FC is the auxiliary source. The WECS is the only renewable source in the AC subgrid. Hence, the requirements for developing the power smoothing control schemes on two subgrids are different.

Therefore, this paper proposes the Kalman filter based control method for power smoothing of the WECS. For the DC subgrid, a simple power smoothing technique is developed so as to facilitate instantaneous power management by the SC as the FC and EL systems are controlled to achieve power balance. The research contributions of this article are as follows.  KF based novel power smoothing technique is

proposed for the WECS.  The impact of intermittent power production from the

PV system and WECS on individual subgrids in a HMG is analyzed and an appropriate power smoothing method is investigated to mitigate power fluctuations.

This paper is organized as follows. In section II, the

configuration of the HMG is presented. In section III, different control schemes are described. The simulation results are analyzed in section IV. Conclusions are presented in section V.

II. CONFIGURATION OF HMG

The schematic illustration of the HMG is presented in Fig. 1. The main energy source on the DC subgrid is a PV array capable of producing 21 kW at standard conditions operating in MPPT mode. A 20 kW PMSG based WECS [14] is coupled to the AC bus, which is integrated with the utility grid. FC system of capacity 10 kW is chosen as the backup source in DC subgrid. EL system of rating 10 kW is used for energy storage. PV array is modelled based on the model described in [20]. Models of the FC stack and electrolyzer system are realized as described in [21-22]. The WECS consists of a rectifier and a boost converter. The output of the boost converter is inverted and delivered to the AC bus. AC bus is integrated with the grid through a 100 kVA transformer (415V/3.3kV). AC and DC subgrids are coupled by a bidirectional ILC and a filter [23]. The SC banks on both subgrids (SC1 and SC2) contain a series combination of 50 modules of rating 58 F and 16 V. The SC banks are controlled by bidirectional DC-DC converters.

Figure 1. Schematic illustration of the HMG

100

[Downloaded from www.aece.ro on Thursday, June 10, 2021 at 14:03:12 (UTC) by 97.69.214.164. Redistribution subject to AECE license or copyright.]

Advances in Electrical and Computer Engineering Volume 21, Number 2, 2021

III. CONTROL SCHEMES

A. Control Schemes for the Power Balance of DC Subgrid

The control schemes of FC and EL systems realized for the power management in the DC bus are illustrated in Fig. 2. In the control strategy of the FC system, the difference between the DC load current (IDC LOAD) and the current sensed at the output terminals of the PV system’s converter (IPV) is the reference signal. The current sensed from the output of the converter of the FC system (IFC) is the feedback.

The difference between the power output of the PV array (PPV) and DC load demand (PDC LOAD) is the reference signal for the controller of the EL. The control strategies of both FC and EL systems employ the PI regulator. The FC system delivers power when the load demand on the DC bus goes beyond the output of the PV array. Surplus power generated is absorbed by the EL system (PEL). Based on the current through the EL, hydrogen is produced, which is stored in a storage tank.

PI PWMIDC LOAD

IPV

+ -

IFC

+ -

PI PWMPPV + -

+ -

PDC LOAD PEL

To boost converter of FC system

To buck converter of EL system

Figure 2. Control schemes of FC and EL systems

B. The Control Scheme of ILC

The DC subgrid is integrated with the AC subgrid through an ILC. The PQ control scheme [14] maintains the DC bus voltage (Vdc) at the desired value (800 V) and enables a proper exchange of power amongst the subgrids. The PQ controller of the ILC is shown in Fig. 3. The inverter of WECS is also regulated by a similar PQ control scheme.

Vdc (ref)

Vdc Outer regulator

(PI)

abc to dq

Inner regulator

(PI)

id(ref)

id, iq

iq(ref)

iabc

to ILC

dq to abc

PWM

VabcPLL Figure 3. The PQ control scheme of ILC

The power in a three-phase network is given by equation (1):

(1) ccbbaa ivivivtP )( Active and reactive power (P and Q) are computed based

on d and q axis voltages vd and vq using the equations (2) and (3) respectively:

 qqdd viviP  2

3 (2)

 dqqd viviQ  2

3 (3)

When the reference frame and grid voltage are synchronized, P and Q can be expressed as per equations (4) and (5) respectively.

)( 2

3 dd viP  (4)

)( 2

3 dq viQ  (5)

where, id and iq are the currents corresponding to the d and q axis. The outer loop of the control scheme maintains the DC bus voltage and the inner current control loop regulates id and iq. The iq (ref) is made zero to ensure the operation at UPF. The phase-locked loop is used to measure the phase angle of grid voltage to synchronize ILC with the grid.

The voltage balance across the filter [24] is given by equation (6):

 

  

 

  

  

  

  

  

 

  

 

  

  

q

d

q

d f

q

d f

q

d f

q

d

v

v

i

i L

i

i

dt

d L

i

i R

v

v

0

0 1

1

  (6)

where, Lf and Rf represent the total inductance and resistance of filter, respectively and ω is the angular frequency.

C. Power Smoothing in the DC Subgrid

The power smoothing controller on the DC subgrid is designed to provide or absorb power during sudden power variations in the DC bus. This control technique is depicted in Fig. 4. By incorporating the smoothing controller on the DC bus, power stress on the ILC can be minimized. When the power balance is disturbed by fluctuating load or generation, the controller decides the power to be absorbed or supplied by the SC1. A selection switch is introduced as depicted in Fig. 4 so as to transfer the excess power to the utility grid through the ILC whenever the surplus power produced in the PV system exceeds 10 kW. Without the selection switch, power would be absorbed by the SC bank.

+

-

SC1

PWM

VSC1

PI+-

ISC1

Vdc

PPV + PFC - PEL +-

PDC LOAD

Selection s witch

(PPV - PDC LOAD) < 10 kW

0

ISC1

VSC1

Figure 4. The control scheme for power smoothing in DC bus

101

[Downloaded from www.aece.ro on Thursday, June 10, 2021 at 14:03:12 (UTC) by 97.69.214.164. Redistribution subject to AECE license or copyright.]

Advances in Electrical and Computer Engineering Volume 21, Number 2, 2021

D. Proposed Power Smoothing Method for the WECS

The power output of WECS is highly fluctuating due to continuously changing wind velocity. Since the WECS is coupled to the AC bus, it affects the operation of the DC subgrid and utility grid. If PW is the actual power of WECS, Psmooth is the smooth power reference and VSC2 is the voltage of SC2, then the current reference to the SC bank is computed by equation (7):

2

)(2 SC

Wsmooth refSC

V

PP I

  (7)

A PI regulator is used to compute the duty ratio to the BDC based on the current output required from the SC system. The smoothing control technique developed for the WECS is depicted in Fig. 5.

+

-

SC2

PWM PI + - Power

computation 

VSC2ISC2

+ -

Pw

Psmooth

KF

vn un

vsmooth

Figure 5. The control scheme for power smoothing of WECS by KFV approach

The KFV smoothing technique is developed to compute

smooth power reference for power smoothing of the WECS. In the proposed approach, the intermittent wind velocity signal is processed by the KF to compute the smooth power reference. The KF is mainly used to estimate the state of the system when it contains random noises. This approach is inspired by the application of KF described in [25]. If x represents the state, z represents the output, w and m represent noises, the system equations are given by equations (8) and (9):

(8) nnnn wBuAxx 1 nnn mHxz  (9)

In this method, the wind velocity (vn) and rate of change of vn (an) are chosen as the state variables as represented in equation (10):

(10)  

  

 

n

n n

a

v x

If the rate of change of an is the input un to the system and T is the time for which the values are computed (2 µs), then the matrices are given by equation (11) and (12):

(11) nnn u T

x T

x  

  

 

  

 

0

10

1 1

(12)   nn xz 01 The prediction equations of the KF can be expressed by

equations (13) and (14):

(13) 11ˆˆ    nnn BuxAx

(14) k T

nn QAAPP   

1

The correction equations [25-26] of the KF can be expressed by equations (15), (16) and (17):

(15) 1)(   k T

n T

nn RHHPHPK

(16) )ˆ(ˆˆ   nnnnn xHzKxx

(17)  nnn PHKIP )(

where, and are the priori and posterior estimate error

covariance, respectively. Kn is the Kalman gain; Rk and Qk are measurement noise covariance and process noise covariance, respectively.

 nP nP

Based on the smoothened velocity (vsmooth), the smooth power reference for the WECS (Psmooth) is computed based on equation (18):

 psmoothsmooth CAvP 3

2

1  (18)

where,  is the density of air (1.225 kg/m3), A is the swept area of the rotor blades, Cp is the coefficient of performance and represents the average efficiency of generator and power conversion system (0.85). The maximum value of CP is 0.47 [14]. The values of Rk and Qk are carefully selected to ensure smooth power output without resulting in a larger difference between the smooth power and actual power. In

this method, Qk = and Rk =10 3.

  

  

 

12

12

100

010

The proposed KFV approach is compared with the rate- limiter-mean (RL-mean) method, LPF method and Kalman filter based weighted average (KFWA) method.

In the RL-mean approach, the fluctuating power output of WECS is passed through the rate limiter and then the average is computed using the ‘mean’ block of Simulink library [14].

In the LPF based method, smooth power is computed using a low pass filter [12]. The time period corresponding to LPF is chosen as 0.6 s.

In the KF based weighted average approach, the fluctuating output of WECS is used to compute maximum and minimum power for a certain time interval based on which, the weighted average is computed. The KF uses weighted average to calculate the desired power for smoothing [17-18].

IV. RESULTS AND DISCUSSION

The practical data of irradiance and wind velocity for 10 minutes presented in [14] is chosen for analysis. The simulation is performed in MATLAB/Simulink software by scaling down the time to 6 s. The irradiation and wind velocity are illustrated in Fig. 6 and Fig. 7, respectively.

Figure 6. Irradiance

102

[Downloaded from www.aece.ro on Thursday, June 10, 2021 at 14:03:12 (UTC) by 97.69.214.164. Redistribution subject to AECE license or copyright.]

Advances in Electrical and Computer Engineering Volume 21, Number 2, 2021

Figure 7. Wind velocity

A. Analysis of the HMG without Smoothing Controllers

The power balancing of the DC subgrid is achieved by EL and FC systems. Instantaneous power balancing on the DC subgrid is achieved by delivering or absorbing power through ILC (PEX). FC system provides the power (PFC) whenever there is a deficit in generation from PV array (PPV). EL system absorbs power (PEL) when the PV array produces excess power. When the power mismatch surpasses the capacity of FC or electrolyzer, the power balance is achieved with the help of AC subgrid. When the smoothing controller is not included in the system, PEX is characterized by fluctuations that affect the power balance in the DC subgrid, as evident from Fig. 8.

Figure 8. Power management of DC subgrid without smoothing controllers

The power management in the AC subgrid is depicted in

Fig. 9. It can be observed that grid power is highly intermittent in the absence of smoothing controllers.

Figure 9. Power management of AC subgrid without smoothing controllers

B. Comparison of Smoothing Techniques for the WECS

The smooth power obtained by RL-mean, LPF, KFWA and proposed KFV technique are illustrated in Fig. 10. The maximum and minimum values of the power (Pmax and Pmin) are computed for certain duration and the smooth power variation rate (SPVR) [19] is computed as per equation (19).

The SPVR for different methods is presented in Fig. 11. The maximum and minimum SPVR for each method are presented in Table I. The smaller value of SPVR implies better power smoothing.

ratedP

PP SPVR

minmax   (19)

Figure 10. Fluctuating wind power and smooth powers

Figure 11. SPVR of fluctuating power and smooth power

TABLE I. SPVR FOR DIFFERENT METHODS

Method SPVR (max) SPVR (min) LPF 0.080 0.009

RL-mean 0.021 0.001 KFWA 0.032 0.001 KFV 0.021 0.001

The moving average of the error between smooth power

and fluctuating (actual) wind power is observed for each method as depicted in Fig. 12. It can be seen that the KF based methods provide a better estimate of smooth power as the moving average is close to zero. KFV approach is superior to other approaches.

Figure 12. Moving average of the error

The SOC of SC2 with different smoothing techniques is

illustrated in Fig. 13. It can be observed that when the LPF and RL-mean approaches are incorporated, SC bank operates more in the charging mode. However, when the KF

103

[Downloaded from www.aece.ro on Thursday, June 10, 2021 at 14:03:12 (UTC) by 97.69.214.164. Redistribution subject to AECE license or copyright.]

Advances in Electrical and Computer Engineering Volume 21, Number 2, 2021

based approaches are incorporated, SC bank operates in both charge and discharge modes more evenly. This implies that the SC bank uniformly absorbs and delivers power to smoothen the power in KF based methods compared to other techniques. The proposed KFV method is found to be superior with respect to SOC.

Figure 13. SOC of SC2 with different methods

The THD of the grid current obtained with different

smoothing techniques is presented in Table II. THD is computed for ten cycles for the same interval of time for different cases. The THD is found to be low when the KFV method is incorporated for the SC bank in the WECS.

TABLE II. THD OBTAINED WITH DIFFERENT METHODS

Method THD LPF 4.1%

RL-mean 4.62% KFWA 3.84% KFV 3.68%

When various parameters such as SPVR, moving average of the error, SOC and THD are analyzed, the proposed KFV method is found to be superior. Hence it is chosen for smoothing the output of WECS.

C. Analysis of the Impact of Smoothing Controllers

The Vdc in the presence and absence of the smoothing controllers on both subgrids is illustrated in Fig. 14. The voltage deviations in the DC bus are considerably reduced with smoothing controllers.

Figure 14. DC bus voltage with and without smoothing

The active power exchanged with the grid is shown in

Fig. 15. Fluctuations in active power are reduced by incorporating the smoothing controller for WECS. Better smoothing is achieved when smoothing controllers are incorporated for both WECS and DC subgrid.

Figure 15. Active power exchanged with the grid

The reactive power exchanged with the grid is maintained

zero as depicted in Fig. 16.

Figure 16. Reactive power

The power management in DC subgrid in the presence of

smoothing controllers on both the subgrids is depicted in Fig. 17. The smoothing controller of the DC subgrid helps in the reduction of stress on the ILC.

Figure 17. The power balance in DC subgrid with smoothing controllers

The power exchange via the ILC in the presence and

absence of smoothing controllers is depicted in Fig. 18. Since immediate power balancing is achieved by SC1 and fluctuations in the output of WECS are reduced by SC2, the stress on ILC is significantly minimized.

104

[Downloaded from www.aece.ro on Thursday, June 10, 2021 at 14:03:12 (UTC) by 97.69.214.164. Redistribution subject to AECE license or copyright.]

Advances in Electrical and Computer Engineering Volume 21, Number 2, 2021

Figure 18. The power exchanged through ILC

The power balance equations on DC and AC subgrids in

the absence of SC-based smoothing controllers are given by equations (20) and (21):

(20)

(21)

LOADDCEXELFCPV PPPPP 

LOADACEXGRIDW PPPP 

The power balance equations on the DC and AC subgrids in the presence of smoothing controllers are given by equations (22) and (23):

LOADDCEXELSCFCPV PPPPPP  (22)

(23) LOADACEXGRIDsmooth PPPP 

If the power flowing through ILC (PEX) is positive, the power is flowing from DC subgrid to AC subgrid and if it is negative, the power is drawn from AC subgrid. Similarly, if the grid active power (PGRID) is positive, then the power is sent to the grid and if it is negative, the grid supplies the power.

The sinusoidal waveform of the grid current when the smoothing controllers are incorporated in the system is shown in Fig. 19. The current is less distorted. The harmonic spectrum of the grid current in the presence of the smoothing controller for SC1 and KFV based smoothing controller for SC2 is shown in Fig. 20. The THD is within acceptable limits (lesser than 5%).

Figure 19. Grid current

Figure 20. Harmonic spectrum of grid current

V. CONCLUSION

This paper presents a novel power smoothing technique for the grid integrated HMG with PV, WECS and FC based DG system. Distinct power smoothing control schemes are employed on AC and DC subgrids. The SC system is controlled to absorb and deliver power during sudden power deviations on DC subgrid. The output power of WECS coupled to the AC subgrid is smoothened using the SC system by different smoothing techniques. The key findings of this study are summarized below.  When PV array or WECS are integrated on both AC

and DC subgrids, intermittent power production affects the power and voltage profiles of both subgrids.

 KF based velocity smoothing approach is found to be effective in smoothing the power output for WECS.

 By incorporating the smoothing controller developed for the DC subgrid along with the smoothing controller for WECS, stress on the ILC is significantly minimized.

 The smoothing controllers developed in this work have reduced the stress on the individual subgrids and the utility grid; thereby, the power quality is enhanced.

LIST OF ABBREVIATIONS BES Battery Energy Storage EL Electrolyzer FC Fuel Cell HMG Hybrid AC-DC Microgrid ILC Interlinking Converter KF Kalman Filter KFV Kalman Filter Based Velocity

Smoothing KFWA Kalman Filter Based Weighted Average LPF Low Pass Filter MG Microgrid MPPT Maximum Power Point Tracking PMSG Permanent Magnet Synchronous Generator PV Photovoltaic SC Supercapacitor SOC State of Charge SPVR Smooth Power Variation Rate THD Total Harmonic Distortion WECS Wind Energy Conversion System

REFERENCES [1] P. Wang, X. Liu, C. Jin, P. Loh, and F. Choo, “A hybrid AC/DC

micro-grid architecture, operation and control,” in IEEE Power and Energy Society General Meeting, Jul. 2011, pp. 1–8. doi:10.1109/PES.2011.6039453

105

[Downloaded from www.aece.ro on Thursday, June 10, 2021 at 14:03:12 (UTC) by 97.69.214.164. Redistribution subject to AECE license or copyright.]

Advances in Electrical and Computer Engineering Volume 21, Number 2, 2021

106

[2] X. Liu, P. Wang, and P. C. Loh, “A hybrid AC/DC microgrid and its coordination control,” IEEE Transactions on Smart Grid, vol. 2, no. 2, pp. 278–286, Jun. 2011. doi:10.1109/TSG.2011.2116162

[3] F. Nejabatkhah and Y. W. Li, “Overview of power management strategies of hybrid AC/DC microgrid,” IEEE Transactions on Power Electronics, vol. 30, no. 12, pp. 7072–7089, Dec. 2015. doi:10.1109/TPEL.2014.2384999

[4] F. Blaabjerg, R. Teodorescu, M. Liserre, and A. V. Timbus, “Overview of control and grid synchronization for distributed power generation systems,” IEEE Transactions on Industrial Electronics, vol. 53, no. 5, pp. 1398–1409, Oct. 2006. doi:10.1109/TIE.2006.881997

[5] T. D. Hund, S. Gonzalez, and K. Barrett, “Grid-tied PV system energy smoothing,” in Conference Record of the IEEE Photovoltaic Specialists Conference, Jun. 2010, pp. 2762–2766. doi:10.1109/PVSC.2010.5616799

[6] J. Pegueroles-Queralt, F. D. Bianchi, and O. Gomis-Bellmunt, “A power smoothing system based on supercapacitors for renewable distributed generation,” IEEE Transactions on Industrial Electronics, vol. 62, no. 1, pp. 343–350, Jan. 2015. doi:10.1109/TIE.2014.2327554

[7] C. Ceja-Espinosa and E. Espinosa-Juárez, “Smoothing of photovoltaic power generation using batteries as energy storage,” in IEEE PES Innovative Smart Grid Technologies Conference - Latin America, ISGT Latin America, Sep. 2017, pp. 1–6. doi:10.1109/ISGT- LA.2017.8126758

[8] A. M. Howlader, N. Urasaki, A. Yona, T. Senjyu, and A. Y. Saber, “A review of output power smoothing methods for wind energy conversion systems,” Renewable and Sustainable Energy Reviews, vol. 26, pp. 135–146, Oct. 2013. doi:10.1016/j.rser.2013.05.028

[9] H. Lee, M. Hwang, E. Muljadi, P. Sørensen, and Y. C. Kang, “Power- smoothing scheme of a DFIG using the adaptive gain depending on the rotor speed & frequency deviation,” Energies, vol. 10, no. 4, p. 555, Apr. 2017. doi:10.3390/en10040555

[10] X. Li, D. Hui, and X. Lai, “Battery energy storage station (BESS)- based smoothing control of photovoltaic (PV) and wind power generation fluctuations,” IEEE Transactions on Sustainable Energy, vol. 4, no. 2, pp. 464–473, Apr. 2013. doi:10.1109/TSTE.2013.2247428

[11] G. Mandic, A. Nasiri, E. Ghotbi, and E. Muljadi, “Lithium-ion capacitor energy storage integrated with variable speed wind turbines for power smoothing,” IEEE Journal of Emerging and Selected Topics in Power Electronics, vol. 1, no. 4, pp. 287–295, Dec. 2013. doi:10.1109/JESTPE.2013.2284356

[12] M. R. I. Sheikh and N. Mondol, “Wind power smoothing scheme using SMES with reduced capacity,” in International Conference on Informatics, Electronics and Vision, ICIEV, May 2012, pp. 404–410. doi:10.1109/ICIEV.2012.6317425

[13] X. Han, F. Chen, X. Cui, Y. Li, and X. Li, “A power smoothing control strategy and optimized allocation of battery capacity based on hybrid storage energy technology,” Energies, vol. 5, no. 5, pp. 1593– 1612, May 2012. doi:10.3390/en5051593

[14] N. S. Jayalakshmi and D. N. Gaonkar, “A new control method to mitigate power fluctuations for grid integrated PV/wind hybrid power system using ultracapacitors,” International Journal of Emerging Electric Power Systems, vol. 17, no. 4, pp. 451–461, Aug. 2016. doi:10.1515/ijeeps-2015-0183

[15] R. Sharma and S. Suhag, “Supercapacitor utilization for power smoothening and stability improvement of a hybrid energy system in a weak grid environment,” Turkish Journal of Electrical Engineering & Computer Sciences, vol. 26, no. 1, pp. 347–362, Jan. 2018. doi:10.3906/elk-1703-101

[16] M. Y. Worku and M. A. Abido, “Fault ride-through and power smoothing control of PMSG-Based wind generation using supercapacitor energy storage system,” Arabian Journal for Science and Engineering, vol. 44, no. 3, pp. 2067–2078, Mar. 2019. doi:10.1007/s13369-018-3284-1

[17] D. Lamsal, V. Sreeram, and Y. Mishra, “Reducing power fluctuations from wind and photovoltaic systems using discrete Kalman filter,” in Australasian Universities Power Engineering Conference (AUPEC), Sep. 2016, pp. 1–5. doi:10.1109/aupec.2016.7749297

[18] D. Lamsal, V. Sreeram, Y. Mishra, and D. Kumar, “Achieving a minimum power fluctuation rate in wind and photovoltaic output power using discrete kalman filter based on weighted average approach,” IET Renewable Power Generation, vol. 12, no. 6, pp. 633– 638, Jan. 2018. doi:10.1049/iet-rpg.2017.0346

[19] D. Lamsal, V. Sreeram, Y. Mishra, and D. Kumar, “Output power smoothing control approaches for wind and photovoltaic generation systems: A review,” Renewable and Sustainable Energy Reviews, vol. 113, p. 109245, Oct. 2019. doi:10.1016/j.rser.2019.109245

[20] N. S. Jayalakshmi, D. N. Gaonkar, and P. B. Nempu, “Power control of PV/fuel cell/supercapacitor hybrid system for stand-alone applications,” International Journal of Renewable Energy Research, vol. 6, no. 2, pp. 672–679, Jun. 2016

[21] M. Uzunoglu, O. C. Onar, and M. S. Alam, “Modeling, control and simulation of a PV/FC/UC based hybrid power generation system for stand-alone applications,” Renewable Energy, vol. 34, no. 3, pp. 509– 520, Mar. 2009. doi:10.1016/j.renene.2008.06.009

[22] T. Schucan, “Case studies of integrated hydrogen systems,” IEA Hydrogen Implementing Agreement, Final report for Subtask A of Task. 11, Dec. 1999

[23] A. Reznik, M. G. Simoes, A. Al-Durra, and S. M. Muyeen, “LCL Filter design and performance analysis for grid-interconnected systems,” IEEE Transactions on Industry Applications, vol. 50, no. 2, pp. 1225–1232, Mar. 2014. doi:10.1109/TIA.2013.2274612

[24] K. Tatjana “Control of voltage source converters for power system applications,” Master's thesis, Norwegian University of Science and Technology, 2011

[25] D. Simon, “Kalman filtering,” Embedded systems programming, vol. 14, no. 6, pp. 72–79, Jun. 2001

[26] G. Welch and G. Bishop, An Introduction to the Kalman filter, pp. 1– 16, Jul. 2006

[Downloaded from www.aece.ro on Thursday, June 10, 2021 at 14:03:12 (UTC) by 97.69.214.164. Redistribution subject to AECE license or copyright.]

Performance_Analysis_of_Implan.PDF

1

Electrica 2021; 21(1): 1-9

RESEARCH ARTICLE

Performance Analysis of Implanted Hybrid Three Quasi Z source Inverter designed for Renewable Energy Conversion Applications Ramanjaneyulu Alla , Anandita Chowdhury Department of Electrical Engineering, Sardar Vallabhbhai National Institute of Technology, Surat, India

Corresponding Author: Ramanjaneyulu Alla

E-mail: [email protected]

Received: 03.09.2019

Accepted: 12.07.2020

Available Online Date: 10.09.2020

DOI: 10.5152/electrica.2020.19068

Cite this article as: Alla R, Chowdhury A. Performance Analysis of Implanted Hybrid Three Quasi Z source Inverter designed for Renewable Energy Conversion Applications. Electrica, 2021; 21(1): 1-9.

ABSTRACT

Implanted hybrid three quasi Z source inverter has been designed to achieve high voltage gain for a single-stage power conversion system. The net input voltage applied to the converter is the sum of the voltages of four independent sources to which the converter is configured. It is a fact that the power from renewable energy sources will not be constant and varies. So there is uncertainty in the uniform availability of energy from renewable energy sources and the resulting voltage variations in any of the four sources affects net input voltage of the converter. A dual-loop control method has been developed to control the dc-link voltage across the converter to attain a constant ac output voltage across load terminals and meet load requirements. The controller maintains the stable operation of the converter in the absence of any of the four voltage sources. Simulation results using MATLAB software are presented for the validation of the proposed work under the conditions of input voltage variations as well as a sudden change of loads. Keywords: Shoot-through duty ratio, modulation index, simple boost control, voltage gain, Z source converter

Introduction

Power converters are utilized to extract the maximum power from renewable energy sourc- es  (RES). In general, the voltage magnitude at the renewable energy extraction system is at a lower value. So power converters with high voltage gain are required to extract energy from renewable sources. Conventional voltage source and current source converters need secondary power conversion circuits to either reduce or boost the input voltage. These conventional con- verters have operational limitations and can be overcome by the impedance source converters (ISC). The first ISC is known as Z source converter (ZSC) [1]. The problems encountered in ZSC are solved in a quasi-Z source converter (QZSC) [2] and it has ZSC features. Now many configurations of ISC are developed to enhance the converter voltage gain and are reviewed in [3].

Magnetically coupled ISC (MCISC) are discussed in [4] which provide high voltage gains by changing the turns ratio value of the coupled coils and shoot-through duty ratio (D) value. In MCISC, higher voltage gains are achieved at higher modulation index (m) by varying turns ra- tio of the coils rather than D, Converters [4] are designed with a lesser number of circuit ele- ments to get increase in voltage gain. Apart from these advantages, MCISC experiences startling voltage spikes [5] at dc-link terminals due to magnetic leakage. MCISC need clamping circuits [6] to reduce the voltage spikes. The addition of the clamping circuits increases the number of components in converter configuration. The review of pulse width modulation (PWM) methods [7] designed for ISC is discussed. Most researchers use the simple boost control method (SBC) method, as its implementation is very easy. The applications of ZSC in motor drive applications are illustrated in [8]. The application of ZSC for reactive power compensation [9] was explained.

Steady ac voltage across the connected load or for grid-connected applications is needed to regulate the dc-link voltage of the converter. As the dc-link voltage varies from zero to a peak value, the capacitor voltage is considered as a feedback voltage, both of them are expressed in terms of D. Controlling the dc-link voltage with improved transient response [10] was dis- cussed, but the effect of inductor current ripples are not considered. The inductor current

Content of this journal is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.

2

Electrica 2021; 21(1): 1-9 Alla and Chowdhury. Performance Analysis of ImHTQZSI for RES Applications

ripples are controlled in the inner loop and dc-link voltage is controlled in the outer loop [11] with the implementation of dual-loop digital control.

Hybrid three quasi Z source dc-dc converter provides higher voltage gain [12]. Z source half-bridge inverter [13] has been de- signed to reduce voltage stress on capacitors and the cascade connection with the inverter was presented to get the higher voltage gain. An embedded ZSC [14] is formed by placing the half dc voltage at the inductors helps to draw smoother current at a voltage gain equals to the voltage gain in ZSC. Further, var- ious topologies are designed to address the various network problems [15-20], resulting in improving the voltage gain of the ISC. Still, research is continuing toward the development of high gain and efficient topologies. The ISC working at low power fac- tor loads or lower inductance value in the impedance network of the converter causes higher current ripples, as the converter might be working in discontinuous conduction mode [21].

Implanted hybrid three quasi Z source Inverter (ImHTQZSI) [15] is developed by using hybrid three quasi Z source dc-dc converter [12] by connecting four individual voltage sources in series with the inductors in impedance circuit. Although MCISC provides higher gain, it faces the voltage spikes at dc-link terminals. So clamping circuits are required to reduce voltage spikes in MCISC and it makes the circuit very complex. Hence, ImHTQZSI has been considered as it provides higher voltage gain compared with existing non-coupled coil ISC. This paper illustrates the configuration of ImHTQZSI and operating modes in Section 2. Input voltage disturbances due to uncertainties in the renew- able energy sources are controlled by using a dual-loop control method on the dc-side of the converter is explained in Section 3. The performance of the converter with simulation results in the proposed control method is discussed in Section 4.

Configuration and Working of Proposed Inverter

QZSC with two configurations of the [2] are used to develop the present converter is shown in Figure 1. Here V

1 , V

2 , V

3, and V

4 are

the dc voltage sources extracted from either renewable energy sources or direct dc sources. The network having components L

1 , C

1 , D

1 , L

2, and C

2 represent the discontinuous current mode

(DCM) operation of [2]. Similarly, the network formed with L 3 ,

C 5 , D

3 , L

4, and C

6 is another DCM [2]. The final appearance of the

converter with C 3 , C

4, and diode D

2 is visualized as continuous

current QZSC [2]. As a whole, the present QZSC has three con-

figurations. A dc voltage source in series with the inductor is connected to get continuous input current from the sources. It can work in two modes; in one mode it works as normal in- verter at both active and zero states known as the non-shoot- through state and in the second mode an extra zero state is added known as shoot-through zero state.

Mode-1— Shoot-through zero state

The equivalent circuit of Figure 1 in shoot-through zero state operation is shown in Figure 2. In this mode all diodes are in off state and the converter terminals are short-circuited during the interval of T

st in a switching period of T. Converter terminals

are short-circuited with turning on all switching devices at any of the leg or all legs or combination of legs. In mode-1, the in- ductors are energized by the capacitors leading to increase in current passing through inductors.

The voltage across the inductors in mode-1 is expressed in equations (1)–(4). Here V

L1 , V

L2 , V

L3, and V

L4 are voltage across the

respective inductors. Similarly V C1

, V C2

, V C3

, V C4

, V C5,

and V C6

are voltage across the respective capacitors.

L1 1 C2 C4V =V +V +V (1)

L2 2 C1 C4V =V +V +V (2)

L3 3 C3 C6V =V +V +V (3)

L4 4 C3 C5V =V +V +V (4)

Mode-2— non-shoot-through state

In non-shoot-through state all the diodes are on state and con- verter terminals are connected to load. They are represented with an equivalent current source shown in Figure 3 at a termi- nal voltage of V

LK during the interval of T

nst in a switching peri-

od of T. In this mode switching will happen as in convention- al voltage source converter. The voltage across the inductors in mode-2 is expressed in equations (5)–(8). V

LK is the voltage

across the dc-link terminals of the converter.

L1 1 C1V =V -V (5)

L2 2 C2V =V -V (6)

Figure 2. Mode-1 representation of ImHTQZSIFigure 1. Implanted Hybrid Three Quasi Z source Inverter

3

Electrica 2021; 21(1): 1-9 Alla and Chowdhury. Performance Analysis of ImHTQZSI for RES Applications

L3 3 C5V =V -V (7)

L4 4 C6V =V -V (8)

C3 C4 LKV +V =V (9)

The shoot-through duty ratio of the converter is represented as

(10)

According to volt-second balance principle across the inductor, the individual capacitor and dc-link voltages are expressed in equations (11) – (17)

(11)

(12)

(13)

(14)

(15)

(16)

(17)

Equation (17) indicates the relation between dc-link voltage of the converter and four dc voltage sources. The net input dc voltage applied to the converter is the sum of the voltages of four individual dc sources as seen in equation (18).

(18)

(19)

where B denotes the voltage boost factor.

The value of inductances L 1 -L

4 are considered according to

equation (20)

L,i i

L,i%i v DT

L = i=1,2,3, and 4 (20)

Where %i L,i

is the percentage of ripple allowed in the corre- sponding inductor current.

The value of capacitances C 1 -C

6 is considered according to

equation (21) C,i

i C,i%v

i DT C = i=1,2,3, 4, 5, and 6 (21)

Where i C,i

is current through the ith capacitor and %v C,i

is the percentage of ripple allowed in the corresponding capacitor voltage.

Dual-Loop Control of ImHTQZSI

The family of ISC is designed to provide a high voltage gain in a single-stage conversion system. In this circuit, shoot-through zero state is inserted in place of conventional zero states to get boost voltages at dc-link. Three basic PWM methods, namely, SBC [2, 7], maximum boost control, and maximum constant boost control methods are used to get a peak output ac volt- age from a boosted dc-link voltage. This paper uses SBC meth- od to show the ImHTQZSI performance.

The relation between m and D in SBC control is

D+m=1 (22) Output peak ac voltage can be represented as

LK ac

V V =m

2 (23)

The disturbances on input energy sources cause fluctuations in dc voltage across the converter terminals, which in turn vary the output ac voltage. To get a steady output ac voltage, the designer needs to control dc-link voltage by controlling ‘D’ on dc-side and load disturbances on ac-side by controlling ‘m’.

Dc-side control

Dc-link voltage V LK

is maintained on dc-side to make the convert- er to provide required v

ac , according to equation (23). V

LK depends

on D. V LK

is varying from zero to a peak value just like a pulse sig- nal. V

LK is zero in STS and has a finite value in nSTS state. One PI

controller is enough to control V Lk

, by sensing the peak value of the dc-link voltage. For that, an extra sensing circuit is required. To avoid more or additional sensing circuit, capacitor voltage V

C3 is considered to control V

LK , as V

LK and capacitor voltages are

functions of D. One more PI controller is considered to control inductor current ripples and the bandwidth of the overall con- troller is increased [10, 11]. Hence two PI controllers required. V

C3

is considered as a feedback element instead of V LK

, and it means indirect controlling of dc-link voltage as shown in Figure 4. Apart

Figure 3. Mode-2 representation of ImHTQZSI

Figure 4. Dual Loop Control of of ImHTQZSI

4

Electrica 2021; 21(1): 1-9 Alla and Chowdhury. Performance Analysis of ImHTQZSI for RES Applications

from controlling V LK

, inductor current ripples are also regulated with the inner current loop. The outcome of the dc-side control is D. Outer voltage loop provides reference inductor current (I

Lref )

given in equation (24) where S is the load in volt-amperes.

(24)

Ac-side control

Load currents i abc

are considered as a feedback signal to gener- ate modulating signals. The load current and its reference cur- rents are transformed into dq components and are supplied to PI control. The output of PI control is then converted into abc components to provide modulating signals as shown in Figure 4. With the interaction of m and D values in the SBC method, the converter operates at a steady operating point. D and m are interrelated, as seen from equation (22). Disturbances on the dc-side of the converter influence ac-side performance of the converter as well as disturbances on the ac-side are also affect dc-side performance.

The reference peak ac current (i peak

) given in equation (25)

(25)

Here V l is the line voltage and cosφ is the load power factor.

The voltage stress across the switching devices are expressed in equation (26)

(26)

Here V SW

is the voltage stress across the switches in the inverter

The current stress across the switching devices are expressed in equation (27)

= = inD1 D2 D3 I

i =i i 1-D

(27)

Simulation Results

The performance of the proposed converter with the dual-loop control method has to be tested and is carried out with MAT- LAB/simulink. The system parameters are listed in Table 1. Here step change in applied input voltage and load volt-ampere are considered disturbances for the validation of the controller

Figure 5. Steady state response of ImHTQZSI to step change in dc voltage and load VA

Table 1. System parameters

Parameter Value

Inductance 2 mH

Capacitance 470 µf

Internal Resistance of Inductor 0.1 Ω

Internal Resistance of Capacitor 0.1 Ω

Switching Frequency 10 kHz

Output Voltage 70.7 V(ph)

Load 250-375 VA, 0.8 pf lag

Table 2. Variation of converter parameters with input voltage and load changes

V 1

V 2

V 3

V 4

VC 3

VC 4

V LK

D G S i abc

25 25 25 25 116.7 116.7 233.3 0.143 2 250 1.67

25 0 25 25 108.5 133.5 241.9 0.173 2.67 250 1.67

25 0 0 25 125 125 250 0.2 4 250 1.67

25 0 0 25 125 125 250 0.2 4 375 2.5

25 0 25 25 108.5 133.5 241.6 0.173 2.67 375 2.5

25 25 25 25 116.7 116.7 233.3 0.143 2 375 2.5

V 1 , V

2 , V

3 and V

4 are the input side dc source voltages in volts. V

C3 and V

C4 are the voltage across the capacitors C

3 and C

4 in volts. V

LK is the voltage across the dc- link terminals

in volts. D, G and S are the shoot-through duty ratio, Converter voltage gain and load Volt-Amperes, respectively. i abc

is the current through the load in amperes.

5

Electrica 2021; 21(1): 1-9 Alla and Chowdhury. Performance Analysis of ImHTQZSI for RES Applications

performance. Figure 5 shows the steady-state response of the converter with disturbances at every 0.5 sec and the perfor- mance details are given in Table 2.

There is no load change in ac-side of the converter for 0 to 1.5 sec when input dc voltages are reduced by 25V at every 0.5 sec. At 1.5 sec load is increased by 50% and after at every 0.5sec dc voltage is increased by 25V. In all the cases, the converter meets the load requirements with four dc sources or three dc sources or two dc sources.

To better analyze waveforms, they are considered as five case studies to illustrate the transient response. I

L1 and S are the cur-

rent through the inductor L 1 and load Volt-Amperes, respec-

tively.

Case i: V 1 =25 V, V

2 =0 V, V

3 =25 V, and V

4 =25 V at 250 VA, 0.8

pf load

Out of four dc voltage sources in the converter, V 2 is considered

inactive in this case due to general maintenance issues or main- tenance issues because of breakdown in the renewable energy extraction system, makes net input voltage is at 75 V (V

dc ) in

the impedance network. Step change in V dc

value from 100V to 75V is indicated in the simulation. Change in voltage on dc-side at 0.5 sec results in immediate reduction of capacitor voltage and increases afterward, finally reaching a particular value due to the non-minimum phase phenomena. The dc-link voltage changes in the transient period are observed in Figure 6.

As the V dc

is decreased by 25 V, D is increased from 0.143 to 0.173. Hence, the ac per phase output voltage at 70.7 V is main- tained uniformly. Correspondingly, capacitor voltage distribu- tion in the network is also varied. Here V

C3 value is changed to

108.5 V from 116.7 V and V LK

is changed to 241.6 V from 233.3 V. As the load on the converter is not changed, there is no change in load current magnitude. Under step change in dc voltage on the input side, based on equation (24), inductor current I

L1

jumps from 2.5 A to 3.33 A and the transient period continues with the duration from 0.5 sec to 0.6 sec and then the converter stays at the steady-state condition.

Case ii: V 1 =25 V, V

2 =0 V, V

3 =0 V, and V

4 =25 V at 250 VA, 0.8

pf load

A step change in voltage from 75 V to 50 V has occurred due to the inactiveness of voltage source V

3 and meets the load

Volt-Amperes. The overall change in voltage is from 100 V to 50 V. ImHTQZSI has two dc voltage sources out of four. Two of the dc voltage sources V

2 and V

3 are inactive which makes net

input dc voltage V dc

as 50 V. Change in voltage on dc-side at the instant 1 sec causes immediate changes in capacitor voltages and the dc-link voltage, and the transient response is observed in Figure 7. As the V

dc is decreased by 25 V, D has been increased

from 0.173 to 0.2 to maintain the ac per phase output voltage at 70.7 V. Here V

C3 value has been changed to 125 V from 108.5

V and V LK

changed to 250 V from 241.6 V.

In this case, also there are no load changes on the converter as there is no change in i

abc magnitude. Under step change in V

dc ,

I L1

jumps from 3.33 A to 5 A and the transient period continues with the duration from 1 sec to almost 1.15 sec and then the converter reaches the steady-state condition. Figure 8 shows very low total harmonic distortion (THD) in current waveform on load side during the interval zero sec to 1.5 sec. It is evi- dent from the case i and case ii that the proposed converter can reach the load requirements even when any of the input voltage source is not available which provides flexibility in the maintenance of the input dc voltage source mechanisms.

Figure 6. Response of ImHTQZSI at 100V to 75V step change in dc voltage

Figure 7. Response of ImHTQZSI at 75V to 50V step change in dc voltage

6

Electrica 2021; 21(1): 1-9 Alla and Chowdhury. Performance Analysis of ImHTQZSI for RES Applications

Case iii: V 1 = 25 V, V

2 = 0 V, V

3 =0 V, and V

4 =25 V at 375 VA, 0.8

pf load

As in the previous case, there is no change in dc input voltage but 50% step change of load from 250 VA to 375 VA at 0.8 pf lag occurred at 1.5 sec. It influences I

L1 to

increase from 5 A to 7.5 A

and also i abc

changes from 1.667 A, peak current to 2.5 A, peak val- ue as shown in Figure 9. The load disturbance also causes tran- sient condition on the dc-side. As the m and D are interdepen-

dent on one another, disturbances on either side are transferred and influence performance on both sides. In this case, there is no change in input dc voltage and hence the dc-link voltage and capacitor voltages continue to be as in the previous case.

Case iv: V 1 =25 V, V

2 =0 V, V

3 =25 V, and V

4 =25 V at 375 VA, 0.8

pf load

Just as in the previous case, step change in voltage from 50V to 75V is done without any change in load at 2 sec. As the voltage on dc-side increases, I

L1 will decrease from 7.5 A to 5 A as shown

in Figure 10, whereas in case ii voltage decreased to 50 V from 75 V. As there is change in dc voltage, the capacitor voltages and dc-link voltage increases and then decreases and settles at a particular value due to non-minimum phase phenomena. V

C3

changes from 125 V to 108.5 V and V LK

changes to 241.6 V from 250 V, respectively.

It is evident from the case iii and case iv that the proposed con- verter can overcome the load variations when any one of the input voltage sources is not available.

Case v: V 1 =25 V, V

2 =25 V, V

3 =25 V, and V

4 =25 V at 375 VA,

0.8 pf load

In case i the change in dc voltage was decreased whereas in case v it was increased at 2.5 sec, but the loads at both the cases are different. Here all the dc sources have met the load requirements with a change in load from 250 VA to 375 VA as shown in Figure 11. I

L1 is decreased from 5 A to 3.75 A. Figure 12

shows very less THD in the current waveform on the load side with a peak current value of 2.5 A after 2.5 sec. As dc voltage changes, V

C3 and V

LK are also changed from 108.5 V to 116.7 V

and 241.6 V to 233.3 V, respectively. Figure 8. Current THD at 250VA, 0.8pf lag

Figure 10. Response of ImHTQZSI for a step change in VA from 250VA to 375VA as well as change in voltage from 50V to 75V

Figure 9. Response of ImHTQZSI for a step change in VA from 250VA to 375VA as well as change in voltage continues

7

Electrica 2021; 21(1): 1-9 Alla and Chowdhury. Performance Analysis of ImHTQZSI for RES Applications

The dc-link voltage V LK

has been varying from zero to a peak value as shown in Figure 13. It has a zero value in the shoot- through period and a peak value in the non-shoot-through pe- riod. Figure 14 shows the voltage boost factor in comparison with other converters. When D is in the range of 0 to 0.25, the proposed converter provides a higher voltage gain. It has been observed that for an incremental step change of input dc volt-

ages or an incremental change in load disturbance can cause overshoots in capacitor and dc-link voltages. Whereas a decre- ment step change of input dc voltages can cause undershoots in the capacitor and dc-link voltages is observed in Figure 5. It has been observed from all the cases, the converter has been working well to meet load requirements with dual-loop control method even when variations occur in source voltage and load voltage.

Conclusion

An implanted hybrid three QZSC has been designed for utiliz- ing its four dc voltage sources efficiently to meet load require- ments. This converter is providing higher voltage gain when compared to the non-coupled ISC. The net input dc voltage applied to the converter is the sum of all voltage sources that existed in the converter configuration. The steady-state and dy- namic response of the converter has been illustrated with the dual-loop control method. The converter has been providing a constant ac voltage across the load terminals by controlling the dc-link voltage. By controlling capacitor voltage as well as the inductor current ripples with the dual-loop control method on dc-side, we are able to control the dc-link voltage. The control- ler tracks the reference current on the load side under the sud- den change of load on the ac-side as well as the change in net dc voltage. The controller has been providing lower THD in the load current. The converter has been working in the absence

Figure 11. Response of ImHTQZSI for a step change in VA from 250VA to 375VA as well as change in voltage from 75V to 100V

Figure 13. Closed observation of dc-link voltage

Figure 12. Current THD at 375VA, 0.8pf lag Figure 14. Voltage boost factor comparision among the con- verters

8

Electrica 2021; 21(1): 1-9 Alla and Chowdhury. Performance Analysis of ImHTQZSI for RES Applications

of any of the voltage sources and so it provides flexibility of operation even when any one of the input source is not avail- able due to maintenance, faults, or any other reason that occur during the energy extraction mechanisms. The four dc voltages are considered to be extracted from various renewable energy sources.

Peer-review: Externally peer-reviewed.

Conflict of Interest: The authors have no conflicts of interest to de- clare.

Financial Disclosure: The authors declared that this study has re- ceived no financial support.

References

1. F. Z. Peng, “Z-source inverter”, IEEE Transactions on Industry Appli- cations, vol. 39, No.2, pp. 504-510, March-April, 2003. [CrossRef]

2. J. Anderson, F.Z. Peng, “Four Quasi-Z-Source Inverters”, 2008 IEEE Power Electronics Specialists Conference, Rhodes, Greece, August 2008, pp. 2743-2749. [CrossRef]

3. O. Ellabban, H. Abu-Rub, “Z Source Inverter: Topology Improve- ments Review” IEEE Industrial Electronics magazine, vol.10, No.1, pp. 6-24, March, 2016. [CrossRef]

4. Yam P. Siwakoti, F. Blaabjerg, P. C. Loh, “New Magnetically Cou- pled Impedance (Z-) Source Networks” IEEE Transactions on Pow- er Electronics, Vol. 31, No. 11, pp. 7419-7435, November, 2016. [CrossRef]

5. Y. P. Siwakoti, P. C. Loh, F. Blaabjerg, G. E. Town, “Effects of Leakage Inductances on Magnetically Coupled Y-Source Network”, IEEE Transactions on Power Electronics, Vol. 29, No. 11, pp. 5662-5666, November, 2014. [CrossRef]

6. H. Liu, Z. Zhou, K. Liu, P. C. Loh, W. Wang, D. Xu, F. Blaabjerg, “A Family of High Step-Up Coupled-Inductor Impedance-Source Inverters With Reduced Switching Spikes”, IEEE Transactions on Power Electronics, Vol. 33, No. 11, pp. 9116 – 9121, November, 2018. [CrossRef]

7. A. Abdelhakimr, F. Blaabjerg, P. Mattavelli, “Modulation Schemes of the Three-Phase Impedance Source Inverters—Part I: Classifica- tion and Review”, IEEE Transactions on Industrial Electronics, Vol. 65, No. 8, pp. 6309 – 6320, August, 2018. [CrossRef]

8. O. Ellabban, H. Abu-Rub, “An overview for the Z-Source Converter in motor drive applications”, Renewable and Sustainable Energy Reviews Vol.61 pp. 537–555, August, 2016. [CrossRef]

9. J. Zhang, “Unified control of Z-source grid-connected photovol- taic system with reactive power compensation and harmonics

restraint: design and application”, IET Renewable Power Genere- ration, Vol. 12, No. 4, pp. 422-429, March, 2018. [CrossRef]

10. Q. Tran, T. Chun, J. Ahn, H. Lee, “Algorithms for Controlling Both the DC Boost and AC Output Voltage of Z-Source Inverter”, IEEE Trans- actions on Industrial Electronics, Vol. 54, No. 5, pp. 2745 – 2750, October, 2007. [CrossRef]

11. H. Liu, P. Liu, Y. Zhang, “Design and digital implementation of voltage and current mode control for the quasi-Z-source convert- ers”, IET Power Electronics, Vol. 6, No. 5, pp. 990–998, June, 2013. [CrossRef]

12. H. Shen, B. Zhang, D. Qiu, “Hybrid Z-Source Boost DC–DC Convert- ers”, IEEE Transactions Industrial Electronics, Vol. 64, No. 1 pp. 310 – 319, January, 2017. [CrossRef]

13. E. Babaei, E. S. Asl, “A new topology for Z-source half-bridge invert- er with low voltage stress on capacitors”, Electric Power Systems Research, vol. 140, pp722–734, November, 2016. [CrossRef]

14. P. C. Loh, F. Gao, F. Blaabjerg, “Embedded EZ-Source Inverters”, IEEE Transactions on Industry Applications, Vol. 46, No. 1, pp 256 – 267, January-February, 2010. [CrossRef]

15. R. Alla, A. Chowdhury, “An Implanted Hybrid Three Quasi Z source Inverter for Photovoltaic Power Generation Applications”, IEEE International Students’ Conference on Electrical, Electronics and Computer Science (SCEECS), NIT Bhopal, India February 2018, pp. 1- 6. [CrossRef]

16. V. Jagan, J. Kotturu, S. Das, “Enhanced-Boost Quasi-Z-Source Inverters With Two-Switched Impedance Networks”, IEEE Trans- actions on Industrial Electronics, Vol. 64, No. 9, pp. 6885 – 6897, September, 2017. [CrossRef]

17. X. Zhu, B. Zhang, D. Qiu, “A High Boost Active Switched Quasi-Z- Source Inverter With Low Input Current Ripple”, IEEE Transactions on Industrial Informatics, Vol. 15, No. 9, pp. 5341 – 5354, Septem- ber, 2019. [CrossRef]

18. S. Rostami, V. Abbasi, F. Blaabjerg, “Implementation of a common grounded Z-source DC–DC converter with improved operation factors”, IET Power Electronics, Vol. 12, Iss. 9, pp. 2245-2255, Au- gust, 2019. [CrossRef]

19. M. H. B. Nozadian, E. Babaei, E. S. Asl, S. H. Hosseini, “Switched Z-source networks: a review”, IET Power Electronics, Vol. 12, Iss. 7, pp. 1616-1633, July, 2019. [CrossRef]

20. M. Veerachary, P. Kumar, “Analysis and Design of Sixth Order Qua- si-Z-Source DC-DC Boost Converter”, IEEE International Conference on Sustainable Energy Technologies and Systems (ICSETS) Bhu- baneswar, India February-March, 2019 pp. 347 – 352. [CrossRef]

21. M. Shen, F. Z. Peng, “Operation Modes and Characteristics of the Z-Source Inverter With Small Inductance or Low Power Factor”, IEEE Transactions on Industrial Electronics, Vol. 55, No. 1, pp. 89 – 96, January, 2008. [CrossRef]

9

Electrica 2021; 21(1): 1-9 Alla and Chowdhury. Performance Analysis of ImHTQZSI for RES Applications

Ramanjaneyulu Alla received the B.Tech. degree in electrical and electronics engineering from Jawahar Lal Nehru Technological University Hyderabad, India, in 2005, and the M.Tech. degree in Power Electronics from Jawahar Lal Nehru Technological University Kakinada, India, in 2012, Currently he pursuing Ph.D. degree in the Electrical Engineering Department at S.V. National Institute of Technology, Surat in India, from July 2017. His current research interests include power electronics, impedance source converters and its applications.

Anandita Chowdhury received her B.E. and M.E. degree from the University of Calcutta, and the Ph.D. degree from the Indian Institute of Technology, Kharagpur. Presently she is working as Professor in the Department of Electrical Engineering of S. V. National Institute of Technology, Surat, India. She has more than twenty five years of teaching experience. Her area of research interest includes electrical machines, drives and power system stability.

Tidal Energy.PDF

Journal of

Marine Science and Engineering

Article

Modeling Assessment of Tidal Energy Extraction in the Western Passage

Zhaoqing Yang 1,*, Taiping Wang 1 , Ziyu Xiao 1 , Levi Kilcher 2 , Kevin Haas 3, Huijie Xue 4

and Xi Feng 5

1 Pacific Northwest National Laboratory, 1100 Dexter Avenue North, Suite 500, Seattle, WA 98109, USA; [email protected] (T.W.); [email protected] (Z.X.)

2 The National Renewable Energy Laboratory, Golden, CO 80401, USA; [email protected] 3 School of Civil and Environmental Engineering, Georgia Institute of Technology, North Ave NW,

Atlanta, GA 30332, USA; [email protected] 4 School of Marine Sciences, University of Maine, 168 College Ave, Orono, ME 04469, USA; [email protected] 5 Key Laboratory of Coastal Disaster and Defence, Hohai University, Ministry of Education, 1 Xikang Road,

Nanjing 210098, China; [email protected] * Correspondence: [email protected]; Tel.: +1-206-528-3057

Received: 19 May 2020; Accepted: 3 June 2020; Published: 5 June 2020 ���������� �������

Abstract: Numerical models have been widely used for the resource characterization and assessment of tidal instream energy. The accurate assessment of tidal stream energy resources at a feasibility or project-design scale requires detailed hydrodynamic model simulations or high-quality field measurements. This study applied a three-dimensional finite-volume community ocean model (FVCOM) to simulate the tidal hydrodynamics in the Passamaquoddy–Cobscook Bay archipelago, with a focus on the Western Passage, to assist tidal energy resource assessment. IEC Technical specifications were considered in the model configurations and simulations. The model was calibrated and validated with field measurements. Energy fluxes and power densities along selected cross sections were calculated to evaluate the feasibility of the tidal energy development at several hotspots that feature strong currents. When taking both the high current speed and water depth into account, the model results showed that the Western Passage has great potential for the deployment of tidal energy farms. The maximum extractable power in the Western Passage was estimated using the Garrett and Cummins method. Different criteria and methods recommended by the IEC for resource characterization were evaluated and discussed using a sensitivity analysis of energy extraction for a hypothetical tidal turbine farm in the Western Passage.

Keywords: tidal energy; Western Passage; resource characterization; numerical modeling; FVCOM

1. Introduction

Instream tidal energy is one of the most popular marine renewable energy sources because it is highly predictable and the associated technology is relatively mature [1–4]. One of the important steps toward harvesting instream tidal energy is resource characterization and assessment, at either the project-design or regional scales. Although significant efforts have been made to assess the maximum potential of tidal stream energy at a system-wide scale using theoretical methods or numerical models [5–12], the accurate assessment of tidal energy resources at the project-design scale requires detailed tidal hydrodynamic information obtained from intense field measurements [13–15] and high-resolution simulations using three-dimensional (3-D) numerical models [16–21]. Field measurements can not only provide direct and accurate resource assessment at a specific site but also support model validation and build confidence in the models. On the other hand,

J. Mar. Sci. Eng. 2020, 8, 411; doi:10.3390/jmse8060411 www.mdpi.com/journal/jmse

J. Mar. Sci. Eng. 2020, 8, 411 2 of 21

numerical models can be used to generate rich data sets at larger spatial–temporal scales to support resource assessment and guide further field measurements. For example, numerical models can be used to evaluate the efficiency of multiple tidal farms [22,23], the effects of tidal energy extraction on vertical mixing [24], and flushing time [25,26], and provide better coverage in both the temporal and spatial domains than measurements. Extended model validation using high-quality measured data is essential to minimize the uncertainties of resource assessment. The International Electrotechnical Commission (IEC) Technical Specification (TS) recommends that tidal resource characterization be conducted based on detailed and accurate hydrodynamic information, which can be obtained from either direct field measurements or hydrodynamic modeling [27]. Specifically, IEC recommends that for small projects whose total extracted power is less than 10 MW or 2% of the theoretical tidal resource, extractable energy at the project site should be estimated based on the modeled undisturbed flow or direct velocity measurements at the turbine locations. However, for large projects whose total expected power output is greater than 10 MW or 2% of the total theoretical undisturbed resource, high-resolution modeling should be used, along with adequate model validation using measurement data and energy extraction simulations [27].

Clearly, to follow the IEC recommendations for tidal resource characterization and assessment, the tidal hydrodynamic model should be developed with a high resolution and be validated using field measurements, which could be challenging at times because of the need for high-performance computing resources and the lack of field measurements. A previous modeling study was carried out to evaluate the efficiency of energy extraction in the Western Passage in the U.S. state of Maine [28]. However, this model was validated with limited measured data, and the theoretical resources in the Western Passage were not assessed. Over the last decade, a number of numerical models have been developed to assess the amount of energy extraction by adding a momentum sink term in the governing equations based on the actuator-disc theory [7,17,28–31]. Numerical models that have energy extraction capabilities enable accurate and realistic resource characterization and assessment, as well as environmental interactions, at any desired project scale [32,33]. This study presents a modeling effort conducted to assess the tidal energy resources at a highly energetic site—the Western Passage [12,34]—using a high-resolution hydrodynamic model. This study followed the IEC recommendations for model configurations and model validation. The theoretical resources in the Western Passage were estimated based on the theory developed by Garrett and Cummins [5] and the model results. Finally, a sensitivity analysis with energy extraction from a tidal turbine farm was conducted to demonstrate the advantage of using numerical modeling to support tidal energy resource assessment by providing essential information.

2. Methods

2.1. Study Site

The Western Passage is an energetic tidal channel between New Brunswick, Canada, and the state of Maine in the United States (Figure 1). It is part of a large and complex coastal system, the Passamaquoddy–Cobscook Bay archipelago, which consists of many tidal channels and coastal bays, including Passamaquoddy Bay, Head Harbor Passage, Cobscook Bay, and South Bay. The Passamaquoddy–Cobscook Bay archipelago is connected to the mouth of the Bay of Fundy, which has one of the largest tidal ranges in the world [29]. Because of the large tidal range in the coastal bay system, many of the tidal channels show strong tidal currents and, therefore, hold great promise for tidal energy development [28,35]. The Western Passage has been identified as one of the top-ranked sites for tidal stream energy development in U.S. coastal waters, based on the initial nation-wide tidal resource assessment [12] and a number of criteria, including tidal power density, market value, energy price, shipping cost, and transmission distance [34]. Moreover, its relatively deep channel makes the Western Passage a favorable site for tidal power development.

J. Mar. Sci. Eng. 2020, 8, 411 3 of 21J. Mar. Sci. Eng. 2020, 8, x FOR PEER REVIEW 3 of 23

Figure 1. Map of the study site and locations of the observation stations.

Although the focus of this study was on the tidal hydrodynamics in the Passamaquoddy–

Cobscook Bay archipelago, to minimize the open-boundary effect on the area of interest, i.e., the

Western Passage and other tidal channels, and to better understand the tidal wave propagation from

the Gulf of Maine into the Bay of Fundy and the archipelago, the model domain was extended farther

out to cover the entire Bay of Fundy and the northern part of the Gulf of Maine. Figure 2. shows the

model domain and bathymetry distribution. Bathymetric data were obtained from various data

resources. Bathymetry in the large area of the Bay of Fundy and Gulf of Maine was interpolated from

the National Oceanic and Atmospheric Administration’s (NOAA’s) ETOPO1 1-arc-minute Global

Relief Model [36]. Inside the Passamaquoddy–Cobscook Bay archipelago, bathymetric data were

digitized from NOAA and Canadian Hydrographic Service navigation charts and supplemented

with multi-beam data for the Western Passage and Cobscook Bay [28]. Water depths in most of the

coastal areas are generally shallow and less than 100 m. The deepest area in the model domain is near

the open boundary and the southeast side of Grand Manan Island.

Figure 1. Map of the study site and locations of the observation stations.

Although the focus of this study was on the tidal hydrodynamics in the Passamaquoddy–Cobscook Bay archipelago, to minimize the open-boundary effect on the area of interest, i.e., the Western Passage and other tidal channels, and to better understand the tidal wave propagation from the Gulf of Maine into the Bay of Fundy and the archipelago, the model domain was extended farther out to cover the entire Bay of Fundy and the northern part of the Gulf of Maine. Figure 2. shows the model domain and bathymetry distribution. Bathymetric data were obtained from various data resources. Bathymetry in the large area of the Bay of Fundy and Gulf of Maine was interpolated from the National Oceanic and Atmospheric Administration’s (NOAA’s) ETOPO1 1-arc-minute Global Relief Model [36]. Inside the Passamaquoddy–Cobscook Bay archipelago, bathymetric data were digitized from NOAA and Canadian Hydrographic Service navigation charts and supplemented with multi-beam data for the Western Passage and Cobscook Bay [28]. Water depths in most of the coastal areas are generally shallow and less than 100 m. The deepest area in the model domain is near the open boundary and the southeast side of Grand Manan Island.

J. Mar. Sci. Eng. 2020, 8, 411 4 of 21

J. Mar. Sci. Eng. 2020, 8, x FOR PEER REVIEW 4 of 23

Figure 2. Model domain and bathymetry. The three blue circles represent the tide stations used for

model calibration of the water level.

2.2. Tidal Hydrodynamic Model

The unstructured-grid, finite-volume, community ocean model (FVCOM) [37,38] was used in

this study to simulate the tidal hydrodynamics in the Western Passage and its adjacent coastal waters,

including the Bay of Fundy. FVCOM solves the 3-D primitive Navier–Stokes equations to simulate

water surface elevation, velocity, salinity, temperature, and other transport constituents. FVCOM

employs a number of ocean turbulence closure models for the parameterization of the vertical eddy

viscosity and thermal diffusion coefficient. In the present study, the Mellor–Yamada level 2.5 (MY-

2.5) turbulent closure model was used [39]. The unstructured-grid modeling framework in FVCOM

is particularly suitable for resolving complex coastlines and providing accurate simulations with

great computational efficiency. FVCOM has been widely used around the world to simulate a wide

range of coastal problems, including coastal hydrodynamics [40–43], water quality and sediment

transport [44–47], storm surge and coastal flooding [48–52], and tidal energy [7,18,20,21,28].

To enable the assessment of future scenarios of proposed tidal turbine array development, a

numerical model that has energy extraction capabilities is required. To simulate tidal energy

extraction, a marine hydrokinetic (MHK) module was implemented in the FVCOM model based on

the momentum sink approach [7,28]. The FVCOM-MHK model was validated in an idealized tidal

channel and bay system against an analytical solution [7] and applied to simulate tidal energy

extraction and the associated environmental impacts [53–55]. The total extracted tidal power by a

tidal turbine farm at any location with a velocity �⃗� can be calculated using the following formula [7]:

𝑃𝑒𝑥𝑡 = ∑ (𝑁 × 1

2 𝜌𝐶𝑇𝐴𝑠|�⃗� |

3̅̅ ̅̅ ̅) 𝑖

𝑀

𝑖=1

(1)

Figure 2. Model domain and bathymetry. The three blue circles represent the tide stations used for model calibration of the water level.

2.2. Tidal Hydrodynamic Model

The unstructured-grid, finite-volume, community ocean model (FVCOM) [37,38] was used in this study to simulate the tidal hydrodynamics in the Western Passage and its adjacent coastal waters, including the Bay of Fundy. FVCOM solves the 3-D primitive Navier–Stokes equations to simulate water surface elevation, velocity, salinity, temperature, and other transport constituents. FVCOM employs a number of ocean turbulence closure models for the parameterization of the vertical eddy viscosity and thermal diffusion coefficient. In the present study, the Mellor–Yamada level 2.5 (MY-2.5) turbulent closure model was used [39]. The unstructured-grid modeling framework in FVCOM is particularly suitable for resolving complex coastlines and providing accurate simulations with great computational efficiency. FVCOM has been widely used around the world to simulate a wide range of coastal problems, including coastal hydrodynamics [40–43], water quality and sediment transport [44–47], storm surge and coastal flooding [48–52], and tidal energy [7,18,20,21,28].

To enable the assessment of future scenarios of proposed tidal turbine array development, a numerical model that has energy extraction capabilities is required. To simulate tidal energy extraction, a marine hydrokinetic (MHK) module was implemented in the FVCOM model based on the momentum sink approach [7,28]. The FVCOM-MHK model was validated in an idealized tidal channel and bay system against an analytical solution [7] and applied to simulate tidal energy extraction and the associated environmental impacts [53–55]. The total extracted tidal power by a tidal turbine farm at any location with a velocity

→ u can be calculated using the following formula [7]:

J. Mar. Sci. Eng. 2020, 8, 411 5 of 21

Pext = M∑

i=1

( N ×

1 2 ρCT As|

→ u |3

) i

(1)

where N is the number of turbines in each grid element, M is the total number of elements containing turbines, ρ is sea water density, CT is the turbine thrust coefficient, As is the flow-facing area swept by turbines, and

→ u is the velocity vector at the turbine hub height.

2.3. Model Configurations and Boundary Conditions

The Passamaquoddy–Cobscook Bay archipelago is strongly dominated by tidal forcing, with little influence of river discharge and wind-induced wave action [28]. There are two major rivers in the Passamaquoddy–Cobscook Bay archipelago system. The St. Croix River, which forms part of the United States–Canada border between Maine and New Brunswick, discharges into the Passamaquoddy Bay. The Dennys River discharges into the Dennis Bay in Maine. The long-term annual means of the streamflow are 21.24 m3/s and 5.6 m3/s for the St. Croix River and Dennys River, respectively, which are relatively small compared to the model domain [28]. Wind-driven circulation in the archipelago system is also small compared to the strong tidal current. Therefore, similar to [28], the sea surface wind, salinity, and temperature effects are not considered in this study.

The model grid for the Western Passage was initially developed by Rao et al. [28], and the model domain was further extended out to cover the entire Bay of Fundy and northern Gulf of Maine (Figure 2). According to IEC standards, a minimum 50 m grid resolution is required for a Stage 2 design layout study. To meet the IEC recommendations, the model grid resolution in this study varies from 20 m in the Western Passage to approximately 1000 m near the mouth of the Bay of Fundy and 2000 m along the open boundary in the northern Gulf of Maine. The model mesh consists of about 231,000 elements and 120,000 grid nodes. The model grid for the entire model domain is presented in Figure 3a, and Figure 3b shows a zoom-in grid for the Passamaquoddy–Cobscook Bay archipelago, where the Western Passage, Cobscook Bay, and Head Harbor Passage have the highest grid resolution of 20 m. Although stratification in the Passamaquoddy–Cobscook Bay archipelago is weak because the influence of river discharge is small, a depth-averaged modeling analysis is not accurate enough because tidal energy extraction is designed to occur at a specific water depth (hub height) of the water column. Therefore, all simulations in this study were run in 3-D mode. The model vertical resolution was specified to have 15 uniform sigma layers. Open-boundary conditions were specified with tidal elevation time histories generated using 13 tidal constituents obtained from the TPXO7.2 global ocean tide database [53]. The 13 tidal constituents were M2, S2, N2, K2, K1, O1, P1, Q1, M4, MS4, MN4, Mm, and Mf.

The wetting and drying process was simulated in the model domain with a wetting–drying criterion of 0.05 m. The model was run with an external-mode time step of 0.05 s, and a multiplier of 5 was used for the internal mode simulation. The initial hydrodynamic conditions were a null velocity and water level throughout the model domain. A one-week spin-up period was used such that the model reached dynamic equilibrium. Model simulations were conducted for three separate periods: July 2000 and September 2001 for model calibration and from April to June 2017 (three months) for model validation.

2.4. Observation Data for Model Calibration and Validation

Field measurements are important not only for site-specific tidal energy resource characterization but also for numerical model validation. Two types of data were used in this study to support model calibration and validation: water surface elevation and tidal current. Field data in the study area are very limited. There is only one NOAA real-time tidal gauge (8410140) in the model domain, located at Eastport, Maine. Therefore, additional water-level data were obtained from XTide predictions to support model calibration. Two XTide stations were selected from outside of the archipelago—at

J. Mar. Sci. Eng. 2020, 8, 411 6 of 21

Cutler, Maine, and Port Greville, Nova Scotia (Figure 2). Measurements of historical currents in the Passamaquoddy–Cobscook Bay archipelago are also very limited.J. Mar. Sci. Eng. 2020, 8, x FOR PEER REVIEW 6 of 23

Figure 3. (a) Model grid of the finite-volume community ocean model (FVCOM) for the model

domain that consists of the Passamaquoddy–Cobscook Bay archipelago (blue box), Bay of Fundy, and

northern Gulf of Maine; (b) zoomed-in model grid with bathymetry for the Passamaquoddy–

Cobscook Bay archipelago.

2.4. Observation Data for Model Calibration and Validation

Field measurements are important not only for site-specific tidal energy resource

characterization but also for numerical model validation. Two types of data were used in this study

to support model calibration and validation: water surface elevation and tidal current. Field data in

the study area are very limited. There is only one NOAA real-time tidal gauge (8410140) in the model

domain, located at Eastport, Maine. Therefore, additional water-level data were obtained from XTide

predictions to support model calibration. Two XTide stations were selected from outside of the

archipelago—at Cutler, Maine, and Port Greville, Nova Scotia (Figure 2). Measurements of historical

currents in the Passamaquoddy–Cobscook Bay archipelago are also very limited.

Historical current data collected by NOAA were available through the Currents Measurement

Interface for the Study of Tides, a web-based data management system maintained by NOAA to

analyze and disseminate current data from coastal and estuarine collections. Current measurements

were conducted by NOAA during the period of June to September 2000 at Estes Head (EP0003) and

St. Croix River (EP0004), both in Maine (Figure 1). Two bottom-mounted acoustic Doppler current

profilers (ADCPs) were deployed at depths of 34.1 m and 32 m at EP0003 and EP0004, respectively.

Current speed and direction data were obtained at 6 min intervals and a vertical bin resolution of 2

m. A historical current buoy J02 was deployed by the Northeastern Regional Association of Coastal

Ocean Observation Systems (NERACOOS) in Cobscook Bay for the period of September 2001 to June

2007 (Figure 1). Current data from EP0003, EP0004, and J02 were used for model calibration.

To better characterize the tidal hydrodynamics and turbulence characteristics, a more recent

survey was conducted by the National Renewable Energy Laboratory (NREL) at station WP1 in the

Western Passage (Figure 1). Two bottom-mounted ADCPs (a 600 kHz Workhorse Sentinel and a 500

kHz Nortek Signature) were deployed for a period of three months from April to June 2017. The

ADCP was deployed at a water depth of 45 m and configured at a sampling interval of 6 min and

vertical bin size of 1 m. The water level was also measured at the site during the ADCP deployment

period. The observed water-level and velocity data at the WP1 station were used for model

validation.

In this study, the water-level data at Eastport, Cutler, and Port Greville, as well as the current

data from NOAA and NERACOOS, were used for model calibration. The most recent data, for both

the water-level and velocity measurements collected by NREL at station WP1, were used for model

Figure 3. (a) Model grid of the finite-volume community ocean model (FVCOM) for the model domain that consists of the Passamaquoddy–Cobscook Bay archipelago (blue box), Bay of Fundy, and northern Gulf of Maine; (b) zoomed-in model grid with bathymetry for the Passamaquoddy–Cobscook Bay archipelago.

Historical current data collected by NOAA were available through the Currents Measurement Interface for the Study of Tides, a web-based data management system maintained by NOAA to analyze and disseminate current data from coastal and estuarine collections. Current measurements were conducted by NOAA during the period of June to September 2000 at Estes Head (EP0003) and St. Croix River (EP0004), both in Maine (Figure 1). Two bottom-mounted acoustic Doppler current profilers (ADCPs) were deployed at depths of 34.1 m and 32 m at EP0003 and EP0004, respectively. Current speed and direction data were obtained at 6 min intervals and a vertical bin resolution of 2 m. A historical current buoy J02 was deployed by the Northeastern Regional Association of Coastal Ocean Observation Systems (NERACOOS) in Cobscook Bay for the period of September 2001 to June 2007 (Figure 1). Current data from EP0003, EP0004, and J02 were used for model calibration.

To better characterize the tidal hydrodynamics and turbulence characteristics, a more recent survey was conducted by the National Renewable Energy Laboratory (NREL) at station WP1 in the Western Passage (Figure 1). Two bottom-mounted ADCPs (a 600 kHz Workhorse Sentinel and a 500 kHz Nortek Signature) were deployed for a period of three months from April to June 2017. The ADCP was deployed at a water depth of 45 m and configured at a sampling interval of 6 min and vertical bin size of 1 m. The water level was also measured at the site during the ADCP deployment period. The observed water-level and velocity data at the WP1 station were used for model validation.

In this study, the water-level data at Eastport, Cutler, and Port Greville, as well as the current data from NOAA and NERACOOS, were used for model calibration. The most recent data, for both the water-level and velocity measurements collected by NREL at station WP1, were used for model validation. Station information for all the measurements used for model calibration and validation is listed in Table 1.

J. Mar. Sci. Eng. 2020, 8, 411 7 of 21

Table 1. List of observed data stations for model calibration and validation.

Station Type Longitude Latitude Depth (m) Year

Eastport, ME Tide Gage −66.985 44.903 6.7 2000 Cutler, ME XTide −64.967 45.567 5.9 2000

Port Greville, NS XTide −64.550 45.400 11.8 2000 EP0003 ADCP −66.996 44.888 34.1 2000 EP0004 ADCP −67.101 45.076 32.0 2000

JO2 ADCP −67.017 44.891 32.0 2001 WP1 ADCP −66.989 44.920 45.0 2017

3. Results and Discussion

3.1. Model Calibration

To accurately assess the tidal energy resources at any specific site or the maximum tidal energy potential (i.e., a theoretical undisturbed resource) using a modeling approach, it is important to calibrate and validate the numerical model using field observations. In this study, model calibration was performed by comparing modeled tidal elevations and velocities with field measurements at various locations during different time periods. Three representative error statistical parameters, the root-mean-square-error (RMSE), the scattered index (SI), and the coefficient of determination (R2), were calculated to assess the model’s ability to reproduce the characteristics in the study domain. Model parameters, such as the bottom roughness, vertical layers, and open-boundary sponge layer (radius, and friction coefficient), were also adjusted iteratively during the calibration runs to achieve an overall satisfactory agreement between the model predictions and field observations. The final calibrated bottom roughness height was 0.005 m. The calibrated radius and friction coefficient of the open-boundary sponge layer were 1500 m and 0.001, respectively. Two model runs with 15 and 30 uniform sigma layers were conducted to evaluate the effect of the total number of vertical layers on model accuracy. The model results with 30 vertical layers showed little improvement in model performance error statistics over 15 vertical layers. Therefore, 15 vertical layers were selected in all the model runs to achieve better runtime efficiency while maintaining the same level of model accuracy.

The simulated water levels at Cutler, Eastport, and Port Greville were compared with the data for a 14-day period in October 2001 (Figure 4). Overall, the model reproduced the tide-level variations precisely in both the tidal range and phase in the Passamaquoddy–Cobscook Bay archipelago and Bay of Fundy. The water-level time history at all three stations indicated that the tide in the model domain is dominated by semi-diurnal tide (M2) with a strong signal of spring-neap tidal cycle. The tidal wave is amplified as it propagates from the Gulf of Maine (near the open boundary) (Figure 4a) into the Passamaquoddy–Cobscook Bay archipelago (Figure 4b) and toward the Bay of Fundy (Figure 4c). The root-mean-square errors (RMSEs) at the three tidal stations ranged between 0.15 m at Cutler station and 0.41 m at Port Greville station, where the tidal range is greater than 10 m during spring tide (Table 2). The correlation coefficients (R2) were persistently higher than 99% at all three stations (Table 2). The scatter indexes (SIs) were relatively small, indicating that the model was able to reproduce the water level accurately and consistently (Table 2).

Table 2. Error statistics of water-level predictions for model calibration.

Water Level Cutler Eastport Port Greville

RMSE (m) 0.15 0.26 0.41 SI 0.12 0.15 0.13 R2 0.99 0.99 0.99

The modeled depth-averaged principal velocities were compared with the measured data at stations EP0003 and EP0004 for July 2000 and at station J02 for October 2001 (Figure 5). Overall,

J. Mar. Sci. Eng. 2020, 8, 411 8 of 21

the model results matched the measurements well, especially during flood tides. However, the model overpredicted the ebb currents at EP0003, where the measurements showed tidal current asymmetry with flood currents larger than ebb currents (Figure 5a). The error statistics for current predictions at EP0003, EP0004, and J02 are presented in Table 3. The largest RMSE is 0.38 m/s at station EP0003, which is consistent with the time–series plot shown in Figure 5a. The RMSEs at EP0004 and J02 are 0.14 m/s and 0.27 m/s, respectively, which are satisfactory for the current predictions. The correlation coefficients (R2) for the current predictions are all above 0.94, indicating that the model predictions are highly correlated to the measurements at all three stations. J. Mar. Sci. Eng. 2020, 8, x FOR PEER REVIEW 8 of 23

Figure 4. Comparison of the modeled and observed water levels at (a) Cutler, ME; (b) Eastport, ME;

and (c) Port Greville, NS. Station locations are shown in Figure 1 and 2.

Table 2. Error statistics of water-level predictions for model calibration.

Water Level Cutler Eastport Port Greville

RMSE (m) 0.15 0.26 0.41

SI 0.12 0.15 0.13

R2 0.99 0.99 0.99

The modeled depth-averaged principal velocities were compared with the measured data at

stations EP0003 and EP0004 for July 2000 and at station J02 for October 2001 (Figure 5). Overall, the

model results matched the measurements well, especially during flood tides. However, the model

overpredicted the ebb currents at EP0003, where the measurements showed tidal current asymmetry

with flood currents larger than ebb currents (Figure 5a). The error statistics for current predictions at

EP0003, EP0004, and J02 are presented in Table 3. The largest RMSE is 0.38 m/s at station EP0003,

which is consistent with the time–series plot shown in Figure 5a. The RMSEs at EP0004 and J02 are

0.14 m/s and 0.27 m/s, respectively, which are satisfactory for the current predictions. The correlation

coefficients (R2) for the current predictions are all above 0.94, indicating that the model predictions

are highly correlated to the measurements at all three stations.

Figure 4. Comparison of the modeled and observed water levels at (a) Cutler, ME; (b) Eastport, ME; and (c) Port Greville, NS. Station locations are shown in Figures 1 and 2.

J. Mar. Sci. Eng. 2020, 8, x FOR PEER REVIEW 9 of 23

Figure 5. Comparison of the modeled and observed depth-averaged tidal currents (principal

component) at (a) EP0003, (b) EP0004, and (c) J02. A positive value denotes the flood current. Station

locations are shown in Figure 1.

Table 3. Error statistics of the current predictions at stations EP0003, EP0004, and J02.

Depth-Average Velocity EP0003 EP0004 J02

RMSE (m/s) 0.38 0.14 0.27

SI 0.42 0.41 0.33

R 2 0.97 0.94 0.96

3.2. Model Validation

Once the model was calibrated, model validation was conducted for a different simulation

period using the calibrated model parameters. The recent observed data, including water level and

velocity collected at station WP1 in the Western Passage, were used for model validation. The model

was run for three months, corresponding to the measurement period of April to June 2017.

The long (three-month) record of field measurements at WP1 allowed for accurate harmonic

analysis. Therefore, the model skills for predicting water level and currents at WP1 were evaluated

by comparing the tidal level harmonic constituents derived from the measurements and model

results. Water levels were decomposed into 10 tidal components (M2, N2, S2, K1, O1, P1, Q1, M4, M6,

and MK3) using harmonic analysis. Comparisons of the constituents of each observed and predicted

tidal level are provided in Table 4. The maximum difference between modeled and measured tidal

constituents is −0.11 m for the principal lunar semi-diurnal tide M2, which is about 3.9%

underpredicted relative to the observed M2 constitute (2.62 m). Differences between the model

predictions and measurements for other tidal constituents are all small; even the percentage error is

high because of the small magnitudes of the constituents. Therefore, the model performed very well

in simulating the tidal elevation as part of the model validation.

Table 4. Comparison of the observed and modeled water level at WP1 in the Western Passage.

Water Level (m) M2 N2 S2 K1 O1 P1 Q1 M6 MK3 MS4

Data (WP1) 2.72 0.38 0.54 0.09 0.14 0.06 0.06 0.15 0.02 0.18

Model 2.62 0.38 0.52 0.08 0.15 0.06 0.02 0.17 0.02 0.09

Figure 5. Comparison of the modeled and observed depth-averaged tidal currents (principal component) at (a) EP0003, (b) EP0004, and (c) J02. A positive value denotes the flood current. Station locations are shown in Figure 1.

J. Mar. Sci. Eng. 2020, 8, 411 9 of 21

Table 3. Error statistics of the current predictions at stations EP0003, EP0004, and J02.

Depth-Average Velocity EP0003 EP0004 J02

RMSE (m/s) 0.38 0.14 0.27 SI 0.42 0.41 0.33 R2 0.97 0.94 0.96

3.2. Model Validation

Once the model was calibrated, model validation was conducted for a different simulation period using the calibrated model parameters. The recent observed data, including water level and velocity collected at station WP1 in the Western Passage, were used for model validation. The model was run for three months, corresponding to the measurement period of April to June 2017.

The long (three-month) record of field measurements at WP1 allowed for accurate harmonic analysis. Therefore, the model skills for predicting water level and currents at WP1 were evaluated by comparing the tidal level harmonic constituents derived from the measurements and model results. Water levels were decomposed into 10 tidal components (M2, N2, S2, K1, O1, P1, Q1, M4, M6, and MK3) using harmonic analysis. Comparisons of the constituents of each observed and predicted tidal level are provided in Table 4. The maximum difference between modeled and measured tidal constituents is −0.11 m for the principal lunar semi-diurnal tide M2, which is about 3.9% underpredicted relative to the observed M2 constitute (2.62 m). Differences between the model predictions and measurements for other tidal constituents are all small; even the percentage error is high because of the small magnitudes of the constituents. Therefore, the model performed very well in simulating the tidal elevation as part of the model validation.

Table 4. Comparison of the observed and modeled water level at WP1 in the Western Passage.

Water Level (m) M2 N2 S2 K1 O1 P1 Q1 M6 MK3 MS4

Data (WP1) 2.72 0.38 0.54 0.09 0.14 0.06 0.06 0.15 0.02 0.18 Model 2.62 0.38 0.52 0.08 0.15 0.06 0.02 0.17 0.02 0.09

Difference −0.11 0.01 −0.03 −0.01 0.01 0.01 −0.04 0.03 0.00 −0.09 Percentage Error 3.9 1.5 4.8 10.9 5.4 12.8 66.9 18.7 3.1 53.3

To better visualize a comparison between the modeled and observed velocity through the water column, velocity contours were generated with respect to water depth and time (Figure 6). The distribution pattern of the modeled velocity contours for the u (east) and v (north) components (Figure 6c,d) is very similar to the observations (Figure 6a,b). Both model results and observations show consistently strong currents through the entire water column, with distinct flood and ebb tidal signals. In general, both the model results and the observed data showed a stronger flood current (negative u and positive v) than an ebb current (positive u and negative v). The velocity magnitudes were mostly uniformly distributed through the water column, except for the modeled v-velocity, which showed a slightly weaker current near the surface during the flood tide (positive value) (Figure 6d).

Figure 7 shows a comparison of the vertical profiles of the velocity percentiles based on the three-month model results and observed data. Velocity percentiles were calculated as 10%, 25%, 50%, 75%, and 90%. Overall, the model results are similar to the observations in terms of the range of the velocity percentiles. Strong tidal asymmetry in the velocity profiles is seen in both the modeled results and the observations, where currents are stronger during flood tides (negative value) and weaker during ebb tides (positive value) (Figure 7c,f). The mean error of the predicted mean velocity-magnitude profile is −0.134 m/s, where the negative sign indicates model underprediction, and the percentage error is 11%. Although the model showed good overall skill in predicting the velocity vertical profiles in the Western Passage, it overpredicted the v-velocity component during flood tides (negative value) and underpredicted it during ebb tides (positive value) (Figure 7b,e). In particular, the model predicted

J. Mar. Sci. Eng. 2020, 8, 411 10 of 21

a strong ebb current near the bottom (Figure 7e), which was not shown in the observed data (Figure 7b). The cause of such a discrepancy between the model results and observed data is likely due to the accuracy of model bathymetry, which was mainly interpolated from bathymetric datasets digitized from old NOAA navigation charts [28] that may not accurately represent the real bathymetry in the Western Passage when the ADCP data were collected in 2017.

J. Mar. Sci. Eng. 2020, 8, x FOR PEER REVIEW 10 of 23

Difference −0.11 0.01 −0.03 −0.01 0.01 0.01 −0.04 0.03 0.00 −0.09

Percentage Error 3.9 1.5 4.8 10.9 5.4 12.8 66.9 18.7 3.1 53.3

To better visualize a comparison between the modeled and observed velocity through the water

column, velocity contours were generated with respect to water depth and time (Figure 6). The

distribution pattern of the modeled velocity contours for the u (east) and v (north) components

(Figure 6c,d) is very similar to the observations (Figure 6a,b). Both model results and observations

show consistently strong currents through the entire water column, with distinct flood and ebb tidal

signals. In general, both the model results and the observed data showed a stronger flood current

(negative u and positive v) than an ebb current (positive u and negative v). The velocity magnitudes

were mostly uniformly distributed through the water column, except for the modeled v-velocity,

which showed a slightly weaker current near the surface during the flood tide (positive value) (Figure

6d).

Figure 6. Comparison of the observed and modeled velocity time series through the water column at

station WP1 in the Western Passage. (a,b) The observed east (u) and north (v) velocity components;

(c,d) the modeled east (u) and north (v) velocity components.

Figure 7 shows a comparison of the vertical profiles of the velocity percentiles based on the three-

month model results and observed data. Velocity percentiles were calculated as 10%, 25%, 50%, 75%,

and 90%. Overall, the model results are similar to the observations in terms of the range of the velocity

percentiles. Strong tidal asymmetry in the velocity profiles is seen in both the modeled results and

the observations, where currents are stronger during flood tides (negative value) and weaker during

ebb tides (positive value) (Figure 7c,f). The mean error of the predicted mean velocity-magnitude

profile is −0.134 m/s, where the negative sign indicates model underprediction, and the percentage

error is 11%. Although the model showed good overall skill in predicting the velocity vertical profiles

in the Western Passage, it overpredicted the v-velocity component during flood tides (negative value)

and underpredicted it during ebb tides (positive value) (Figure 7b,e). In particular, the model

Figure 6. Comparison of the observed and modeled velocity time series through the water column at station WP1 in the Western Passage. (a,b) The observed east (u) and north (v) velocity components; (c,d) the modeled east (u) and north (v) velocity components.

The error statistics calculated at WP1 show the model’s excellent ability to simulate a tidal current (Table 5). Similar to the water-level results, the model underpredicted the M2 tidal current by 0.21 m/s, which is 12.4% of the M2 current speed. All other predicted tidal constituents are no more than 0.07 m/s smaller than the observed tidal current constituents. The model validation results in this study are comparable to those presented in [28]. Because the tide is the dominant force in the system, the model is considered fully capable of reproducing tidal hydrodynamics and characterizing the tidal energy resources in the Western Passage.

Table 5. Comparison of the observed and modeled tidal current constituents at WP1 in the Western Passage.

Current (m/s) M2 N2 S2 K1 O1 P1 Q1 M6 MK3 MS4

Data (WP1) 1.67 0.25 0.34 0.07 0.04 0.21 0.11 0.07 0.06 0.12 Model 1.46 0.23 0.31 0.04 0.04 0.20 0.04 0.06 0.04 0.03

Difference −0.21 −0.02 −0.03 −0.02 −0.01 −0.01 −0.07 −0.01 −0.02 −0.09 Percentage Error 12.4 8.5 7.8 32.1 13.9 3.9 62.9 13.2 26.6 71.9

J. Mar. Sci. Eng. 2020, 8, 411 11 of 21

J. Mar. Sci. Eng. 2020, 8, x FOR PEER REVIEW 11 of 23

predicted a strong ebb current near the bottom (Figure 7e), which was not shown in the observed

data (Figure 7b). The cause of such a discrepancy between the model results and observed data is

likely due to the accuracy of model bathymetry, which was mainly interpolated from bathymetric

datasets digitized from old NOAA navigation charts [28] that may not accurately represent the real

bathymetry in the Western Passage when the ADCP data were collected in 2017.

Figure 7. Comparison of the observed and modeled vertical profiles of velocity percentiles at station

WP1 in the Western Passage based on three months of records: (a,b) the observed east (u) and north

(v) velocity components; (d,e) the modeled east (u) and north (v) velocity components; (c,f) the

observed and modeled velocity magnitudes, with positive as the ebb current and negative as the flood

current. Velocity percentiles were calculated as 10%, 25%, 50%, 75%, and 90%.

The error statistics calculated at WP1 show the model’s excellent ability to simulate a tidal

current (Table 5). Similar to the water-level results, the model underpredicted the M2 tidal current

by 0.21 m/s, which is 12.4% of the M2 current speed. All other predicted tidal constituents are no

more than 0.07 m/s smaller than the observed tidal current constituents. The model validation results

in this study are comparable to those presented in [28]. Because the tide is the dominant force in the

system, the model is considered fully capable of reproducing tidal hydrodynamics and characterizing

the tidal energy resources in the Western Passage.

Table 5. Comparison of the observed and modeled tidal current constituents at WP1 in the Western

Passage.

Current (m/s) M2 N2 S2 K1 O1 P1 Q1 M6 MK3 MS4

Data (WP1) 1.67 0.25 0.34 0.07 0.04 0.21 0.11 0.07 0.06 0.12

Model 1.46 0.23 0.31 0.04 0.04 0.20 0.04 0.06 0.04 0.03

Difference −0.21 −0.02 −0.03 −0.02 −0.01 −0.01 −0.07 −0.01 −0.02 −0.09

Percentage Error 12.4 8.5 7.8 32.1 13.9 3.9 62.9 13.2 26.6 71.9

3.3. Characteristics of Tidal Hydrodynamics

Once the model was calibrated and validated, the model results could be used to characterize

the resource and identify hotspots for tidal energy development in the region. Figure 8 shows the

horizontal distribution of the depth-averaged velocities during peak flood and ebb tides. Clearly,

strong currents exist along the main channel of the Western Passage, Cobscook Bay, and Head Harbor

Figure 7. Comparison of the observed and modeled vertical profiles of velocity percentiles at station WP1 in the Western Passage based on three months of records: (a,b) the observed east (u) and north (v) velocity components; (d,e) the modeled east (u) and north (v) velocity components; (c,f) the observed and modeled velocity magnitudes, with positive as the ebb current and negative as the flood current. Velocity percentiles were calculated as 10%, 25%, 50%, 75%, and 90%.

3.3. Characteristics of Tidal Hydrodynamics

Once the model was calibrated and validated, the model results could be used to characterize the resource and identify hotspots for tidal energy development in the region. Figure 8 shows the horizontal distribution of the depth-averaged velocities during peak flood and ebb tides. Clearly, strong currents exist along the main channel of the Western Passage, Cobscook Bay, and Head Harbor Passage. In general, these areas with strong currents during both the flood and ebb tides represent potential hotspots for future tidal energy development. However, tidal asymmetry also exists in some areas between the flood and ebb tides. For example, in the tidal channel between Deer Island and Indian Island, the tidal currents are much stronger during the flood tide (Figure 8b) than those during the ebb tide (Figure 8a). On the other hand, the ebb currents (Figure 8b) are significantly stronger than the flood currents in the Head Harbor Passage (Figure 8a).

Because tidal turbines are installed in the water column to assess the feasibility of a project site for tidal energy extraction, the vertical structure of the current magnitude and the water depth distribution should be considered. Based on the results of the depth-averaged velocity distribution (Figure 8), the current magnitudes along the four cross sections with strong tidal currents were generated. Cross sections XS1 and XS2 are located in the Western Passage, cross section XS3 is in the Cobscook Bay, and cross section XS4 is in the Head Harbor Passage. Areas where strong currents occur only during flood or ebb tides, such as the south part of the Head Harbor Passage and the channel between Deer Island and Indian Island, were not selected for cross-sectional velocity analysis.

Figure 9 shows the normal velocity magnitudes along XS1 in the south end of the Western Passage. The water depth in the deep channel is about 120 m. A positive value indicates the flood current flowing away from the reader (Figure 9a). Strong flood and ebb currents occupy most of the deep channel and the U.S. side of the passage (left side) through the entire water column; the maximum current magnitude is greater than 2.5 m/s during the flood tide. The current distribution on the north end of the cross section is complicated; consistent ebb currents (negative value) occur during both the flood and ebb tides, and reverse flow (positive value) occurs at the bottom layer of the water column

J. Mar. Sci. Eng. 2020, 8, 411 12 of 21

during the ebb tide. The consistently strong tidal currents in the deep channel and the U.S. side of the passage indicate that the area is a good candidate for tidal turbine farm deployment.

J. Mar. Sci. Eng. 2020, 8, x FOR PEER REVIEW 12 of 23

Passage. In general, these areas with strong currents during both the flood and ebb tides represent

potential hotspots for future tidal energy development. However, tidal asymmetry also exists in some

areas between the flood and ebb tides. For example, in the tidal channel between Deer Island and

Indian Island, the tidal currents are much stronger during the flood tide (Figure 8b) than those during

the ebb tide (Figure 8a). On the other hand, the ebb currents (Figure 8b) are significantly stronger

than the flood currents in the Head Harbor Passage (Figure 8a).

Because tidal turbines are installed in the water column to assess the feasibility of a project site

for tidal energy extraction, the vertical structure of the current magnitude and the water depth

distribution should be considered. Based on the results of the depth-averaged velocity distribution

(Figure 8), the current magnitudes along the four cross sections with strong tidal currents were

generated. Cross sections XS1 and XS2 are located in the Western Passage, cross section XS3 is in the

Cobscook Bay, and cross section XS4 is in the Head Harbor Passage. Areas where strong currents

occur only during flood or ebb tides, such as the south part of the Head Harbor Passage and the

channel between Deer Island and Indian Island, were not selected for cross-sectional velocity

analysis.

Figure 8. Depth-averaged velocity magnitudes in the Western Passage, Cobscook Bay, and Head

Harbor Passage during (a) peak flood and (b) peak ebb tides in spring tide on April 28, 2017.

Figure 9 shows the normal velocity magnitudes along XS1 in the south end of the Western

Passage. The water depth in the deep channel is about 120 m. A positive value indicates the flood

Figure 8. Depth-averaged velocity magnitudes in the Western Passage, Cobscook Bay, and Head Harbor Passage during (a) peak flood and (b) peak ebb tides in spring tide on April 28, 2017.

Strong currents were observed in XS2 (Figure 10), especially on the U.S. side (the left side of the plot). The current speed in the main channel is approximately 1.75 m/s during both the flood and ebb tidal phases. The maximum current speed (>2 m/s) occurs in the U.S. side of the passage where the water depth is relatively shallow—about 30 m. Again, the current structure on the Canadian side is small and complicated, showing strong vertical shear during the flood tide (Figure 10a).

J. Mar. Sci. Eng. 2020, 8, 411 13 of 21

J. Mar. Sci. Eng. 2020, 8, x FOR PEER REVIEW 13 of 23

current flowing away from the reader (Figure 9a). Strong flood and ebb currents occupy most of the

deep channel and the U.S. side of the passage (left side) through the entire water column; the

maximum current magnitude is greater than 2.5 m/s during the flood tide. The current distribution

on the north end of the cross section is complicated; consistent ebb currents (negative value) occur

during both the flood and ebb tides, and reverse flow (positive value) occurs at the bottom layer of

the water column during the ebb tide. The consistently strong tidal currents in the deep channel and

the U.S. side of the passage indicate that the area is a good candidate for tidal turbine farm

deployment.

Strong currents were observed in XS2 (Figure 10), especially on the U.S. side (the left side of the

plot). The current speed in the main channel is approximately 1.75 m/s during both the flood and ebb

tidal phases. The maximum current speed (>2 m/s) occurs in the U.S. side of the passage where the

water depth is relatively shallow—about 30 m. Again, the current structure on the Canadian side is

small and complicated, showing strong vertical shear during the flood tide (Figure 10a).

Figure 9. Normal velocity magnitudes at (a) peak flood and (b) peak ebb tides along cross section XS1

in the Western Passage. Positive velocity is away from the reader. The cross-section location is shown

in Figure 1.

Figure 9. Normal velocity magnitudes at (a) peak flood and (b) peak ebb tides along cross section XS1 in the Western Passage. Positive velocity is away from the reader. The cross-section location is shown in Figure 1.

J. Mar. Sci. Eng. 2020, 8, x FOR PEER REVIEW 14 of 23

Figure 10. Normal velocity magnitudes at (a) peak flood and (b) peak ebb tides along cross section

XS2 in the Western Passage. Positive velocity is away from the reader. The cross-section location is

shown in Figure 1.

Figure 11 shows the current distribution along XS3 in Cobscook Bay. Although strong currents

(over 2.5 m/s) are present during both the flood and ebb tides in the main channel, the water depth

is very shallow. Most of the cross section is less than 10 m deep, and the deepest water depth is 25 m,

which makes the area challenging for the deployment of tidal turbine farms.

Figure 10. Normal velocity magnitudes at (a) peak flood and (b) peak ebb tides along cross section XS2 in the Western Passage. Positive velocity is away from the reader. The cross-section location is shown in Figure 1.

J. Mar. Sci. Eng. 2020, 8, 411 14 of 21

Figure 11 shows the current distribution along XS3 in Cobscook Bay. Although strong currents (over 2.5 m/s) are present during both the flood and ebb tides in the main channel, the water depth is very shallow. Most of the cross section is less than 10 m deep, and the deepest water depth is 25 m, which makes the area challenging for the deployment of tidal turbine farms.

J. Mar. Sci. Eng. 2020, 8, x FOR PEER REVIEW  15 of 23 

 

Figure 11. Normal velocity magnitudes at (a) peak flood and (b) peak ebb tides along cross section 

XS3 in Cobscook Bay. Positive velocity is away from the reader. The cross‐section location is shown 

in Figure 1. 

XS4  in  the Head Harbor Passage  (Canada) shows  the strongest current magnitude and  the 

largest area of high velocity among the four cross sections. The current speeds in most parts of this 

cross section are greater than 1.5 m/s during the flood tide (Figure 12a) and greater than 2 m/s during 

the ebb tide (Figure 12b). The seabed in the cross section is relatively flat, but the water depth is 

shallow (approximately 30 m deep for most of the cross section). Again, this shallow water depth 

may pose a challenge for the installation of tidal turbine farms, unless small tidal turbines (e.g., rotor 

diameters less than 15 m) are considered. 

Figure 11. Normal velocity magnitudes at (a) peak flood and (b) peak ebb tides along cross section XS3 in Cobscook Bay. Positive velocity is away from the reader. The cross-section location is shown in Figure 1.

XS4 in the Head Harbor Passage (Canada) shows the strongest current magnitude and the largest area of high velocity among the four cross sections. The current speeds in most parts of this cross section are greater than 1.5 m/s during the flood tide (Figure 12a) and greater than 2 m/s during the ebb tide (Figure 12b). The seabed in the cross section is relatively flat, but the water depth is shallow (approximately 30 m deep for most of the cross section). Again, this shallow water depth may pose a challenge for the installation of tidal turbine farms, unless small tidal turbines (e.g., rotor diameters less than 15 m) are considered.

J. Mar. Sci. Eng. 2020, 8, 411 15 of 21

J. Mar. Sci. Eng. 2020, 8, x FOR PEER REVIEW  16 of 23 

 

Figure 12. Normal velocity magnitudes at (a) the peak flood and (b) peak ebb tides along cross section 

XS4 in Head Harbor Passage. Positive velocity is toward the reader. The cross‐section location is 

shown in Figure 1. 

3.4. Along‐Channel Kinetic Energy Flux 

The predicted tidal currents within the potential deployment area can be used to estimate the 

tidal kinetic‐energy flux through a channel cross section. First, the temporal‐averaged power density 

at any location can be calculated using the predicted tidal currents as follows: 

𝑃 1

1000 ∙

1 𝑁

1 2

𝜌𝑈   (2) 

where  𝑃   is the temporal‐averaged power density (kW/m2) at any location or grid cell in the model,  N is the total number of velocity output timesteps over the simulation period,  𝜌  is the seawater  density (1025 kg/m3), and U is the current magnitude normal to the cross section (m/s). 

Figure 13 shows the distribution of the mean power density at each of the four cross sections 

(XS1–XS4). XS1 maintains around 1.0 kW/m2, which is evenly distributed in the middle–bottom layers 

of the channel that are deep enough for device deployment (Figure 13a). XS2 and XS3 show a range 

of 1–2 kW/m2 power density available near the surface (Figure 13b,c). The highest power density 

value, up to 3.0 kW/m2, was identified at XS4, which presented the strongest current at a magnitude 

of 2.5 m/s (Figure 13d). 

The mean tidal kinetic‐energy flux Pxs across a cross section can be estimated by multiplying the 

mean power density  𝑃   with the grid cell area and integrating it over the entire cross section using  the following formula: 

𝑃 ∑ 𝑃 𝐴     (3) 

where Ncell is the total number of model grid cells projected along the cross section, and Acell is the 

projected area for each grid cell. Based on Equation (3), the tidal kinetic energy flux Pxs for each of the 

four cross sections was estimated to be 45.8 × 103 (kW) for XS1, 22.3 × 103 (kW) for XS2, 9.03 × 103 (kW) 

for XS2, and 48.6 × 103 (kW) for XS4, respectively. 

Figure 12. Normal velocity magnitudes at (a) the peak flood and (b) peak ebb tides along cross section XS4 in Head Harbor Passage. Positive velocity is toward the reader. The cross-section location is shown in Figure 1.

3.4. Along-Channel Kinetic Energy Flux

The predicted tidal currents within the potential deployment area can be used to estimate the tidal kinetic-energy flux through a channel cross section. First, the temporal-averaged power density at any location can be calculated using the predicted tidal currents as follows:

Pw = 1

1000 ·

1 N

i=N∑ i=1

1 2 ρU3 (2)

where Pw is the temporal-averaged power density (kW/m2) at any location or grid cell in the model, N is the total number of velocity output timesteps over the simulation period, ρ is the seawater density (1025 kg/m3), and U is the current magnitude normal to the cross section (m/s).

Figure 13 shows the distribution of the mean power density at each of the four cross sections (XS1–XS4). XS1 maintains around 1.0 kW/m2, which is evenly distributed in the middle–bottom layers of the channel that are deep enough for device deployment (Figure 13a). XS2 and XS3 show a range of 1–2 kW/m2 power density available near the surface (Figure 13b,c). The highest power density value, up to 3.0 kW/m2, was identified at XS4, which presented the strongest current at a magnitude of 2.5 m/s (Figure 13d).

The mean tidal kinetic-energy flux Pxs across a cross section can be estimated by multiplying the mean power density Pw with the grid cell area and integrating it over the entire cross section using the following formula:

Pxs = ∑Ncell

j=1

( PwAcell

) (3)

J. Mar. Sci. Eng. 2020, 8, 411 16 of 21

where Ncell is the total number of model grid cells projected along the cross section, and Acell is the projected area for each grid cell. Based on Equation (3), the tidal kinetic energy flux Pxs for each of the four cross sections was estimated to be 45.8 × 103 (kW) for XS1, 22.3 × 103 (kW) for XS2, 9.03 × 103 (kW) for XS2, and 48.6 × 103 (kW) for XS4, respectively.J. Mar. Sci. Eng. 2020, 8, x FOR PEER REVIEW 17 of 23

Figure 13. Predicted tidal power density along cross sections (a) XS1 and (b) XS2 in the Western

Passage, (c) XS3 in Cobscook Bay, and (d) XS4 in Head Harbor Passage. The locations of the cross

sections are shown in Figure 1.

3.5. Energy Extraction in the Western Passage

As described in Section 3.4, the tidal energy fluxes in the Western Passage (XS1) and the Head

Harbor Passage (XS4) are much greater than those in XS2 and XS3. However, compared to XS4 in the

Head Harbor Passage, the Western Passage (XS1) is more favorable for tidal energy deployment

because of its greater water depth. In this section, the tidal energy extraction capacity in the Western

Passage was evaluated through a theoretical analysis and the use of a hypothetical tidal turbine farm.

The maximum theoretical extractable tidal power can be estimated based on the formula developed

by Garrett and Cummins [5] for a tidal channel connected to two basins:

𝑃𝑚𝑎𝑥 = 𝛾𝜌𝑔𝑎0𝑄𝑚𝑎𝑥 [1 + 9

16 (∑ (

𝑎𝑖 𝑎0

) 2𝑀𝑡

𝑖=1 )] (4)

where γ is a dimensionless constant in the range of 0.21 to 0.24, 𝑎0 is the largest tidal constituent

amplitude (M2) in the channel, and 𝑎𝑖 (𝑖 = 1, 2, … , 𝑀𝑡) are the additional Mt tidal constituent

amplitudes. Specifying γ = 0.22, 𝜌 = 1025 kg/m3, g = 9.81 m/s2, Qmax = 74,388 m3/s, and 𝑎𝑖 (𝑖 =

1, 2, … , 9) from Table 4 into Equation (4) yields Pmax = 447,825 (kW). The Pmax value can be used as a

reference (upper limit) for the development of tidal energy projects in the Western Passage. It should

be noted that Equation (4) likely overestimates the Pmax value due to the simplification and

assumptions made in the derivation of the equation [5].

Figure 13. Predicted tidal power density along cross sections (a) XS1 and (b) XS2 in the Western Passage, (c) XS3 in Cobscook Bay, and (d) XS4 in Head Harbor Passage. The locations of the cross sections are shown in Figure 1.

3.5. Energy Extraction in the Western Passage

As described in Section 3.4, the tidal energy fluxes in the Western Passage (XS1) and the Head Harbor Passage (XS4) are much greater than those in XS2 and XS3. However, compared to XS4 in the Head Harbor Passage, the Western Passage (XS1) is more favorable for tidal energy deployment because of its greater water depth. In this section, the tidal energy extraction capacity in the Western Passage was evaluated through a theoretical analysis and the use of a hypothetical tidal turbine farm. The maximum theoretical extractable tidal power can be estimated based on the formula developed by Garrett and Cummins [5] for a tidal channel connected to two basins:

Pmax = γρga0Qmax

1 + 916 ∑Mti=1

( ai a0

)2  (4)

J. Mar. Sci. Eng. 2020, 8, 411 17 of 21

where γ is a dimensionless constant in the range of 0.21 to 0.24, a0 is the largest tidal constituent amplitude (M2) in the channel, and ai (i = 1, 2, . . . , Mt) are the additional Mt tidal constituent amplitudes. Specifying γ = 0.22, ρ = 1025 kg/m3, g = 9.81 m/s2, Qmax = 74,388 m3/s, and ai (i = 1, 2, . . . , 9) from Table 4 into Equation (4) yields Pmax = 447,825 (kW). The Pmax value can be used as a reference (upper limit) for the development of tidal energy projects in the Western Passage. It should be noted that Equation (4) likely overestimates the Pmax value due to the simplification and assumptions made in the derivation of the equation [5].

A hypothetical tidal turbine farm was considered to simulate the energy extraction at a realistic scale in the Western Passage. The tidal farm is located at the south end and U.S. side of the Western Passage, where strong tidal currents occur. The tidal farm consists of a total of 19 tidal turbines, with along-channel spacing of 160 m and cross-channel spacing of 80 m (Figure 14). The tidal turbine hub height and diameter are 15 m and 20 m, respectively. The configuration of the tidal turbine farm is listed in Table 6.

The tidal energy extraction was simulated using the FVCOM-MHK model [7] based on Equation (1). The extractable energy at any specific site depends on not only the current speed but also the thrust coefficient, which is a function of the turbine design and flow speed for characterizing power extraction efficiency. Many far-field modeling studies showed that a turbine can potentially reach peak efficiency for extracting the maximum power, i.e., 59% of the available power in a system (called the Betz Limit) when the thrust coefficient is specified to be in the range of 0.8–1 [30,33,54–56]. Based on the values reported in the previous studies, a thrust coefficient of 0.9 was chosen in this study. For simplicity, the cut-in and cut-out velocities were not considered. The simulated temporally averaged extracted energy was 4810 (kW) for the tidal farm or 253 (kW) per turbine. Because the theoretical resources in the Western Passage are estimated to be 447,825 (kW), the extracted energy by the hypothetical tidal turbine farm is 1.07% of the theoretical resources, which is below the 2% threshold specified in the IEC standards [27].

Another method recommended in the IEC TS for tidal energy resource assessment is to estimate the extractable power of a tidal farm using measured or modeled undisturbed flow when the expected extractable power is smaller than 10 MW or 2% of the theoretical resource in the system. For the purpose of comparison, the extractable power was also estimated using Equation (1) and the modeled undisturbed velocity. All parameters in the calculation were kept the same as those specified in the FVCOM-MHK simulations, including the thrust coefficient (0.9), turbine blade diameter (20 m) in Equation (1), and undisturbed velocity at the turbine hub height (15 m from the seabed) at the 19 turbine locations. The estimated mean power extracted by the tidal farm based on the modeled undisturbed flow field was 5198 (kW), which is 1.2% of the theoretical resources. The extractable power estimated based on the FVCOM-MHK model run with energy extraction is slightly smaller than that directly calculated using undisturbed flow, which is expected because of the flow reduction caused by the turbines and the interactions among the turbines in the tidal farm in the FVCOM-MHK simulations.

Clearly, by doubling the number of tidal turbines in the existing tidal farm, i.e., increasing the turbine distribution density twofold, the estimated extractable power using modeled undisturbed flow will likely exceed the 10 MW and the 2% thresholds. However, the extractable power for the same tidal farm calculated using the FVCOM-MHK model with energy extraction may not exceed both thresholds because the efficiency of energy extraction decreases as the number of turbines increases in the tidal farm [7,22]. Therefore, it is better to use the numerical model with energy extraction to accurately estimate the extractable energy, especially when the energy extraction by the tidal farm is expected to exceed 10 MW or 2% of the theoretical resources.

Table 6. Tidal turbine configuration and simulated extractable power in the Western Passage.

Total Turbines

Turbine Spacing

Hub Height

Turbine Diameter

Avg. Extracted Power (kW)

Power per Turbine (kW)

19 80 m × 160 m 15 m 20 m 4810 253

J. Mar. Sci. Eng. 2020, 8, 411 18 of 21

J. Mar. Sci. Eng. 2020, 8, x FOR PEER REVIEW 19 of 23

Figure 14. Layout of the hypothetical tidal turbine farm in the Western Passage.

Table 6. Tidal turbine configuration and simulated extractable power in the Western Passage.

Total

Turbines

Turbine

Spacing

Hub

Height

Turbine

Diameter

Avg. Extracted

Power (kW)

Power per

Turbine (kW)

19 80 m × 160 m 15 m 20 m 4810 253

4. Summary

A high-resolution tidal hydrodynamic model using the unstructured-grid FVCOM was

developed to simulate the tidal hydrodynamics and characterize the tidal energy resource in the

Passamaquoddy–Cobscook Bay archipelago. The model was calibrated and validated with

observation data. The model performance error statistics show good model skills in reproducing the

tidal hydrodynamics in the study domain. The mean power densities along selected cross sections

were calculated and analyzed. The model results show multiple hotspots that have strong tidal

currents where potential tidal instream energy projects could be developed. However, the shallow

water depths in some strong current areas, such as Cobscook Bay and Head Harbor Passage, may

limit the potential deployment of tidal turbines unless small turbines are considered. The Western

Passage on the U.S. side shows great potential for tidal energy extraction because of its strong tidal

currents and greater water depth.

The theoretical resources, i.e., the upper limit of extractable energy, in the Western Passage, were

estimated to be 447,825 (kW) based on the model results validated using the Garrett and Cummins

method [5]. The extractable power for the hypothetical tidal farm was calculated using two different

methods recommended by IEC TS, one with undisturbed flow and the other with the FVCOM-MHK

model that simulates energy extraction directly. While these two methods provide similar results

when the extracted energy is small, the extractable energy estimated using the undisturbed flow

method is generally greater because the flow field interaction with the tidal turbine farm is not

considered. It should be noted that although IEC recommends two criteria (the extraction of 10 MW

or 2% of the theoretical resources) for determining the need to model energy extraction for resource

characterization, the 2% criterion could be exceeded before the 10 MW threshold when the system’s

Figure 14. Layout of the hypothetical tidal turbine farm in the Western Passage.

4. Summary

A high-resolution tidal hydrodynamic model using the unstructured-grid FVCOM was developed to simulate the tidal hydrodynamics and characterize the tidal energy resource in the Passamaquoddy– Cobscook Bay archipelago. The model was calibrated and validated with observation data. The model performance error statistics show good model skills in reproducing the tidal hydrodynamics in the study domain. The mean power densities along selected cross sections were calculated and analyzed. The model results show multiple hotspots that have strong tidal currents where potential tidal instream energy projects could be developed. However, the shallow water depths in some strong current areas, such as Cobscook Bay and Head Harbor Passage, may limit the potential deployment of tidal turbines unless small turbines are considered. The Western Passage on the U.S. side shows great potential for tidal energy extraction because of its strong tidal currents and greater water depth.

The theoretical resources, i.e., the upper limit of extractable energy, in the Western Passage, were estimated to be 447,825 (kW) based on the model results validated using the Garrett and Cummins method [5]. The extractable power for the hypothetical tidal farm was calculated using two different methods recommended by IEC TS, one with undisturbed flow and the other with the FVCOM-MHK model that simulates energy extraction directly. While these two methods provide similar results when the extracted energy is small, the extractable energy estimated using the undisturbed flow method is generally greater because the flow field interaction with the tidal turbine farm is not considered. It should be noted that although IEC recommends two criteria (the extraction of 10 MW or 2% of the theoretical resources) for determining the need to model energy extraction for resource characterization, the 2% criterion could be exceeded before the 10 MW threshold when the system’s theoretical resources are smaller than 500 MW. Therefore, the best practice for tidal resource characterization is to use the method of hydrodynamic modeling with energy extraction.

In this study, the turbine cut-in and cut-out velocities were not considered in the energy extraction simulations, which could overestimate the extractable energy by the tidal turbine farm. In addition, the energy loss due to the effect of friction by the turbine supporting structure was also not considered in the simulations. These factors should be incorporated in future studies to improve the accuracy of the extractable power estimates. As a best practice, tidal resource assessments using model simulations

J. Mar. Sci. Eng. 2020, 8, 411 19 of 21

should be conducted with the support of field measurements to ensure quality control because the model results always contain errors and uncertainties. Finally, the Garrett and Cummins’ method (Equation (4)) was developed based on a 1-dimensional governing equation for the tidal flow. Further, the water levels at both ends of the channel are independent, and the head difference can be represented by a sinusoidal function with a tidal amplitude [5]. However, in reality, tidal currents typically vary across the channel, and the tides at both ends of the channel may not be fully independent. Therefore, future studies are needed to modify the Garrett and Cummins method to improve the accuracy of the Pmax estimate using theoretical analysis and numerical modeling approaches.

Author Contributions: Conceptualization, Z.Y., K.H. and L.K.; methodology, Z.Y., T.W., K.H., L.K. and H.X.; validation, Z.Y., T.W., Z.X. and X.F.; formal analysis, Z.Y., T.W. and Z.X.; investigation, Z.Y. and T.W.; resources, Z.Y., L.K. and H.X.; data curation, Z.Y., T.W., H.X. and L.K.; writing—original draft preparation, Z.Y.; writing—review and editing, Z.Y., T.W., K.H. and L.K.; visualization, T.W. and Z.X.; supervision, Z.Y.; project administration, Z.Y.; funding acquisition, Z.Y. and L.K. All authors have read and agreed to the published version of the manuscript.

Funding: The U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy, Water Power Technologies Office funded the research performed by Pacific Northwest National Laboratory, operated by Battelle Memorial Institute under Contract DE-AC05-76RL01830, and by the National Renewable Energy Laboratory, operated by Alliance for Sustainable Energy, LLC under Contract DE-AC36-08GO28308. The views expressed in this article do not necessarily represent the views of the DOE or the U.S. Government. The publisher, by accepting the article for publication, acknowledges that the U.S. Government retains a non-exclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this work, or allow others to do so, for U.S. Government purposes.

Acknowledgments: All model simulations were performed using the Institutional Computing Facility at Pacific Northwest National Laboratory. The authors thank the technical steering committee, chaired by Tuba Ozkan-Haller at the Oregon State University, for their oversight and input during this study.

Conflicts of Interest: The authors declare no conflict of interest.

References

1. Uihlein, A.; Magagna, D. Wave and tidal current energy—A review of the current state of research beyond technology. Renew. Sustain. Energy Rev. 2016, 58, 1070–1081. [CrossRef]

2. Bahaj, A.S. New research in tidal current energy. Philos. Trans. R. Soc. A Math. Phys. Eng. Sci. 2013, 371, 20120501. [CrossRef] [PubMed]

3. Bryden, I.G.; Couch, S.I.; Owen, A.; Melville, G. Tidal current resource assessment. Proc. Inst. Mech. Eng. Part A J. Power Energy 2007, 221, 125–135. [CrossRef]

4. Hussain, A.; Arif, S.M.; Aslam, M. Emerging renewable and sustainable energy technologies: State of the art. Renew. Sustain. Energy Rev. 2017, 71, 12–28. [CrossRef]

5. Garrett, C.; Cummins, P. The power potential of tidal currents in channels. Proc. R. Soc. A Math. Phys. Eng. Sci. 2005, 461, 2563–2572. [CrossRef]

6. Garrett, C.; Cummins, P. Maximum power from a turbine farm in shallow water. J. Fluid Mech. 2013, 714, 634–643. [CrossRef]

7. Yang, Z.Q.; Wang, T.P.; Copping, A.E. Modeling tidal stream energy extraction and its effects on transport processes in a tidal channel and bay system using a three-dimensional coastal ocean model. Renew. Energy 2013, 50, 605–613. [CrossRef]

8. Blanchfield, J.; Garrett, C.; Wild, P.; Rowe, A. The extractable power from a channel linking a bay to the open ocean. Proc. Inst. Mech. Eng. Part A J. Power Energy 2008, 222, 289–297. [CrossRef]

9. Bahaj, A.S.; Myers, L. Analytical estimates of the energy yield potential from the Alderney Race (Channel Islands) using marine current energy converters. Renew. Energy 2004, 29, 1931–1945. [CrossRef]

10. Vennell, R.; Funke, S.W.; Draper, S.; Stevens, C.; Divett, T. Designing large arrays of tidal turbines: A synthesis and review. Renew. Sustain. Energy Rev. 2015, 41, 454–472. [CrossRef]

11. Adcock, T.A.A.; Draper, S. Power extraction from tidal channels—Multiple tidal constituents, compound tides and overtides. Renew. Energy 2014, 63, 797–806. [CrossRef]

12. Defne, Z.; Haas, K.A.; Fritz, H.M.; Jiang, L.D.; French, S.P.; Shi, X.; Smith, B.T.; Neary, V.S.; Stewart, K.M. National geodatabase of tidal stream power resource in USA. Renew. Sustain. Energy Rev. 2012, 16, 3326–3338. [CrossRef]

J. Mar. Sci. Eng. 2020, 8, 411 20 of 21

13. Thiebaut, M.; Filipot, J.F.; Maisondieu, C.; Damblans, G.; Duarte, R.; Droniou, E.; Chaplain, N.; Guillou, S. A comprehensive assessment of turbulence at a tidal-stream energy site influenced by wind-generated ocean waves. Energy 2020, 191, 116550. [CrossRef]

14. Thiebaut, M.; Sentchev, A.; du Bois, P.B. Merging velocity measurements and modeling to improve understanding of tidal stream resource in Alderney Race. Energy 2019, 178, 460–470. [CrossRef]

15. Lewis, M.; McNaughton, J.; Marquez-Dominguez, C.; Todeschini, G.; Togneri, M.; Masters, I.; Allmark, M.; Stallard, T.; Neill, S.; Goward-Brown, A.; et al. Power variability of tidal-stream energy and implications for electricity supply. Energy 2019, 183, 1061–1074. [CrossRef]

16. Wilcox, K.W.; Zhang, J.T.; McLeod, I.M.; Gerber, A.G.; Jeans, T.L.; McMillan, J.; Hay, A.; Karsten, R.; Culina, J. Simulation of device-scale unsteady turbulent flow in the Fundy Tidal Region. Ocean Eng. 2017, 145, 59–76. [CrossRef]

17. Work, P.A.; Haas, K.A.; Defne, Z.; Gay, T. Tidal stream energy site assessment via three-dimensional model and measurements. Appl. Energy 2013, 102, 510–519. [CrossRef]

18. Cowles, G.W.; Hakim, A.R.; Churchill, J.H. A comparison of numerical and analytical predictions of the tidal stream power resource of Massachusetts, USA. Renew. Energy 2017, 114, 215–228. [CrossRef]

19. Marta-Almeida, M.; Cirano, M.; Soares, C.G.; Lessa, G.C. A numerical tidal stream energy assessment study for Baia de Todos os Santos, Brazil. Renew. Energy 2017, 107, 271–287. [CrossRef]

20. Li, X.R.; Li, M.; McLelland, S.J.; Jordan, L.B.; Simmons, S.M.; Amoudry, L.O.; Ramirez-Mendoza, R.; Thorne, P.D. Modelling tidal stream turbines in a three-dimensional wave-current fully coupled oceanographic model. Renew. Energy 2017, 114, 297–307. [CrossRef]

21. Yang, Z.; Wang, T.; Copping, A.; Geerlofs, S. Modeling of in-stream tidal energy development and its potential effects in Tacoma Narrows, Washington, USA. Ocean Coast. Manag. 2014, 99, 52–62. [CrossRef]

22. Sutherland, G.; Foreman, M.; Garrett, C. Tidal current energy assessment for Johnstone Strait, Vancouver Island. Proc. Inst. Mech. Eng. Part A J. Power Energy 2007, 221, 147–157. [CrossRef]

23. Wang, T.P.; Yang, Z.Q. A modeling study of tidal energy extraction and the associated impact on tidal circulation in a multi-inlet bay system of Puget Sound. Renew. Energy 2017, 114, 204–214. [CrossRef]

24. Yang, Z.; Wang, T. Modeling the effects of tidal energy extraction on estuarine hydrodynamics in a stratified estuary. Estuaries Coasts 2013, 38, 187–202. [CrossRef]

25. Wang, T.; Yang, Z.; Copping, A. A modeling study of the potential water quality impacts from in-stream tidal energy extraction. Estuaries Coasts 2013, 38, 173–186. [CrossRef]

26. Kadiri, M.; Ahmadian, R.; Bockelmann-Evans, B.; Rauen, W.; Falconer, R. A review of the potential water quality impacts of tidal renewable energy systems. Renew. Sustain. Energy Rev. 2012, 16, 329–341. [CrossRef]

27. IEC. Marine Energy—Wave, Tidal and Other Water Current Converters—Part 201: Tidal Energy Resource Assessment and Characterization; IEC TS 62600-201; International Electrotechnical Commission: Geneva, Switzerland, 2015.

28. Rao, S.; Xue, H.J.; Bao, M.; Funke, S. Determining tidal turbine farm efficiency in the Western Passage using the disc actuator theory. Ocean Dyn. 2016, 66, 41–57. [CrossRef]

29. Karsten, R.; Swan, A.; Culina, J. Assessment of arrays of in-stream tidal turbines in the Bay of Fundy. Philos. Trans. R. Soc. A Math. Phys. Eng. Sci. 2013, 371, 20120189. [CrossRef]

30. Phoenix, A.; Nash, S. Specification of the thrust coefficient when using the momentum sink approach for modelling of tidal turbines. J. Renew. Sustain. Energy 2018, 10, 044502. [CrossRef]

31. Roc, T.; Conley, D.C.; Greaves, D. Methodology for tidal turbine representation in ocean circulation model. Renew. Energy 2013, 51, 448–464. [CrossRef]

32. Yang, Z.; Copping, A. Marine Renewable Energy—Resource Characterization and Physical Effects; Springer: Berlin/Heidelberg, Germany, 2017.

33. Ahmadian, R.; Falconer, R.; Bockelmann-Evans, B. Far-field modelling of the hydro-environmental impact of tidal stream turbines. Renew. Energy 2012, 38, 107–116. [CrossRef]

34. Kilcher, K.; Thresher, R.; Tinnesand, H. Marine Hydrokinetic Energy Site Identification and Ranking Methodology Part II: Tidal Energy; National Renewable Energy Laboratory: Golden, CO, USA, 2016; p. 30.

35. Brooks, D.A. The tidal-stream energy resource in Passamaquoddy-Cobscook Bays: A fresh look at an old story. Renew. Energy 2006, 31, 2284–2295. [CrossRef]

36. Amante, C.; Eakins, B.W. ETOPO1 1 Arc-Minute Global Relief Model: Procedures, Data Sources and Analysis; NOAA Technical Memorandum NESDIS NGDC-24 2009; National Geophysical Data Center, NOAA: Boulder, CO, USA, 2009.

J. Mar. Sci. Eng. 2020, 8, 411 21 of 21

37. Chen, C.; Liu, H.; Beardsley, R.C. An unstructured grid, finite-volume, three-dimensional, primitive equations ocean model: Application to coastal ocean and estuaries. J. Atmos. Ocean. Technol. 2003, 20, 159–186. [CrossRef]

38. Chen, C.B.; Robert, C.; Cowles, G. An unstructured grid, finite-volume coastal ocean model (FVCOM) system. Oceanography 2006, 19, 78–89. [CrossRef]

39. Mellor, G.L.; Yamada, T. Development of a turbulence closure-model for geophysical fluid problems. Rev. Geophys. 1982, 20, 851–875. [CrossRef]

40. Xuan, J.; Yang, Z.; Huang, D.; Wang, T.; Zhou, F. Tidal residual current and its role in the mean flow on the Changjiang Bank. J. Mar. Syst. 2016, 154, 66–81. [CrossRef]

41. Lai, Z.; Ma, R.; Gao, G.; Chen, C.; Beardsley, R.C. Impact of multichannel river network on the plume dynamics in the Pearl River estuary. J. Geophys. Res. Ocean. 2015, 120, 5766–5789. [CrossRef]

42. Chen, T.Q.; Zhang, Q.H.; Wu, Y.S.; Ji, C.; Yang, J.S.; Liu, G.W. Development of a wave-current model through coupling of FVCOM and SWAN. Ocean Eng. 2018, 164, 443–454. [CrossRef]

43. Xiao, Z.Y.; Wang, X.H.; Roughan, M.; Harrison, D. Numerical modelling of the Sydney Harbour Estuary, New South Wales: Lateral circulation and asymmetric vertical mixing. Estuar. Coast. Shelf Sci. 2019, 217, 132–147. [CrossRef]

44. Khangaonkar, T.; Sackmann, B.; Long, W.; Mohamedali, T.; Roberts, M. Simulation of annual biogeochemical cycles of nutrient balance, phytoplankton bloom(s), and DO in Puget Sound using an unstructured grid model. Ocean Dyn. 2012, 62, 1353–1379. [CrossRef]

45. Khangaonkar, T.; Nugraha, A.; Hinton, S.; Michalsen, D.; Brown, S. Sediment transport into the Swinomish Navigation Channel, Puget Sound—Habitat restoration versus navigation maintenance needs. J. Mar. Sci. Eng. 2017, 5, 19. [CrossRef]

46. Luo, L.; Wang, J.; Schwab, D.J.; Vanderploeg, H.; Leshkevich, G.; Bai, X.Z.; Hu, H.G.; Wang, D.X. Simulating the 1998 spring bloom in Lake Michigan using a coupled physical-biological model. J. Geophys. Res. Ocean. 2012, 117. [CrossRef]

47. Hakim, A.R.; Cowles, G.W.; Churchill, J.H. The impact of tidal stream turbines on circulation and sediment transport in Muskeget Channel, MA. Mar. Technol. Soc. J. 2013, 47, 122–136. [CrossRef]

48. Yang, Z.Q.; Wang, T.P.; Leung, R.; Hibbard, K.; Janetos, T.; Kraucunas, I.; Rice, J.; Preston, B.; Wilbanks, T. A modeling study of coastal inundation induced by storm surge, sea-level rise, and subsidence in the Gulf of Mexico. Nat. Hazards 2014, 71, 1771–1794. [CrossRef]

49. Wang, T.P.; Yang, Z.Q. The nonlinear response of storm surge to sea-level rise: A modeling approach. J. Coast. Res. 2019, 35, 287–294. [CrossRef]

50. Yoon, J.J.; Shim, J.S. Estimation of storm surge inundation and hazard mapping for the southern coast of Korea. J. Coast. Res. 2013, 65 (Suppl. 1), 856–861. [CrossRef]

51. Ge, J.; Much, D.; Kappenberg, J.; Nino, O.; Ding, P.; Chen, Z. Simulating storm flooding maps over HafenCity under present and sea level rise scenarios. J. Flood Risk Manag. 2014, 7, 319–331. [CrossRef]

52. Chen, C.S.; Beardsley, R.C.; Luettich, R.A.; Westerink, J.J.; Wang, H.; Perrie, W.; Xu, Q.C.; Donahue, A.S.; Qi, J.H.; Lin, H.C.; et al. Extratropical storm inundation testbed: Intermodel comparisons in Scituate, Massachusetts. J. Geophys. Res. Ocean. 2013, 118, 5054–5073. [CrossRef]

53. Egbert, G.D.; Erofeeva, S.Y. Efficient inverse modeling of barotropic ocean tides. J. Atmos. Ocean. Technol. 2002, 19, 183–204. [CrossRef]

54. Murray, R.O.; Gallego, A. A modelling study of the tidal stream resource of the Pentland Firth, Scotland. Renew. Energy 2017, 102, 326–340. [CrossRef]

55. Haverson, D.; Bacon, J.; Smith, H.C.M.; Venugopal, V.; Xiao, Q. Modelling the hydrodynamic and morphological impacts of a tidal stream development in Ramsey Sound. Renew. Energy 2018, 126, 876–887. [CrossRef]

56. Plew, D.R.; Stevens, C.L. Numerical modelling of the effect of turbines on currents in a tidal channel—Tory Channel, New Zealand. Renew. Energy 2013, 57, 269–282. [CrossRef]

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

  • Introduction
  • Methods
    • Study Site
    • Tidal Hydrodynamic Model
    • Model Configurations and Boundary Conditions
    • Observation Data for Model Calibration and Validation
  • Results and Discussion
    • Model Calibration
    • Model Validation
    • Characteristics of Tidal Hydrodynamics
    • Along-Channel Kinetic Energy Flux
    • Energy Extraction in the Western Passage
  • Summary
  • References