Review on Energy Resilience

Energy Economics 50 (2015) 215–226

Contents lists available at ScienceDirect

Energy Economics

journal homepage: www.elsevier.com/locate/eneco

Energy import resilience with input–output linear programming models

Peijun He a, Tsan Sheng Ng a,⁎, Bin Su b a Department of Industrial and Systems Engineering, National University of Singapore, Singapore b Energy Studies Institute, National University of Singapore, Singapore

⁎ Corresponding author. Tel.: +65 6516 2562; fax: +6 E-mail addresses: [email protected] (P. He), ise

[email protected] (B. Su).

http://dx.doi.org/10.1016/j.eneco.2015.05.007 0140-9883/© 2015 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history: Received 16 January 2015 Received in revised form 9 May 2015 Accepted 20 May 2015 Available online 6 June 2015

JEL classification: C44 C61 C67 Q40 Q41 Q48

Keywords: Energy resilience Input–output modeling Linear programming

In this work we develop a new approach to study the energy import resilience of an economy using linear programming and economic input–output analysis. In particular, we propose an energy import resilience index by examining the maximum level of energy import reduction that the economy can endure without sacrificing domestic demands. A mixed integer programming model is then developed to compute the resilience index efficiently. The methodology is applied to a case study using China input–output data to study the energy import resilience under different power generation portfolio assumptions. We demonstrate how our models can be used to uncover important inter-sectoral dependencies, and to guide decision-makers in improving the energy resilience in a systematic manner.

© 2015 Elsevier B.V. All rights reserved.

1. Introduction

The central theme of concern in this work is succinctly described by IEA's definition of national energy security (IEA, 2014), where in partic- ular “short-term energy security focuses on the ability of the energy system to react promptly to sudden changes in the supply–demand balance”. In- deed, unanticipated disruptions in energy supplies (primary fuels, elec- tricity, high energy-embodied goods) can lead to cascading problems in the economy due to the complex interdependency between different industry sectors. For example, the U.S. witnessed a huge decline in eco- nomic output during the 1973 oil supply crisis (Akins, 1973). In 2008, a shortage of hydroelectricity supply in Central Asia caused severe dam- age to the national economy, especially in agriculture sectors, which led to a major food crisis in the same year (Libert et al., 2008).

Energy import reliance is often identified as an important determi- nant of national energy security, as evident from the various published studies on energy security performance indicators (Geng and Ji, 2014; Kruyt et al., 2009; Le Coq and Paltseva, 2009; Löschel et al., 2010; Vivoda, 2009). It is clear that an economy highly dependent on imports for its primary energies faces greater exposure to risks of exogenous supply shocks. On the other hand, two economies with similar degree

5 6777 1434. [email protected] (T.S. Ng),

of import dependencies may have very different reactions when ex- posed to the same import supply shocks. This can be due simply to their inherently different inter-sectoral dependency structure. A desir- able structure should allow effective mitigation of the energy import loss via redistribution of production resources to restore supply and demand balance. This is coherent with the concept of energy resilience as espoused in other works (Flynn, 2012; Molyneaux et al., 2012; OBrien and Hope, 2010; Thomas and Kerner, 2010). It is of interest in this work to quantify and evaluate such a concept of energy resilience using a model-based approach.

In this paper, we propose an input–output based linear program- ming model that focuses on the relationship between energy imports, industrial production technologies and capacities. Basically, the model can be used to simulate the impact of specified energy import losses on the sectoral production levels, and consequently, the final supply– demand balance. In reality, energy import losses may arise from various importing sectors and cannot be anticipated precisely in advance. A key technical contribution of our work is that the proposed model can be used to evaluate the worst-case impact over a family of (possibly infi- nite) import loss scenarios. This family of loss scenarios is defined, for instance, as the set of all loss scenarios that has the same total tonnes of coal equivalent (TCE) of energetic content. In the proposed model, the impact of an energy import loss on the economy is defined as the total final demands deficit, that is, the amount of final demands of goods that cannot be balanced by the given supply and production in

216 P. He et al. / Energy Economics 50 (2015) 215–226

the short run. This is then extended to define an energy import resilience indicator, which essentially evaluates the highest level of energy import loss sustainable by the economy. Such an indicator provides decision- makers with a simple way to rank the related performance of an economy.

The outline of the rest of the paper is organized as follows. The next section provides a literature review of related works in the application of input–output models for energy supply studies, and also the use of linear programming models in input–output analysis. Section 3 intro- duces our proposed input–output linear programming model, the uncertainty model for defining the family of energy import loss scenar- ios, and also our proposed energy import resilience indicator. Moreover, to evaluate the worst-case impact of the economic system from the set of import loss scenarios, we propose an equivalent formulation of the model in the format of a mixed linear integer programming model. We also extend the methodology to compute production capacity de- signs that achieves the maximum possible energy import resilience of a given input–output structure. In Section 4, we provide a case example based on China's input–output and power generation data to demon- strate the application of our approach. Our basic results show that ener- gy import resilience can be significantly influenced by production capacities and technologies of non-energy producing sectors. The nu- merical studies also reveal that in the case of China, over-reliance on natural gas for power generation (by displacement of coal-fired gener- ation) may compromise the energy import resilience of the economy if no changes to the existing production structure are made. Different approaches are also discussed to improve the energy resilience. Finally, Section 5 concludes this work and suggests future research directions.

2. Literature review

An important application of input–output (IO) analysis is to study the impact of energy shortages on the economic system. For example, Penn and Irwin (1977) demonstrated the use of input–output models for studying energy constrained scenarios, including reduction of crude petroleum imports, shortage of refined petroleum, and restriction on natural gas imports. It was noted that the impacts of shortages were often not obvious and may present unexpected results due to the inter- sectoral dependencies. In another study, Kerschner and Hubacek (2009) applied IO analysis to investigate the impacts of oil supply shortage to the economy. The authors show that their methodology can help iden- tify sectors most vulnerable to oil supply disruptions.

Combining linear programming (LP) (Dantzig, 1949) with IO model- ing is a popular approach to study various energy–economy–environ- ment problems in the research literature. Chenery and Clark (1960) proposed an LP model to optimize the use of primary resources such as labor, capital and land by maximizing the total value of total output. Just (1974) presented a methodology to quantify the impact of techno- logical change on the economy. In particular, the authors studied how the adoption of High-BTU or Low-BTU coal gasification and combined gas and steam cycle generation impacts the output of several sectors and change energy consumption, using 1985 projections of U.S. data. Schluter and Dyer (1976) in a later work extended the study by intro- ducing additional constraints on the total output level and maximizing the gross domestic production.

James et al. (1986) proposed an integration of the input–output model with a dynamic energy technology optimization model to project the change in total energy demand and technological mix. The authors then identified some economic implications of technological change and inter-fuel substitution using the model. Leung and Hsu (1984) pro- posed the use of an LP model to study the economic impacts of energy supply reductions on the Hawaiian economy. The authors used the LP model to compute shadow prices for different levels of gasoline avail- ability, and they demonstrated that the LP solution can be used as an ef- ficient allocation of energy resources to various industry sectors during energy shortages. Other works proposing IO–LP models include Wang

and Miller (1995), who analyzed the economic impact of a transporta- tion and energy supply bottleneck on the economy of Taiwan. Rose et al. (1997) used IO–LP to study the regional economic impact of an earthquake and they found that a disruption in electricity was caused but reallocation of the scarce electricity across sectors could reduce the impacts substantially. Wilting et al. (2008) propose a dynamic IO– LP model used to perform scenario analysis with various levels of pro- duction and carbon emissions based on different assumptions of future (2030) technological change. Vogstad (2009) provided a more recent review of the modeling capability of LP in IO analysis. Tunkay and Bilge (2012) performed a case study using data from Turkey to develop a IO–LP model for optimizing the distribution of economic resources on sectoral basis to maximize national income.

Most reviewed works on IO–LP in the above focused on using the proposed LP model to optimize some socio-economic objective function (e.g. GDP), assuming that energy availability will remain the same as the data. They have not considered the issues of energy shortages and ener- gy resilience under uncertainty. In our opinion, the latter problems are of greater interest from an energy security point of view, and warrant a deeper study. Hence, in this work, we propose the advancement and enhancement of the IO–LP modeling approach for the analysis of energy resilience. In particular, our focus in this work is to use the IO–LP model as a means to develop an energy resilience indicator of the economy in the event of uncertain energy import shortages.

3. Methodology

In this section, our major goal is to develop a quantitative indicator for evaluating the energy import resilience of a given economy. More specifically, the energy import resilience indicator measures the highest level of energy import losses that can be absorbed by the economy, while maintaining production supply and demand balance in the short run. To achieve a rigorous and meaningful definition of the energy import resilience indicator, however, we first need to develop various related components based on the traditional economic IO model. The relevant notations in this paper are stated as follows. Other notation will be defined as when appropriate and necessary.

sj Demand deficit in commodity j dj Domestic demand of commodity j (the domestic exports are

included in the domestic demand) yj,t Output of commodity j using technology t. uj Import of commodity j in physical unit (tonnes of coal equiv-

alent, TCE) for energy sectors, while in monetary unit (in monetary value) for non-energy sectors

aji t Amount of good j consumed for each $1 production of good i

using technology t rj Price for energy import j (in monetary value), while it equals

one for non-energy imports εj Import loss of energy sector j, it equals to zero for non-energy

sectors ε Vector of import loss P Total output J Set of economic sectors in the input–output model JE Set of energy goods sectors of interest Tj Set of technologies of sector j Ū Import limit for energy sectors uj_max Import limit for non-energy sectors ωj Import limit of commodity, ωj = uj_max, forj ∉ JE; ω =

Ū, forj ∈ JE xj,t Production capacity of commodity j using technology t x Vector of production capacityλ, π, δ, μ Dual variables

The economic IO model is developed by Wassily Leontief in the 1930s to study the interdependencies between different sectors using a system of linear equations. It is well-known that the IO model can be

217P. He et al. / Energy Economics 50 (2015) 215–226

formulated as a special case of a linear programming model, with the extension of an objective function over which some performance measure can be optimized. In fact, one of the earliest applications of linear programming is related to technology selection in IO analysis (Dantzig, 1963). For example, the following LP can be solved to yield the choice of production levels yj,t to minimize total imports reliance:

minimize X j

uj ð1ÞX X X

subject to

t∈T j

yj;t ≥ i t∈Ti

atjiyi;t þ dj−r juj ∀j∈J ð2Þ

uj ≤ω j ∀j ∈J ð3Þ

X j∈J

X t∈T j

yj;t ≤P ð4Þ

yj;t ≤xj;t ∀t∈T j; ∀j ∈J ð5Þ

y; u≥0 ð6Þ

In the above constraint (2) models the supply–demand balances for all sectors, i.e. total production of a sector j plus imports should be no less than the sum of intermediate demands and final demands. We as- sume energy imports uj for all j ∈ JE, are defined in physical energy units (e.g. TCE), while domestic production levels yj,t are defined in their monetary values (for a given base year price). Hence, energy im- ports are monetized in Eq. (2) using the energy import price rj. Con- straint (3) states that the level of import for energy commodity j can be no greater that ωj. Constraint (4) states that total domestic produc- tion of the economy should be bounded above by some common re- source limit, for instance in the total available industrial land space or labor. Finally, Eq. (5) states that the domestic production of each com- modity j using technology t should be no greater than the available pro- duction capacity xj,t.

In what follows, we extend the application of the IO–LP model in Eqs. (1)–(6) to define a final demands deficit function which is used to evaluate the level of demand deficit an economy suffers under energy import loss events. We then propose an approach to quantify the energy import resilience based on considering the worst possible import loss scenario that the economy can tolerate. However, evaluating the resil- ience indicator value involves solving a series of non-linear optimization problems that can be computationally challenging for large models. Fi- nally, we present a reformulation of the model that allows it to be solved efficiently using off-the-shelf mixed integer programming solvers.

3.1. Modeling of demand deficits

It is clear that economies highly reliant on primary energy imports can incur serious economic repercussions when an unanticipated shock in imported energy supplies occurs. In this work, we focus on the impact of energy import losses on the supply–demand balance in the economy. We define a demand deficit as that level of final demand that cannot be fulfilled in the short term by domestic production or available imports. Let εj be the level of import loss for some energy com- modity j ∈ JE, and ϵ∈ℜ JEj j be the vector of all such energy import loss levels. A given value of ε is regarded as a specific import loss event. For a given production capacity x and loss event ε, define the final de- mands deficit function, f(x, ε) as the optimal objective function value of the following:

f x; ϵð Þ ¼ minimize X j

sj ð7Þ

subject to X t∈T j

yj;t ≥ X i

X t∈Ti

atjiyi;t þ dj−r juj−sj ∀j ∈J ð8Þ

uj ≤ ω j−ϵj ∀j ∈J ð9Þ X j∈ J

X t∈T j

yj;t ≤ P ð10Þ

yj;t ≤ xj;t ∀t ∈T j; ∀j ∈J ð11Þ y; s; u ≥ 0 ð12Þ

The optimization model (7)–(12) is a linear programming model which when solved, minimizes the total demand deficits incurred when the energy import loss event ε takes place. Eq. (8) are the sup- ply–demand balance conditions of the IO model, as described in Eq. (2). Constraint (9) states that the level of import for energy com- modity j can now be no greater that the nominal level ωj less the loss level εj. For the values of the import limits ωj, one can set these as the highest import levels over a set of historical data as a reasonable refer- ence point. Eqs. (10) and (11) are production capacity constraints that are reproduced from Eqs. (4) and (5).

The concept of defining f(x, ε) in the above is based on the following inquiry: Given an import loss event ε and capacities x, does there exist a feasible production y and import level u so that all final demands for goods are supported (i.e. f(x, ε) = 0) ? Hence, the optimal solution y⁎ in Eqs. (7)–(12) can be regarded as a form of mitigating response of the economy in the event of the energy import loss ε. For example, let yj,t and yj;t0 be the initial equilibrium levels of power generation output (commodity j) using coal-fired (technology t) and natural gas-fired gen- eration (technology t′) respectively. When a severe import loss of coal happens (and assuming no domestic source of coal is available), the mitigated level of coal-fired generation yj,t⁎ is necessarily curtailed due to the supply limitation, i.e. yj,t⁎ ≤ yj,t and the level of natural gas-fired generation is increased, i.e. y�j;t0 ≥yj;t0 in order to restore the supply– demand balance. The level of fuel switching that can be achieved, of course depends on the available installed production capacity at the time of the import loss crisis (and possibly depends also on the produc- tion capacities of other inter-dependent sectors). This reiterates the conventional wisdom of diversifying the power generation mix to hedge against supply risks.

The above proposed model (7)–(12) can easily be extended to ac- count for many different issues with the modifications of the constraints or objective function. For example, considerations such as minimum production level requirements, economic targets, CO2 emission targets, can be accounted for with additional linear constraints, and does not change the concept of the approach. It can also be extended to include strategic petroleum reserves explicitly, which is also an important in- strument in mitigating energy import losses. In this work, we focus on the basic model (7)–(12) for the purpose of presenting our methodolo- gy clearly. Even though rather simplistic in structure, several interesting insights can be generated as will be elaborated in our case study. Finally, in this work, we focus exclusively on the supply–demand balance of goods in the economy in the short run in physical units and do not consider price changes. In our opinion, the economics of supply and demand imbalance are the fundamental mechanisms that trigger ob- served inflation in prices of import energies and related goods in reality. These supply and demand imbalances are explicitly captured in our def- inition of the final demands deficits.

3.2. Modeling energy import resilience

Based on the aforementioned concept of demand deficits, we now define the energy import resilience of a given economy as the highest level of total energy import loss that can be absorbed by that economy without resulting in positive demand deficits. The same level of total

218 P. He et al. / Energy Economics 50 (2015) 215–226

energy import loss, can however, arise from many different possible sce- narios. Indeed, in reality, energy import losses can occur in various sectors such as primary fuels (e.g. coal, oil, natural gas), electricity, or even other secondary energy goods such as gasoline and petroleum products, or other high energy-embodied goods. Import losses are also often events that can have very little historical bearing. In particular, it is difficult to predict how much or where the energy import shortages can occur. It is also possible that import losses can take place in various importing sectors simultaneously. With these considerations, we define Φ(θ) as the follow- ing set of import loss scenarios parameterized by the scalar θ:

Φ θð Þ ¼ ϵ j ��X

j

ϵ j ≤ θ � E; ϵ j ≤ U; ∀j ∈ JE

8< :

9= ; ð13Þ

where Ū is the maximum energy import level for each individual fuel type, E denotes the maximum total import loss level considered, and θ ∈ [0, 1] is a proportion multiplier that can be used to consider sets of loss scenarios Φ(θ) corresponding to different severity levels. That is, Φ(θ) describes the set of all import loss scenarios ε that has total energy losses no greater than θ ⋅ E (it is assumed that all the energy imports con- sidered are converted to common energy units, e.g. in TCE). Clearly, Φ(θ′) ⊆ Φ(θ) whenever θ′ ≤ θ. The import loss limit Ū can be set as the maximum level of energy import based on historical data across the dif- ferent fuel types considered. The maximum total import loss level E can be calibrated by setting E :¼ ∑ j∈ JE U−∑ju�j , where uj⁎ is retrieved from the optimal solution of models (1)–(6). Recall that since Eqs. (1)–(6) min- imize the total energy imports, this implies that there does not exist any feasible solution, under any scenario, where the economy can do less than this level of imports, i.e. ∑juj⁎. Consequently, there does not exist any feasible solution or scenario whereby energy import loss will be greater than the above choice of E. This calibration states that the total en- ergy equals the total energy import available minus the minimum value of imports required.

The assumption of the import loss scenario set Φ(θ) enables us now to mathematically define the import resilience indicator θ⁎, formulated as follows:

θ� ¼ max θj g θ; xð Þ≤0f g ð14Þ

where g(θ; x) refers to the worst level of demand deficits defined as:

g θ; xð Þ ¼ max ϵ∈Φ θð Þ

f x; ϵð Þ ð15Þ In the above, the worst (highest) level of deficits is evaluated by

maximizing the import loss scenarios ε over Φ(θ). In summary, we say that the import resilience θ⁎ is no smaller than a given θ, if it can mitigate against all loss scenarios arising from Φ(θ) without incurring positive deficits.

Unlike other proposed energy security performance indexes (Geng and Ji, 2014; Kruyt et al., 2009; Le Coq and Paltseva, 2009; Löschel et al., 2010; Vivoda, 2009) which either consider multiple dimensions of energy security with different metrics or propose an aggregated index based different indicators with subjective weights, our energy im- port resilience index is rather simple and objective. Thus, based on the above definition, the value of θ⁎ can be used as a quantitative indicator for an economy's import energy resilience. This is useful in evaluating the impact of changes in input–output structure on the energy resil- ience and security of the economy. It also provides a basis for ranking fu- ture alternatives in their energy import resilience.

3.3. Computing energy import resilience

We now consider the computational issues related to the evaluation of the energy import resilience θ⁎. Note that if g(θ; x) can be computed efficiently, θ⁎ can be evaluated using a binary search on the interval [0,

1], since g(θ; x) is clearly non-decreasing in θ. In particular, the binary search algorithm is proposed as follows.

Algorithm 1. Binary search algorithm for computing θ⁎

Step 1: Initialization. Set k:=1, L:=0, U:=1, ε:=0.0001 Step 2: Check U − L. If U − L = ε, update θ⁎:=L, terminate the algorithm. Else Go to Step 3. End If Step 3: Set θ = (L + U)/2 and compute g(θ; x) If g(θ; x) ≤ 0, update L:=θ. Else, update U:=θ, k:=k + 1, and go to Step 2. End If

Hence, the major computational consideration is the evaluation of g(θ; x), which can be formulated directly as the following max–min objective optimization model:

g θ; xð Þ ¼ maxϵ∈Φ θð Þ f x; ϵð Þ ¼ maxϵ∈Φ θð Þminy ϵð Þ X j∈J

sj j s:t: 8ð Þ− 12ð Þ 8< :

9= ; ð16Þ

Note that if the number of import loss scenarios ε is finite and only a reasonably small number, problem (16) can be solved by simply fixing ε and then solving the corresponding linear programming model. The worst-case demands deficit is then determined by inspecting the largest deficits incurred across the scenarios. The above however, describes the maximization over an infinite set of ε arising from the polyhedral set Φ(θ). The problem can be shown to be equivalent in structure to a spe- cific instance of two-stage adaptable robust linear optimization prob- lem, which is computationally intractable in general (see e.g. Ben-Tal et al. (2009)). While one may suggest the use of a sampling approach (i.e. evaluating g(θ; x) over a given finite sample of ε), there is the risk of using a bad sample which can significantly underestimate the true value of g(θ; x). In the context of energy resilience and security, it is rea- sonable to assume that decision-makers are averse to uncertainty and ‘erring on the wrong side’ of the evaluation cannot be easily justified. Fortunately, in our problem, we are able to show that the max–min op- timization in Eq. (16) can be converted into a single instance of a linear mixed integer programming (MIP) problem, which can be solved quite efficiently using commercial solvers such as CPLEX. This result is pre- sented in the following proposition.

Proposition 1. Define Û = min{θ ⋅ E, Ū}, and Q ¼ ⌊θ�E Û ⌋. Evaluating g(θ; x)

by solving problem (16) is equivalent to solving the following mixed inte- ger program. The detailed equivalent mix integer programming is as follows.

g θ; xð Þ ¼ max λ j;π j;μti ;δ;α j

X j

djλ j þ X j

α j−P � δ− X t∈T j

X j

xj;tμ j;t ð17Þ

subject to λi− X j

atijλ j−δ−μ j;t ≤0 ∀i; j ∈ J ð18Þ

α j ≤ 0−U � �

π j þ 1−kj � �

M ∀j ∈ J ð19Þ � � � �

α j ≤ Û−U π j þ 1−k0j M ∀j ∈ J ð20Þ

α j ≤ θ � E Û

−Q � �

� Û−U

π j þ 1−k″j � �

M ∀j ∈ J ð21Þ

kj þ k0j þ k ″ j ¼ 1; ∀j ∈ J ð22Þ

X j

k0j ≤ Q ∀j ∈ J ð23Þ

219P. He et al. / Energy Economics 50 (2015) 215–226

X j

k″j ≤1 ∀j ∈ J ð24Þ

r jλ j−π j ≤0 ∀j ¼ 1; 2; …; N ð25Þ

λ j ≤1 ∀j ∈ J ð26Þ

λ j; π j; δ; μ j;t ≥0; α j ≤0 ∀j ∈ J ð27Þ

where k, k′, k″ ∈ {0, 1} J are defined as binary variables, and M is a large positive constant.

Proof. See Appendix A. □

With the help of Proposition (1), the original problem can be reformulated to a solvable single MIP with which we are able to com- pute energy import resilience by the binary search algorithm. The prac- tical implication of this result is that the evaluation of the resilience index can now be performed exactly, and need not rely on Monte Carlo simulation approaches, which unavoidably contain sampling errors.

3.4. Optimizing production capacity for energy resilience

We now extend the model developed so far to answer the ques- tion: how can we compute a design of production capacity x that achieves the maximum possible energy import resilience of a given input–output structure? In other words, we are interested in solving for x in the following problem:

θ�� :¼ max x∈X

θ; s:t: g θ; xð Þ≤0 ð28Þ

where θ⁎⁎ denotes the maximum possible level of θ achieved. Since θ⁎⁎ represents an upper bound on the system performance, it is use- ful in defining a normalized energy resilience indicator ρ = θ*/θ⁎ ⁎. In this sense the system performance is measured relative to that of an idealized production capacity design given the same import loss model.

Since g(⋅) is non-decreasing in θ, Eq. (28) can be solved using a bisec- tion search on θ, where in each iteration, the following query is made for

a given value ofθ ¼ θ̂: “Does there exist some x ∈ X such thatg θ̂; x � �

≤ 0?”.

If the answer to the above is yes, then θ�� ≥θ̂, otherwise, θ��bθ̂. The bisec- tion method is shown as follows.

Algorithm 2. Binary search algorithm for computing θ⁎⁎

Step 1: Initialization. Set k:=1, L:=0, U:=1, ε:=0.0001 Step 2: Check U − L. If U − L = ε, update θ⁎⁎:=L, terminate the algorithm. Else Go to Step 3. End If Step 3: Set θ̂ ¼ L þ Uð Þ=2 and compute g θ̂; x

� � If g θ̂; x

� � ≤0, update L :¼ θ̂.

Else, update U :¼ θ̂, k:=k + 1, and go to Step 2. End If

Hence, the main computational hurdle is the resolution of the optimization problem:

min x∈X

g θ; xð Þ ¼ min x∈X

maxϵ∈Φ θð Þ f x; ϵð Þ ¼ minx∈X maxϵ∈Φ θð Þminy ϵð Þ X j∈J

sj j 8ð Þ− 12ð Þ 8< :

9= ; ð29Þ

In general, the solving Eq. (29) is computationally difficult due to the complex of structure of the objective function. In the following, we first develop a re-formulation of Eq. (29), and then propose a constraint gen- eration solution approach which solves the problem via a series of linear and mixed integer problems.

Proposition 2. There exist some extreme point solution maximizing f(ε) over Φ(θ), i.e. the extreme point optimal solution is a vertex of the polyhe- dron Φ(θ).

Proof. See Appendix B. □

With the above result, we can reformulate problem (29) to a simpler form and to locate the extreme points of Φ(θ) which serve as the inputs to the following constraint generation algorithm. The practical implica- tion of this is the optimization of the energy resilience index can now be performed exactly by solving a large scale deterministic mathematical programming problem, instead of relying on simulation-based trial- and-error methods.

Let λk; πk; δk; μk; ϵk � �

∀k∈K n o

be the set of all extreme points of

(A.2)–(A.7), indexed by the set K. From the results of Proposition (2), problem (29) can now be reformulated as the following large-scale lin- ear program:

min x;β

β ð30Þ

s:t: β ≥ XN j¼1

djλ k j þ

XN j¼1

ϵkj−ω j � �

πkj−P � δk− XT t¼1

XN i¼1

xtjμ tk j ∀k ∈ K ð31Þ

xtj; β≥0 ð32Þ

While the above model is linear in x, the number of constraints grows exponentially in the size of the problem instance. Thus, a con- straint generation algorithm is proposed as follows to improve the com- putational efficiency of the model. The algorithm is guaranteed to terminate in a finite time since there is a finite number of (vertices) constraints.

Algorithm 3. Constraint generation algorithm for solving problem (30–32)

Step 1: Initialization. Set K⊂H , i.e., K is a subset of the indices of the extrema of (A.2)–(A.7). Step 2: Solve the relaxed master problem R:

R

min x;β

β

s:t: β≥ XN j¼1

djλ k j þ

XN j¼1

ϵkj−ω j � �

πkj−P � δk− XT t¼1

XN i¼1

xtjμ tk j ; ∀k∈K

xtj; β≥0

8>>>>>< >>>>>:

Step 3: If R is feasible then Extract the solution x̂ as its optimal solution, go to Step 4; Else terminate and report the problem as infeasible. End If Step 4: Letx ¼ x̂and solve the MIP (17)–(27). Based on the results in Proposition 1, obtain the solution (λk, πk, δk, μk, ϵk). If g x̂; ϵk� � ¼ 0 then terminate the algorithm and output x̂ as an optimal solution to problem R Else Update K :¼ K∪k and go to Step 2. End If

Table 1 Production capacity of energy sectors from 2007 data, unit: billion CNY.

Sector Production capacity

Coal 1024.2 Oil & Gas 1191.9 Coke & Petroleum 2634.3 Elec. by coal 2960.5 Elec. by oil 37.6 Elec. by gas 35.3 Elec. by others 542.3

220 P. He et al. / Energy Economics 50 (2015) 215–226

4. Case study

In this section, we apply the proposed methodology to study the energy import resilience of China using the 2007 IO data (NBS, 2007). Rapid economic growth and industrialization has made China one of the world's largest energy consumers. For example, China's natural gas consumption has grown at an annual rate of 17% since 2003, outstripping natural gas produced domestically in 2007 and reaching 5.7 trillion cubic feet in 2013. By 2013, a third of China's natural gas demands are met by imports, the EIA reported. It is thus of interest, in this case study, to investigate China's resil- ience to import disruptions in primary energies and electricity supply.

China's power generation to date is dominated by coal-fired power. However, extensive reliance on coal is associated with many negative environmental and health impacts such as climate change, respiratory diseases, acidic rain, and air pollution. To curb national GHG emissions and other environmental impacts, there are policies already in place to decommission several coal-fired power plants and investing in alterna- tive technologies, such as renewable energies and natural gas-firing. Al- though the total capacity of gas-fired power plants as of 2006 in China is only about 15.6 GW, or 2.5% of the total power generating capacity, it is expected to reach 70 GW by 2020 (a more recent plan shows capacity of 40 GW by 2012) (EIA, 2014b). Recent trends also hint of rapid growth of natural gas consumptions in the near future. For instance, in 2012, China became the world's third largest liquefied natural gas (LNG) importer, and continues to build 10 major LNG terminals in the urban coastal areas. China has also commissioned the installation of natural gas pipe- lines linking the western areas and eastern demand centers (EIA, 2014a). Thus, in this study we are interested to study the energy import resilience when China moves towards a reduced coal and increased nat- ural gas generation portfolio. We remark that it is not the intention of our computational study to analyze and forecast the China energy situ- ation in great accuracy or realism. To do so would be an overwhelming task beyond the scope of this paper. Neither do we suggest or advocate that a natural gas-based power generation portfolio would be feasible for China. Rather, we only assume some basic hypothetical situations in order to demonstrate the working of our model, and to improve the clarity in the explanation of the results and insights. Several different as- sumptions and more realistic scenarios can certainly be set up and stud- ied using our approach.

In the following, we first briefly describe the preparation of the commodity-industry IO model and data for the case study. Some basic scenario analyses are then performed, and the generated insights are discussed. Finally, we conduct numerical studies focusing on strategies to improve China's energy import resilience using the proposed model.

4.1. Data preparation

According to the sector classification by Chinese Input–output Asso- ciation, the major energy commodities in the (42 sector) IO table are Oil & Natural Gas Extraction, Coal Mining, Electricity Generation, and Coke & Petroleum, which make up almost 93% of the primary energy supply for China in 2007 (EIA, 2012a). Since our focus is on the primary energy and power generation technologies, some disaggregation of these sec- tors is required before applying the methodology in Section 3. The fol- lowing steps summarize the few key steps for preparing the IO data.

(a) Modeling different power generation technologies. For the electric- ity generation sector, we disaggregate it into four production technologies based on primary fuels: Coal, Oil, Gas and Others (hydro, geothermal, solar, wind, biofuels, etc.). Although in prac- tice power generation with each fuel may consist of different var- iations of technologies (i.e., coal-fired power generation may have different coal combustion technologies), we assume that power generation with one fuel type is treated as a single

aggregated technology for simplicity (we can consider alterna- tive technologies with respect one fuel type by simply modifying our model and disaggregate more columns in the IO table but which requires more data). Note that we do not disaggregate electricity into different commodities since power consumers are indifferent to the generation technologies. We use regional data to estimate the generation mix by primary fuel technology. The technical coefficients of the production technologies are approximated using some simple but reasonable assumptions, e.g. Power Generation by Oil technology uses mainly oil input; Power Generation by Gas technology only uses mainly gas input, etc. Finally, the RAS (De Mesnard and Miller, 2006) method is then used to balance the new IO table.

(b) Disaggregating Oil & Natural Gas Extraction commodity. Oil and natural gas is disaggregated as two separate commodities in our model, done since Electricity Generation is distinguished by fuel technology. However, as a production sector Oil & Natural Gas Ex- traction is regarded as a single sector since these two fuels essen- tially share the same production technology. The oil production and gas production ratio is computed using the existing data (EIA, 2012a). In particular, the crude oil and gas production are about 266.1742 million TCE and 92.0892 million TCE in 2007. Converting these dollar values using 2007 prices, the oil to gas production ratio is 0.9076 : 0.0924.

(c) Calibrating production capacities. For electricity generation sector, the 2007 total installed capacity is 725.4 million kilowatts (EIA, 2012a). The breakdown by generation mix is: coal-fired (83%), oil-fired (1%), gas-fired (1%), and others (15%). For the Oil & Gas Extraction and Coke & Petroleum sector, the 2007 output di- vided by the utilization rate reported in IMF (2012) is used to es- timate the production capacities. The capacities for the energy sectors in 2007 monetary values are shown in Table 1. A similar approach is applied for estimating the production capacities of the non-energy sectors.

Based on the above calibration processes, a modified 43 row-by-45 column IO table is developed. In particular, rows model different com- modities while columns model production sectors and technologies. A part of the table is presented in Table C.1, and the complete 45 sector listing is given in Table C.2 in the Appendix section of the paper.

4.2. Energy import resilience and generation portfolio

We first consider single generation technology scenarios to illustrate the application of our methodology. Though these scenarios model rather extreme positions, they are useful in revealing some interesting insights related to the economy structure. In the following scenarios using single technology, the production capacity of the technology of concern is assumed to be the sum of the generation production capaci- ties in the 2007 data.

Three different technology scenarios are considered: coal-firing only, oil-firing only, gas-firing only. The 2007 generation portfolio (de- noted as BAU) is also used as a reference case. The results are presented in Table 2. Besides the energy import resilience index θ⁎, we also

Table 2 Results of different technology scenarios, unit for import loss: million TCE.

Import loss in sector

Scenario θ⁎ (ρ) Coal Oil Gas Coke & Pet. Elec. Demand deficit in sector

BAU 1 0.21019 (0.50955) 0 0 16.2 0 0 Gas COA 2 0.21016 (0.50948) 0 0 16.2 0 0 Gas OIL 3 0.1870 (0.4533) 0 0 0 0 41 Other manu. GAS 4 0.0653 (0.1583) 0 0 0 0 14.6 MDG5

1 “BAU” represents the basic scenario by power generation using 2007 actual data. 2 “COA” represents the scenario by using coal only in power generation. 3 “OIL” represents the scenario by using oil only in power generation. 4 “GAS” represents the scenario by using gas only in power generation. 5 Manufacturing & Distribution of Gas sector, also is the 25th sector.

221P. He et al. / Energy Economics 50 (2015) 215–226

compute the normalized energy import resilience ρ = θ*/θ⁎ ⁎ (the compu- tation of θ⁎⁎ is further elaborated in Section 4.4). The third to seventh column in Table 2 presents maximum import loss event (loss levels and fuel type), and the last column shows the impacted sector when the maximum import loss event is realized. For instance, if the gas im- port loss hits 16.2 million TCE in the BAU scenario, demand deficit ap- pears in Gas Extraction sector. Our computations reveal that θ⁎ is the highest under the 2007 generation portfolio, followed closely by the COA, OIL and then lowest in the GAS scenario. For the purpose of sensi- tivity analysis, we also repeated the computations by using perturbed values on the oil-to-gas disaggregation ratio assumptions. The same re- sults in the ranking of the energy import resilience were obtained.

Table 2 shows that the import resilience achieved under the COA sce- nario is very close to that in the BAU scenario. This is not surprising since China's power generation portfolio is largely dominated by coal-fired power. An issue of important interest is the significant degradation in resilience when switching from a coal-based to gas-based generation portfolio. In particular, when the import loss in electricity sector is 14.6 million TCE in the GAS scenario, demand deficit will appear in the Manufacturing and Distribution of Gas (MDG) commodity because its production reaches its full capacity. However, if the same import loss event arises in, for example the COA scenario, the production capacity structure with generation by coal is able to mitigate the import loss so that no demand deficit arises. A closer analysis of this is described in the following.

In Table 3, left columns show production levels of selected industries in the all-gas power generation case (GAS) and all-coal generation case

Table 3 Production levels and interindustry sales of selected industries in (GAS) and (COA) when expo

Production level (billion CNY)

Industry GAS scenario (1) COA scenario (2) Difference (3) = (

1 Agriculture 4802.67 4799.71 −2.96 4 Metal mining 538.37 538.05 −0.32 6 Food products 4098.30 4094.57 −3.73 11 Coke & Petroleum 290.09 271.75 −18.34 12 Chemical 5843.97 5841.43 −2.53 13 Non-metal manu. 2212.81 2212.37 −0.44 14 Metal smelting 5656.39 5654.46 −1.93 16 Machinery 3657.24 3646.53 −10.71 17 Transportation 3187.57 3172.29 −15.28 22 Scrapers 399.21 399.14 −0.08 27 MDG 110.83 99.89 −10.94 29 Construction 6264.26 6264.21 −0.05 31 Post & Tele 69.93 69.89 −0.04 32 Computer 957.33 957.29 −0.03 34 Hotels 1421.04 1417.02 −4.02 38 R & D 126.26 126.26 0.00 41 Water Environment 823.58 821.37 −2.21 42 Education 1299.13 1298.88 −0.25 43 Health 1106.07 1099.48 −6.60 44 Culture, sports 339.42 339.08 −0.34 45 Public admin 1580.37 1580.32 −0.05

(COA), when exposed to the same import loss scenario (14.6 million TCE import electricity). The decrease in production levels in the COA case for these industries results in decreased consumptions of MDG commodity, as shown in the last three columns. The production levels of each of these sectors are lower in the COA compared to GAS scenario, including the MDG sector production. This immediately implies that the interme- diate consumptions of the MDG commodity in the GAS scenario are also higher, since final demands are the same in both scenarios. A closer in- spection of the technological coefficients of the industry sectors in Table 3 reveals that these are in fact major consumers of the MDG com- modity as inputs (i.e. sectors with high MDG (input) to output coeffi- cients). It also turns out that the Generation by Gas scenario requires relatively higher inputs of these sectors' production compared to the Generation by Coal scenario. In summary, when import electricity is re- duced by 14.6 million TCE in the either scenarios, the economy responds by increasing domestic power generation through its available technol- ogy and capacity. In the GAS scenario, gas-fired power generation re- quires high consumption of products that in turn consume high inputs of MDG sector, straining the ability of the sector to serve its final de- mands. This illustrates the important fact that the energy import resil- ience of the economy can be highly dependent on not just the choice of power generation technology, but more importantly on the inter- sectoral dependencies.

The above results reveal the important insight that, under the existing (2007) input–output structure, over-reliance on gas-fired gen- eration technology can considerably weaken the ability of the economy to mitigate energy import reductions. We now extend the study to

sed to the same import loss scenario (14.6 million TCE import electricity).

Sales of MDG commodity to industry (104 CNY)

2) − (1) GAS scenario (4) COA scenario (5) Difference (6) = (5) − (4)

18,136.70 18,125.52 −11.18 277,217.94 277,055.45 −162.49 222,254.87 222,052.59 −202.28 50,280.08 47,101.17 −3178.92

1,337,553.64 1,336,973.54 −580.11 242,748.37 242,699.83 −48.54 707,477.45 707,236.55 −240.90 278,946.60 278,129.71 −816.89 161,645.14 160,870.36 −774.78

7090.44 7089.05 −1.38 491,212.48 442,722.20 −48,490.27 67,725.87 67,725.31 −0.57 5071.07 5067.84 −3.23

45,737.05 45,735.49 −1.56 400,074.89 398,943.16 −1131.73

3464.13 3464.05 −0.08 54,137.13 53,991.89 −145.24

128,069.42 128,045.25 −24.18 67,250.06 66,849.06 −401.00 28,986.67 28,957.67 −29.00 49,955.90 47,575.52 −2380.37

Table 4 Energy resilience θ⁎ as coal-firing capacity is displaced by gas-firing in COA + GAS and BAU scenario.

Coal generation capacity decommissioned (% of coal in BAU) θ⁎ (ρ) in COA + GAS Ratio of coal to gas generation capacity in BAU case θ⁎ (ρ) in BAU

0% 0.2101 (0.5096) 83.89 0.2101 (0.5096) 10% 0.2101 (0.5096) 8.04 0.2101 (0.5096) 20% 0.2101 (0.5096) 3.78 0.2101 (0.5096) 30% 0.2101 (0.5096) 2.24 0.2101 (0.5096) 40% 0.1906 (0.4621) 1.46 0.2101 (0.5096) 50% 0.1692 (0.4102) 0.98 0.2101 (0.5096) 60% 0.1488 (0.3607) 0.65 0.1922 (0.4659) 70% 0.1278 (0.3098) 0.42 0.1712 (0.4150) 80% 0.1061 (0.2572) 0.25 0.1501 (0.3639) 90% 0.0866 (0.2099) 0.11 0.1289 (0.3125) 100% 0.0653 (0.1583) 0 0.1101 (0.2670)

222 P. He et al. / Energy Economics 50 (2015) 215–226

consider different levels of coal and gas in the generation mix. Table 4 shows the energy import resilience θ⁎ achieved as the switching from coal-fired to gas-fired capacity increases. In all these scenarios the total capacity of the power generation sectors are assumed to be equal to the 2007 value. First column shows percentage displaced based on 2007 portfolio. The results for the rows in ‘0%’ and ‘100%’ coal-to-gas switch correspond to the extreme positions discussed in the previous section. The column under the ‘COA + GAS’ refers to the scenarios in which only these two generation technologies are allowed non-zero ca- pacities. It is observed that up to 30% of coal generation can be displaced by gas without sacrificing the import resilience. Beyond that, θ⁎ is de- creasing with more gas-firing technology in the generation mix. Hence, while reducing coal-fired plants generation mix can reduce envi- ronmental impacts on the outset, there are critical levels of displace- ment that policy makers should be aware of, due to the economic input–output inter-dependencies.

The last two columns in Table 4 also show the numerical results for coal-to-gas switching starting from the 2007 generation portfolio (BAU scenario), which would probably be more representative in the real situation. The results obtained are qualitatively similar to the aforemen- tioned. Here up to 50% of coal can be displaced by gas without deterio- ration in import resilience. This improvement can be attributed to the greater diversification of the generation portfolio mix, which allows the economy to mitigate against a larger set of import loss scenarios. In a China energy study by EIA (2012b), coal-firing has already dropped to 66% of the total power generation mix in 2012, and is projected to reach 52% by 2040. This is reasonably agreeable with our recommenda- tions above from the import resilience point of view.

So far our computational studies strongly suggest that, from energy import resilience point of view, a coal-based generation portfolio out- performs that of gas-based portfolio in China. In retrospect, this is not surprising, and the first reason is as follows. Coal-fired plant technology is extremely matured and cost-effective in China. For example, a super- critical coal unit in China, at an overnight cost of roughly $600 kW−1, is around one-fifth to one-sixth of the cost of a new pulverized coal unit in the U.S. (EIA, 2010). In contrast, the Chinese government has in the past given much lower support to advance the development of natural gas generation technologies. All of China's gas turbines, for instance, are imported. As a result, the production cost of coal-fired plant is much lower than that of gas-fired plants (0.3 CNY/kWh for electricity produc- tion cost for coal, and 0.71 CNY for gas) (CGA, 2014). The higher techno- logical efficiency and low production costs of coal-fired generation imply lower intermediate consumptions, and as reasoned earlier allow more effective mitigation of import loss events. The second reason is more subtle, and is related to the inter-sectoral dependencies of the economy and sectoral level production technologies. In summary, ener- gy import resilience depends on the technological efficiencies and ca- pacities of some key sectors beyond that of the power generation sectors. This observation is also instrumental in designing policies to im- prove the energy import resilience, and is studied in detail in the next section.

4.3. Improving energy resilience

4.3.1. Production capacity expansion For the sake of illustration, we first consider again the all-natural gas

power generation position in Table 2 (i.e. ‘GAS’ scenario). The corre- sponding energy import resilience is 0.065, which is almost a 70% dete- rioration of the level achieved with an all-coal portfolio (0.21). The objective of the numerical studies in this section is to identify capacity expansion strategies to restore or improve the energy resilience level. More importantly, these studies can help improve understanding of the linkages between the economic input–output structure, production capacity and energy resilience.

At first sight, increasing the capacity of domestic primary energy pro- duction sectors (in particular, OIL&GAS sector) seems an obvious way to improve energy resilience. However, our computational studies reveal that there are in fact no improvements on the θ⁎ level made regardless of the levels of capacities added. We then performed capacity additions based on sectors facing demand deficits. Table 5 shows the results using this approach. For instance, when θ⁎ = 0.065 demand deficit oc- curs in the MDG sector (this is also the same in the results in Table 2). We thus introduce new capacity to this sector to mitigate the deficit. In our computations, it is also noted that introducing new capacity to all other sectors produce absolutely no improvements at this stage. Produc- tion capacity is added to MDG sector in small increments, and θ⁎ is recomputed until no further improvements are observed. Here, when MDG capacity is increased by 8%, the energy resilience reaches the limit of 0.1203. At this point, demand deficit shifts to the Healthcare sector, and only by improving the capacity of this sector, can θ⁎ be further in- creased (to 0.1289). Repeating this process, it is observed that the energy input resilience can be recovered to the initial level of 0.21.

A similar study with the BAU scenario as the baseline is shown in Table 6, where three different coal-to-gas mixes are assumed. These mixes cause deterioration in energy resilience θ⁎ as shown in Table (4), also reproduced in the second column in Table 6. Again, the goal of the exercise is to restore energy resilience to the original level of 0.2101 using incremental capacity expansions. Generally, the results show similar trends as in the previous case (Table 5). In particular, MDG sector capacity should be increased. Health care sector capacity should also be increased in order for the overall resilience level to be improved.

In summary, capacity expansions can improve the energy import re- silience of the economy, but only if these expansions are based on care- ful identification of the appropriate ‘choke-points’ or bottleneck sectors beyond the energy sectors. These bottlenecks arise due to the complex inter-sectoral dependencies, and their identification suggests to policy-makers that these sectors can highly influence national energy security, and hence may require further detailed analysis. Furthermore, these bottleneck sectors can shift as capacity additions are made. For ex- ample, in Table 5, although Oil & Gas Extraction is not a bottleneck pro- duction sector initially (the bottleneck is MDG), it turns out that the energy resilience cannot be improved beyond 0.209 unless slight capac- ity expansions are made in this sector subsequently.

Table 5 Improving energy resilience through sequential capacity expansions to achieve θ⁎ = 0.2101 in GAS scenario.

Sequence θ⁎(ρ) Sector with demand deficit Capacity expansion in sector and by percentage 1

0 0.0653 (0.1583) MDG MDG (8%) 1 0.1202 (0.2913) Health Health (1%) 2 0.1289 (0.3125) MDG MDG (21%) 3 0.2094 (0.5076) Gas Oil & Gas extraction (1%) 4 0.2101 (0.5096) Gas –

1 This column indicates the expansion performed to increase θ⁎ to the next higher level by sector and percentage of its current capacity.

223P. He et al. / Energy Economics 50 (2015) 215–226

4.3.2. Improving production technology efficiency We next consider the impact of production technology efficiency

improvements on the energy import resilience. Each column vector in the input–output technical matrix models a production technology, and a smaller value of an entry implies lower input requirement from the commodity of the corresponding row to produce one unit of output. We first consider the all-gas firing power generation position, and com- pute the energy import resilience levels θ⁎ (and ρ) when each column of technical coefficients is reduced by 5% across all rows from the 2007 values. Fig. 1 plots the normalized energy import resilience ρ on the ver- tical ordinate and the industry on the horizontal.

The results show that improving the production efficiency in the 12th (Chemicals) industry yields the largest improvement in θ⁎. This is followed by the 14th (Metal Smelting & Pressing), 29th (Construction), 25th (Generation by Gas), and 16th (Machinery & Equipment) industries. The reasoning for these results is suggested by Fig. 2, which shows the corresponding total intermediate consumptions of the MDG commodity in the event of a loss in imported electricity. This loss event was depicted in Table 2 in the scenario of an all-gas generation portfolio. Recall that in order to mitigate the reduction of imported electricity, the economy shifts to increase domestic power generation (by gas), hence straining the capacity of the MDG sector. This consequently results in a demand deficit in the MDG sector. Fig. 2 shows that, improving the technology of the above industries can significantly lower the total intermediate consumption of the MDG commodity during the occurrence of the elec- tricity import loss event. This frees up the production capacity of MDG to serve its final demands. Finally, the fact that improving the production efficiency of Chemicals is effective in reducing intermediate consump- tions of MDG (and hence improving energy resilience) also makes phys- ical sense, since according to the definition of the MDG industry (NBS, 2013), the outputs of this sector (e.g. syn-gas), are highly utilized in the production of chemicals and fuels.

We next perform similar computations when assuming a generation portfolio with different levels of coal-firing displaced by gas-firing from the 2007 generation mix (see Table 4). Fig. 3 shows the results when each technology efficiency is improved by 5%, for cases of 60%–80% coal displacement. For example, the 60% case (top line) represents that 60% of electricity generation by coal capacity in the 2007 generation portfolio (BAU) is displaced to generation by gas. As in the previous case, improving the technology efficiency in 12th (Chemicals) and 14th (Metal Smelting and Pressing) sectors yields the largest improvements of energy resilience. This is in fact consistent with real-world observa- tions. According to the 2007 data, these two industries are also among the two largest energy consumers and top energy-intensive industries

Table 6 Restoring the energy resilience to 0.2101 by incremental capacity additions under differ- ent coal:gas generation mix situations assuming 2007 (BAU) mix.

Coal to gas generation capacity ratio

Initial θ⁎(ρ)

Capacity expansion sequence and level (in parenthesis)

0.65 0.1922 (0.4659) MDG (3%) 0.42 0.1712 (0.4150) MDG (6%) 0.25 0.1501 (0.3639) MDG (7%), Health (1%), MDG (2%)

in China. These are clearly issues of national importance, and the Chi- nese government has targeted to reduce energy intensity of these in- dustries by 16% by the end of the 12th Five-Year Plan (MIIT, 2012), through various policies and instruments, including investment in technological and energy efficiency research. Hence, these strategies are coherent with our recommendations based on the energy import resilience.

4.4. Computing maximum resilience level θ⁎⁎

In this section, we show the results of solving the energy import re- silience optimization problem (28) for the maximum possible resilience level θ⁎⁎. The assumption for this model is that the total production capacities, i.e. ∑ j∈ J∑t∈T j x

t j are the same as that assumed in the basic

scenario studies in Section 4.2, based on the 2007 IO and energy data. For simplicity, no further constraints are imposed, which implies that the optimal solution corresponds to a rather idealistic situation if one had the complete freedom to re-design the entire economy by re- assigning all production capacities in all sectors without regards to other physical or socio-economic considerations. The constraint gener- ation method described in Section 3.4 is applied for solving the models (29) iteratively. The results are shown in Table 7.

The θ⁎⁎ level of 0.4215 corresponds to the ability of the economy (as- sumed in the model) to absorb a level of loss in import energies no greater than 86.3 million TCE with no demand deficits incurred. Al- though this is an idealistic calculation, the results imply that the econo- my will probably never be able to sustain reduction of more that half its energy imports (2007 levels) without other forms of intervention, which is rather reasonable. The worst case scenario in which this hap- pens, as shown in Table 7, is when natural gas imports (as of 2007 level) is totally cut off, and 71.1 million TCE of Coke & petroleum imports are unavailable. Any further losses of this import result in demand def- icits appearing in the Gas extraction sector.

Fig. 1. ρ achieved when improving 5% efficiency of each corresponding technology. Horizontal dashed line indicates the 0.1583 normalized energy resilience level in the baseline GAS case.

Fig. 2. Reduction in intermediate consumption of MDG when each technological coefficient of each industry is reduced by 5%. Vertical bars plot the corresponding reduc- tions of MDG consumption (from GAS case), when exposed to loss of 14.6 MTCE import electricity.

Table 7 Maximum import resilience θ⁎⁎, with corresponding import loss scenario. Units for energy import loss are in million TCE.

θ⁎⁎ Maximum import loss in commodity Demand deficit in commodity

Coal Oil Gas Coke Electricity

0.4125 0 0 16.4 71.1 0 Gas extraction

Table 8 Comparison of production capacities. The second column refers to solutions of Eq. (28), the third column refers to original 2007 data. Units in billion CNY.

Industry Optimized (w.r.t. (28)) Original (2007) Difference

1 48,677.5 48,893.0 −215.5 2 7720.4 10,241.9 −2521.5 3 31,399.5 11,918.6 19,480.9 4 6152.3 6149.3 3.0 5 3851.6 3851.6 0.0 6 41,615.7 41,790.4 −174.7 7 25,146.0 25,197.4 −51.4 8 17,995.2 18,072.6 −77.4 9 10,952.8 10,993.9 −41.1 10 14,807.4 14,933.0 −125.6 11 21,138.6 26,343.2 −5204.6 12 61,941.9 61,998.1 −56.2 13 22,783.6 22,804.4 −20.8 14 61,142.9 61,096.0 46.9 15 17,405.0 17,705.5 −300.5 16 39,880.0 39,486.6 393.4 17 32,520.2 32,978.4 −458.3 18 25,955.5 27,155.0 −1199.5 19 40,782.2 41,190.3 −408.0 20 4740.4 4879.7 −139.2 21 6138.3 6183.4 −45.1 22 4339.7 4366.0 −26.3 23 0.0 29,604.6 −29,604.6

224 P. He et al. / Energy Economics 50 (2015) 215–226

The optimized production capacity levels from the solution of problem (28) are shown in Table 8, where the production capacities based on the 2007 data are also tabulated for comparison. Interestingly, the idealized solution suggests that all power generation capacity should be assigned to oil, gas and renewable generation technologies, (i.e. 24th–26th sectors respectively), and there should be no power gen- eration using coal-firing (sector 23). Furthermore, significant capacity expansion of Oil & Gas (sector 3) is observed in the solution. However, at this point we should not take the model solution values too seriously and apply it out of context since more future work would be required to calibrate the model with respect to the actual economy (e.g. primary re- source reserves, capacity expansion limits of individual sectors). Most of these considerations can be captured as constraints in the set X in problem (28). In this work, our main objective is to propose the model- ing methodology and use θ⁎⁎ as an idealized benchmark.

4.5. Summary of key results and discussion

One of the main findings in our case study is that under China's input–output structure, the energy import resilience of the economy can be highly dependent on the choice of power generation technology in the future. In particular, over-reliance on gas-fired generation tech- nology is shown to considerably weaken the ability of the economy to

Fig. 3. ρ with efficiency increase by 5% in each technology under different generation portfolio scenarios.

mitigate energy import reductions. Hence, continually increasing gas- firing capacity alone to reduce the GHG emissions, will not be a sustain- able strategy based on our results. As China's long-term plan is to cut carbon intensity by 40 to 45% by 2020 compared to 2005 levels (NDRC, 2011), we give the following suggestions which can mitigate carbon intensity as well as enhance energy security in tandem: (1) To improve coal cleaning technology which will not only help reduce the GHG emissions but also help improve the efficiency of electricity gener- ation; (2) To develop high efficiency, environmentally-friendly coal- fired power generation technology such as ultra supercritical power generation technology, integrated gasification combined cycle, and gas turbine generation technology; (3) To continue developing highly effi- cient and affordable renewable electricity generation technologies such as hydraulic, wind, and solar. In fact, China has already formulated

24 1292.6 376.5 916.1 25 1240.2 352.9 887.3 26 26,323.9 5422.7 20,901.1 27 1098.1 1108.3 −10.2 28 1139.3 1178.8 −39.5 29 62,714.0 62,721.7 −7.8 30 31,156.6 31,700.1 −543.5 31 722.6 730.8 −8.2 32 9919.5 10,030.4 −110.9 33 28,670.7 28,832.5 −161.8 34 14,735.6 14,815.4 −79.9 35 19,059.2 19,481.0 −421.8 36 14,710.0 14,774.6 −64.6 37 11,585.7 11,784.6 −198.9 38 1383.9 1379.0 4.9 39 4400.1 4397.1 3.0 40 2142.4 2158.2 −15.9 41 8589.1 8754.4 −165.3 42 13,033.0 13,065.8 −32.8 43 11,065.6 11,122.6 −57.0 44 3493.1 3540.9 −47.8 45 15,817.2 15,817.6 −0.4 sum 831,378.9 831,378.9 0

225P. He et al. / Energy Economics 50 (2015) 215–226

detailed plans for achieving the energy targets in 12th Fifth Year Plan by developing new power generation technologies (NEA, 2011).

Another important insight from our case study in Section 4.2 is that the energy import resilience of the economy can be highly dependent on not just the choice of power generation technology, but more impor- tantly on the inter-sectoral dependency. In our study of China 2007 IO data, the interdependency among the sectors listed in Table 3 has a sig- nification impact on the energy import resilience. Hence, the Chinese government may be able to exploit this to enhance her energy security, by re-designing the inter-sectoral interdependencies. Moreover, we have shown that capacity expansions can improve the energy import resilience only if these are based on careful identification of the critical bottleneck sectors beyond the energy production sectors. In reality, pol- icy makers need to consider the marginal returns of energy resilience improvement by investing in production capacity expansion. For exam- ple in Table 5, investing in 8% capacity expansion in Manufacturing and Distribution of Gas when θ⁎ is 0.0653 can help potentially improve the resilience indicator level by about 84%. Whether such a direction of ex- pansion in the nation's industry is appropriate will need to be discerned in tandem with other factors such as employment, resources, trade and other economic issues.

Furthermore, we have shown that improving the technology effi- ciency in certain sectors can yield improvements of energy resilience. Recall that because smaller technical coefficients in IO table imply that producing the same unit of that commodity requires less inputs from other sectors (including energy), hence improvement of sectors' pro- duction technology efficiency can also be measured by energy intensity (Energy Consumption per Unit of GDP) of a sector. For the past decade (during the 11th and 12th FYP), China has been seeking to improve en- ergy resilience by developing and promoting energy-saving and high energy efficiency technologies in all industries. As a result, the energy intensity has been reduced by 19.1% at the end of 11th FYP and by 3.53% on average from 2011 to 2014 (NBS, 2014). Based on our results, Chemicals industry and Metal smelting & pressing industry are among the top energy-intensive industries in 2007 thus need to be prioritized in efficiency improving, which is consistent with the policies and goals in the 12th FYP.

5. Conclusion

In this work, we have proposed a novel approach integrating linear programming and IO analysis to evaluate the energy import resilience index for an economy. We demonstrate through the use of China's IO data, that the proposed methodology can be effectively used to identify the critical bottlenecks of an economy in the event of energy import losses. As the largest energy consumer in the world, the Chinese govern- ment is deeply interested in strategies to achieve a high level of energy security. The models developed in this work serve as an important decision-support for policy-makers to study the impacts of energy re- ductions on the supply–demand balances of the various economic sec- tors. It can also guide where government budget should be invested, in particular in capacity expansions and technology improvements. Fu- ture research includes enabling the proposed model for multi-regional studies, cross-country comparisons, and also extensions to a dynamic setting. In particular, it is of interest to investigate the medium to long term recovery trajectory of the economy after the import loss event oc- curs through investments in capital stocks.

Acknowledgments

The research published here was conducted at the Future Resilient Systems at the Singapore-ETH Centre (SEC). The SEC was established as a collaboration between ETH Zurich and National Research Founda- tion (NRF) Singapore (FI 370074011) under the auspices of the NRF's Campus for Research Excellence and Technological Enterprise (CREATE) programme.

The authors are also thankful to the anonymous referees for their constructive comments to improve the work.

Appendix A. Supplementary data

Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.eneco.2015.05.007.

References

Akins, J.E., 1973. The oil crisis: this time the wolf is here. Foreign Aff. 462–490. Ben-Tal, A., El Ghaoui, L., Nemirovski, A., 2009. Robust optimization. Ch. 14. Robust

Ajustable Multistage, Optimization. Princeton University Press, pp. 357–413. CGA, 2014. Generati on cost by natural gas doubles or triples than that by coal. China Gas

Association URL http://www.chinagas.org.cn/hangye/news/2014-05-08/15971.html. Chenery, H.B., Clark, P.G., 1960. Interindustry economics. Published Nav. Res. Logist. Q. 7

(4), 385–391. Dantzig, G.B., 1949. Programming of interdependent activities: Ii mathematical model,

Econometrica. J. Econ. Soc. 200–211. Dantzig, G.B., 1963. Linear Programming and Extensions. Princeton University Press. De Mesnard, L., Miller, R.E., 2006. A note on added information in the RAS procedure: re-

examination of some evidence. J. Reg. Sci. 46 (3), 517–528. EIA, 2010. Updated capital cost estimates for electricity generation plants. Washington,

DC, Energy Information Administration. EIA, 2012a. Annual Energy Review 2012. Energy Information Administration, Washington,

DC. EIA, 2012b. China Energy Review, 2012. Energy Information Administration, Washington, DC. EIA, 2014a. Natural gas serves a small, but growing, portion of Chinas total energy de-

mand. Energy Information Administration URL http://www.eia.gov/todayinenergy/ detail.cfm?id=17591.

EIA, 2014b. China Energy Review, 2014. Energy Information Administration, Washington, DC. S. E. Flynn, 2012. Powering Americass energy resilience. Center for National Policy. Geng, J.-B., Ji, Q., 2014. Multi-perspective analysis of China's energy supply security. Ener-

gy 64, 541–550. IEA, 2014. Energy Supply Security: The Emergency Response of IEA Countries — 2014 Edi-

tion, Chapter 1. IMF, 2012. People's Republic of China: Staff Report for the 2012 Article IV Consultation. James, D., deL Musgrove, A., Stocks, K., 1986. Integration of an economic input–output

model and a linear programming technological model for energy systems analysis. Energy Econ. 8 (2), 99–112.

Just, J.E., 1974. Impacts of new energy technology using generalized input–output analy- sis. Comput. Oper. Res. 1 (1), 97–109.

Kerschner, C., Hubacek, K., 2009. Erratum to assessing the suitability of input–output analysis for enhancing our understanding of potential effects of peak-oil. Energy 34 (10), 1662–1668.

Kruyt, B., van Vuuren, D.P., De Vries, H., Groenenberg, H., 2009. Indicators for energy se- curity. Energy Policy 37 (6), 2166–2181.

Le Coq, C., Paltseva, E., 2009. Measuring the security of external energy supply in the European Union. Energy Policy 37 (11), 4474–4481.

Leung, P., Hsu, G.J., 1984. An integrated energy planning model for Hawaii. Energy Econ. 6 (2), 117–121.

Libert, B., Orolbaev, E., Steklov, Y., 2008. Water and energy crisis in Central Asia. China and Eurasia Forum Quarterly vol. 6, pp. 9–20.

Löschel, A., Moslener, U., Rübbelke, D.T., 2010. Indicators of energy security in industrialised countries. Energy Policy 38 (4), 1665–1671.

MIIT, 2012. 12th five-year-plan for petro-chemical & chemical industry. Ministry of In- dustry and Information Technology of the People’'s Republic of China.

Molyneaux, L., Wagner, L., Froome, C., Foster, J., 2012. Resilience and electricity systems: a comparative analysis. Energy Policy 47, 188–201.

NBS, 2007. 2007 Input–output Table of China. China Statistics Press, Beijing. NBS, 2013. China Input–output Table Sector Classification URL. http://www.stats.gov.cn/

tjsj/tjbz/hyflbz/201310/P020131023305612873923.pdf. NBS, 2014. China Input–output Table Sector Classification URL. http://www.stats.gov.cn/

tjsj/sjjd/201503/t20150308_690781.html. NDRC, 2011. 12th fyp for Chinas national economy and social developments. National De-

velopment and Reform Commission of the People’s Republic of China URL http:// news.sina.com.cn/c/2011-03-17/055622129864.shtml.

NEA, 2011. the twelfth five-year plan for national energy technology development URL http://www.nea.gov.cn/131398352_11n.pdf. National Energy Administration of China.

OBrien, G., Hope, A., 2010. Localism and energy: negotiating approaches to embedding re- silience in energy systems. Energy Policy 38 (12), 7550–7558.

Penn, J., Irwin, G.D., 1977. Constrained input–output simulations of energy restrictions in the food and fiber systems. National Economic Analysis Division, Economic Research Service. Agricultural Economic Report No. 280.

Rose, A., Benavides, J., Chang, S.E., Szczesniak, P., Lim, D., 1997. The regional economic im- pact of an earthquake: direct and indirect effects of electricity lifeline disruptions. J. Reg. Sci. 37 (3), 437–458.

Schluter, G., Dyer, D., 1976. The economic interpretation of constrained input–output so- lutions. Rev. Econ. Stat. 245–248.

Thomas, S., Kerner, D., 2010. Defense Energy Resilience: Lessons from Ecology, No. 39. Strategic Studies Institute.

226 P. He et al. / Energy Economics 50 (2015) 215–226

Tunkay, C., Bilge, O., 2012. Input–output analysis with linear programming: the case of Turkey. Int. Res. J. Finance Econ. 89, 138–145.

Vivoda, V., 2009. Diversification of oil import sources and energy security: a key strategy or an elusive objective? Energy Policy 37 (11), 4615–4623.

Vogstad, K.-O., 2009. Input–output analysis and linear programming. Handbook of Input– output Economics in Industrial Ecology. Springer, pp. 801–818.

Wang, T.-F., Miller, R.E., 1995. The economic impact of a transportation bottleneck: an in- tegrated input–output and linear programming approach. Int. J. Syst. Sci. 26 (9), 1617–1632.

Wilting, H.C., Faber, A., Idenburg, A.M., 2008. Investigating new technologies in a scenario context: description and application of an input–output method. J. Clean. Prod. 16 (1), S102–S112.