Review on Energy Resilience

Renewable and Sustainable Energy Reviews 62 (2016) 32–45

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews

http://d 1364-03

n Corr E-m

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

Bayesian networks in renewable energy systems: A bibliographical survey

Mónica Borunda a,b,n, O.A. Jaramillo c, Alberto Reyes b, Pablo H. Ibargüengoytia b

a CONACYT Research Fellow, Consejo Nacional de Ciencia y Tecnología. Av. Insurgentes Sur 1582, Col. Crédito Constructor, D.F., México 03940, Mexico b Instituto de Investigaciones Eléctricas. Reforma 113, Col. Palmira, Morelos, 62490, Mexico c Instituto de Energías Renovables, Universidad Nacional Autónoma de México. Priv. Xochicalco S/n. Temixco, Morelos, 62580, Mexico

a r t i c l e i n f o

Article history: Received 30 September 2015 Received in revised form 8 February 2016 Accepted 12 April 2016 Available online 28 April 2016

Keywords: Renewable energy Sustainable energy Bayesian networks Dynamic Bayesian networks Artificial intelligence Probabilistic graphical models

x.doi.org/10.1016/j.rser.2016.04.030 21/& 2016 Elsevier Ltd. All rights reserved.

esponding author at: Instituto de Investigacio ail address: [email protected] (M. B

a b s t r a c t

For the last years, the research and development in the field of Renewable Energy has been growing due to the need of Renewable Energy as an extended and reliable source of energy. However, the imple- mentation of renewable energy has many complex problems not easily solved with conventional methods. Recently, Artificial Intelligence techniques such as Artificial Neural Networks, Fuzzy Logic and Genetic Algorithms, have been widely used to deal with these problems in the field of Renewable Energy. Nevertheless, issues with a degree of uncertainty need Bayesian Networks since this is one of the most effective theories to face them. This technique can contribute to the Renewable Energy harnessing and other open issues on this field. In this work we show the state of the art of the applications of Bayesian Networks in Renewable Energy, such as solar thermal, photovoltaic, wind, geothermal, hydroelectric energies and biomass. Additionally, we include related topics such as energy storage, smart grids and energy assessment. We classify the literature by areas considering three main subjects: resource eva- luation, operation, and applications, and in each section we describe the possible directions to be taken in the research of the field. We find that the main applications are done for forecasting, fault diagnosis, maintenance, operation, planning, sizing and risk management. We conclude that Bayesian Networks are a promising tool for the field of Renewable Energy with potential applications due to their versatility.

& 2016 Elsevier Ltd. All rights reserved.

Contents

1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 2. Bayesian networks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.1. Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 2.2. Applicability of BNs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 2.3. Extensions of BNs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3. Solar thermal energy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 3.1. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 3.2. Further steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4. Photovoltaic energy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 4.1. Resource evaluation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 4.2. Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 4.3. Further steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

5. Wind energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 5.1. Resource evaluation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 5.2. Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 5.3. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 5.4. Further steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

nes Eléctricas. Reforma 113, Col. Palmira, Morelos, 62490, Mexico. Fax: þ52 777 362 3808. orunda).

M. Borunda et al. / Renewable and Sustainable Energy Reviews 62 (2016) 32–45 33

6. Geothermal energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 6.1. Resource evaluation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 6.2. Further steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

7. Hydroelectric energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 7.1. Resource evaluation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 7.2. Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 7.3. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 7.4. Further steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

8. Biomass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 8.1. Resource evaluation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 8.2. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 8.3. Further steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

9. Energy storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 9.1. Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 9.2. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 9.3. Further steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

10. Smart grids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 10.1. Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 10.2. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 10.3. Further steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

11. Energy assessment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 11.1. Energy market . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

11.1.1. Further steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

11.2. Energy efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

11.2.1. Further steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

12. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Acknowledgments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

1. Introduction

In the field of renewable energy, the first approach to deal with a problem is using analytical methods. However, when analytic methods are not enough, numerical methods become crucial to handle more complex and realistic scenarios. Nevertheless, numerical methods are limited as well, for example, in the number of variables they can deal with, the complexity of the problem and the computing time. Therefore, Artificial Intelligence (AI) opens a window to help in the solution of problems where analytical and/or numerical methods are not enough and it provides tools for the prediction, modeling, opti- mization and control of renewable energy complex processes.

The main techniques in Artificial Intelligence, namely Artificial Neural Networks (ANN) [1–3], Genetic Algorithms (GAs) [4–6], Fuzzy Logic [7,8], and hybrid models – which combine two or more tech- niques, have already been applied to solve many problems in renewable energy. The research in this direction is expanding very quickly and a big quantity of works have already been done [9–11]. For instance, research has been done in the applications of ANN in the prediction of meteorological variables, such as solar radiation, tem- perature and wind, which are relevant to many renewable energy processes [12,13]. Likewise applications in Photovoltaic (PV) and Solar Thermal Energy systems [14,15] as well as many other renewable energy fields have also been performed [16]. GAs have been used for optimization of Solar Thermal systems [17] and other renewable energy systems [18]. Applications of Fuzzy Logic in renewable energy, such as site assessment for installing photovoltaic facilities and wind farms, power tracking in solar photovoltaic/wind devices and opti- mization, have also been developed [19].

Most of the AI techniques aim to reason logically, indeed Logicism dominated AI in the decades of 1960s and 1970s [20]. Nevertheless, there is a big amount of systems with high degree of uncertainty, i.e., incomplete evidence leading to beliefs that lead to short knowledge and therefore incorrect conclusions. In these systems logic is not enough because uncertain reasoning is a key feature of the problem. Bayesian networks (BNs), which represent probabilities, is one of the

most effective theory models in the uncertainty knowledge repre- sentation field [21] and one of the best candidates for simplifying conditionalization, for planning decisions under uncertainty, and for explaining the outcome of stochastic processes.

This work is organized as follows: the next section is dedicated to Bayesian Networks. We present a quick and basic description of BNs and we show their applicabilities as well as their extensions. The next 9 sections show the applications of BNs already done in renewable energy, i.e., solar energy, photovoltaic energy, wind energy, geothermal energy, hydroelectric energy, biomass, energy storage, smart grids and energy assessment. We classify the utilization of BNs in (a) resource evaluation according to energy assessment and availability, (b) operation during the energy conversion performed by the system and, (c) applications to enhance the implementation of the system. As it is shown below, only in few renewable energy fields the three kinds of applications have been exploded. Every subsection begins with a brief summary of the work done in that field, followed by the citations of the corresponding references with a brief description of the work and ends by highlighting the open issues and possible contributions to be done in the field. It is important to point out that at the end of each subsection we summarize the state of the art in a table to emphasize the lacks and highlights in the literature. Finally, we conclude by pointing out our final remarks in Section 12.

2. Bayesian networks

2.1. Formulation

Bayesian networks (BNs) [22] is a technique used in artificial intelligence to deal with problems with uncertainty [23,24]. In particular, BNs are aimed to solve problems with:

� uncertainty due to imperfect understanding, complexity, contra- dictory knowledge between experts or incomplete knowledge,

Table 1 A Priori probabilities of the node X1.

X1

SX1;1 PðX1 ¼ SX1;1Þ SX1;2 PðX1 ¼ SX1;2Þ

Table 2 CPT of the node X2 given the node X1.

X1 SX1;1 SX1;2

X2 SX2;1 PðX2 ¼ SX2;1 jX1 ¼ SX1;1Þ PðX2 ¼ SX2;1jX1 ¼ SX1;2Þ SX2;2 PðX2 ¼ SX2;2 jX1 ¼ SX1;1Þ PðX2 ¼ SX2;2jX1 ¼ SX1;2Þ

M. Borunda et al. / Renewable and Sustainable Energy Reviews 62 (2016) 32–4534

� randomness due to stochastic phenomena, indeterminism, lack of patterns or predictability,

� both.

BNs have been applied to solve problems in various fields. As an example of pioneer works there are applications in medical diagnosis [25,26], map learning [27], language understanding [28,29], vision [30], heuristic search [31], environmental issues [32–35], watershed management [36,37], virtual sensors [38,39] to name some.

A Bayesian network is basically a directed acyclic graph, i.e., a graph made of nodes connected by edges with a direction asso- ciated with them, with no cycles. This graphical structure allows us to represent and reason about an uncertain domain. The nodes of the network represent a set of random variables, X ¼ X1; …Xi; …Xn, which are pair linked with connecting arrows, Xi-Xj, representing the dependence between variables, and each node has associated a conditional probability table that quantifies the probabilistic relation the parent nodes have on the children nodes.

Therefore a BN consists of 2 parts: the structural one and the parametric or quantitative one. The structure, or topology, of the network encodes the qualitative relationships between variables and it is defined by a set of nodes and a set of directed arcs. Nodes are generally discretized by Boolean or scaled values. And con- tinuous variables are usually discretized. Two connected nodes means that one (ancestor) affects or causes the other (descendant), and the arc indicates the direction of the effect. The BNs structure follows the Markov Property which states that all direct depen- dencies in the system are explicitly shown via arcs. The absence of an arc denotes independence between two arcs.

Once the topology of the BN is specified the next step is to quantify the relationships between connected nodes. These rela- tionships are specified by a conditional probability distribution for each node, and in the case of discrete variables they correspond to a conditional probability table (CPT). The CPT is constructed such that (a) each row contains the conditional probability of each node value for each possible combination of values of its parent nodes; (b) each row must sum to 1; and (c) a node with no parents has one row (the prior probabilities). In particular, in the case of Boolean variables with n Boolean parents the table contains 2nþ1

parameters. For instance, consider two nodes X1 and X2 with two states, Sn;1 and Sn;2, each as shown in Fig. 1.

The a priori probabilities of node X1 are defined in Table 1. Then a CPT defined by the conditional probabilities PðX2jX1Þ is

associated to node X2 to define the probability distributions over the states of X2 given the states of X1 as shown in Table 2.

Once a domain and its uncertainty is represented by a BN, the BN reasons about the domain via a flow of new information, evi- dence, i.e., a probabilistic inference system. This system consists of computing the posterior probability distribution for a set of query nodes, given values for some evidence or observation nodes. There are different kinds of evidence: (a) specific evidence: a definite finding that node X has a particular value x; (b) negative evidence: a finding that node Y is not in state y1 but may take any other values and; (c) virtual or likelihood evidence: source of informa- tion which is not sure about it. There are four types of reasoning using BNs: (a) diagnostic: reasoning from symptoms to cause; (b) predictive: reasoning from new information about causes to

Fig. 1. Basic example of BN.

new beliefs about effects; (c) intercausal: reasoning about the mutual causes of a common effect and (d) combined. In the example described above the BN inference computes the marginal distribution PðX2 ¼ SX2;1Þ: PðX2 ¼ SX2;1Þ ¼ PðX2 ¼ SX2;1jX1 ¼ SX1;1ÞPðX1 ¼ SX1;1Þ

þPðX2 ¼ SX2;1jX1 ¼ SX1;2ÞPðX1 ¼ SX1;2Þ: ð1Þ Therefore a BN computes the probabilities attached to a node state given the state of one or several variables, becoming a powerful modeling tool for complex systems.

BNs use a general inference mechanism to collect and incor- porate the new information and evidence gathered in the study through Bayes' theorem. The Bayes theorem defines the condi- tional probability of x given y as follows

PðxjyÞ ¼ PðxÞPðyjxÞ PðyÞ ; ð2Þ

where x and y are events, P(x) and P(y) are the probabilities of x and y, PðxjyÞ is the conditional probability of x given y and PðyjxÞ is the conditional probability of y given x. In this way, BNs update a set of events probabilities according to the observed facts and the BN structure. Depending on the situation there is a inference algorithm appropriated to it.

2.2. Applicability of BNs

BNs are useful in the solution of many problems involving prediction, data analysis and updating, diagnosis, optimization, deviation detection, and decision making based on the best information available [40]. In particular, the main applicability fields of BNs are:

� Providing global reliability estimation – BNs permit one to incorporate different kinds of knowledge in one model such as data from feedback experience, expert's knowledge such as logical rules, equations or probabilities, the behavior of the system either through a functional or dysfunctional analysis, and observations.

� Study and analyze complex systems – BNs establish cause–effect relationships between all kinds of factors involved in the model, and therefore modeling of the interactions is performed.

� Dependability – BNs are capable of providing a prediction of a parameter which is an input data for a decision step. Some aspects such as multi-state elements, failures dependencies, system redundancy, dynamic evolution and operation condi- tions have to be taken into account. Dependability analysis is used to deal with reliability, availability and maintainability [41].

� Risk analysis – BNs are able to identify, characterize, quantify and evaluate critical or hazard event occurrence including the

M. Borunda et al. / Renewable and Sustainable Energy Reviews 62 (2016) 32–45 35

estimation of its likelihood and its consequences due to its capability to quantify low probability events [42].

� Maintenance – BNs are good for taking into account the main factors involved in the maintenance concept: (a) factors that technically describe each system to be maintained, (b) factors that describe the interrelations between the different systems and (c) factors that describe the general organizational structure [43,44]. Direct applications concern maintenance decisions and performance evaluations.

� Fault diagnosis – BNs can be constructed such that all variables and information involved in the process and in the mechanical equipment during a fault process are encoded in the network structure since they are very suitable for dealing with compli- cated problems [45,46].

The strength of BNs lies in their polyvalence which allow them to deal with different issues and their graphical representation allows one to understand the model complexity in a single view. On the other hand, their weak point is that there is no specific semantic to guide the model development and to guarantee its coherence. Therefore, verification and validation of the model with the system reality must be done [47,48].

2.3. Extensions of BNs

The following networks are some extensions of BNs:

� Decision Networks (DNs) or Influence Diagrams: They are made for decision making. DNs are modified BNs which include variables for decision, management, utility, or benefit-cost problems. They are a useful tool for describing decision process into diagrammatic form, holding relationships between vari- ables and analysing the expected effects of management deci- sions taking into account their associated uncertainties [49–51].

� Dynamic Bayesian Networks (DBNs): They are in charge of rea- soning about changes over time. BN variables have no time dependence, they are considered at a fixed point in time. However, DBNs take into account how the variables change with time. DBNs are defined by (a) a prior network, which represents the prior probabilities for all of the variables in the network at t¼0, and (b) a transition network, which encodes for all time slices, t ¼ 1; 2; …; n, the probabilities for each variable conditioned on other variables [52–54]. For a DBN the set of variables and probability distributions are the same for each time slice, with the exception of the network in the initial time slice, which has its own probability distribution.

Table 3 Current state and future directions of BNs in solar thermal power.

Solar thermal power

Highlights Lacks

Resource evaluation Forecasting the solar resource Solar power generation prediction

Operation Thermal load prediction Fault diagnosis Control Management Maintenance

Applications Fault detection in solar power plants

Sizing

Fault diagnosis in solar heat pump systems

Optimization

3. Solar thermal energy

Currently, the applications of BNs in the field of solar thermal energy are incipient. The reported research consists basically of two works, mainly dealing with fault diagnosis as described below.

3.1. Applications

Coleman and Zalewski [55] presented a fault detection and diagnostics system for solar power plants. They used BNs to study the causes of the faults detected. They tested their system by implementing it to detect faults in real time.

Liu et al. [56] used BNs to analyze fault diagnosis for a solar assisted heat pump system. The learning method used to obtain the BNs parameters was based on the back-propagation neural network (to impute the missing data) and the maximum like- lihood estimation. The parameters of the BN were quantified by

Fuzzy set theory and the resulting BN was able to perform fault diagnosis with complete or incomplete data.

3.2. Further steps

Given the previous work we suggest the following applied research lines. In the particular case of solar thermal energy used for electricity generation, BNs can contribute to forecast the solar resource in order to predict the generated power, and to predict and monitor the thermal load management. Likewise, the fault diagnosis methodology can be applied to solar thermal devices at low, medium or high temperatures. A new research area in this topic is solar heat for industrial processes with open issues such as the prediction of the necessary thermal load demand, the sizing and optimization of the system, and the control, management and maintenance of the operation of the plant. Table 3 summarizes the current state of the research done in this area and new lines of inquiry.

4. Photovoltaic energy

In the field of photovoltaic energy we found only two works, one related to resource prediction and the other related to the control for the operation of a solar photovoltaic power plant as described below.

4.1. Resource evaluation

Dong et al. [57] worked a probabilistic forecast approach using DBN to predict the probability density function of PV power pro- duction in China. Their aim was to help the system operator on the generation scheduling and dispatch by predicting short-term PV power production. Their simulations used the meteorological variables that have more influence on the PV power production level and were performed on a realistic photovoltaic system.

4.2. Operation

Oviedo et al. [58] designed a MultiAgent system to improve automatic remote control of solar power plants by either auto- matically triggering actuators or given recommendations to operators. They built the agent with a variety of techniques such as Expert Systems, Neural Networks and Bayesian Networks, to mimic the reasoning process in human operators.

M. Borunda et al. / Renewable and Sustainable Energy Reviews 62 (2016) 32–4536

4.3. Further steps

As in the case of solar thermal energy this field presents many opportunities to apply BNs. As before, the resource prediction is crucial for the operation of a photovoltaic system due to the time response during the operation and it is also crucial for the pre- diction of the generated power. On the other hand, fault diagnosis is very important in photovoltaic systems and a frequent and not easily solved problem in power plants. As before, sizing and optimization of the photovoltaic power plant, as well as manage- ment and maintenance, are open areas for BNs. BNs are also useful to plan the generation dispatch and to monitor and control the fluctuations during the power generation. Table 4 summarizes the current state of the research done in PV power and new lines of inquiry.

5. Wind energy

One of the most outstanding applications of BNs in Renewable Energy is in the field of wind energy. As we see below there are many works in resource prediction, operation, failure diagnosis, maintenance and risk management of wind turbines and wind farms, onshore and offshore, and in other applications.

5.1. Resource evaluation

Cano et al. [59] used models with Bayesian Networks in Meteorology as a data mining techniques. One of their applications was a weather generator, they generated rain values with their techniques for 200 days. They also applied their method to fill missing data or remove incoherent data and illustrated it in a wind speed analysis. Finally their techniques also provided local weather forecast.

De la Torre-Gea et al. [60] used Bayesian Networks to deter- mine spatial and temporal dependencies among climatic variables. They considered incomplete data sets of temperature, dew point, humidity, pressure and wind speed, from three years and three sites in the surroundings of Querétaro, México. They found the following: (a) for humid temperature climate, the atmospheric pressure increases as the air becomes cooler; (b) for semi-dry temperature climate, the precipitation and the maximum wind speed are impacted by the maximum dew point; and (c) for warm humid climate, there are no influences between dew point and humidity, and the pressure is autonomous from the temperature.

Carta et al. [61] estimated long-term mean wind speed histo- gram with a method based in probabilistic Bayesian networks

Table 4 Applications of BNs in PV power, research done and things to do.

Photovoltaic energy

Highlights Lacks

Resource evaluation PV power generation prediction

Forecasting the solar resource

Operation Load prediction Fault diagnosis

Control Management Maintenance Monitoring and control of power fluctuations

Applications Sizing Optimization Planning of dispatch

using long histories of wind speed and wind direction measure- ments from Spain. They compared the errors between the real and the predicted estimated long-term mean wind turbine energy output, computing the last one with a BN with three reference stations and using two measure-correlated-prediction algorithms. They found that the BN method results are better.

Ibarguengoytia et al. [62] developed a dynamic Bayesian net- work for a short-term 5 h forecast of wind velocity. They con- sidered all relevant variables for wind production and generated a probability distribution for the predictions in order to provide information for decision making. Their method was validated experimentally with data from a wind farm in Mexico.

5.2. Operation

Zitrou et al. [63] used Dynamic Bayesian Networks to support decisions regarding the operation and maintenance of offshore wind turbines. In particular, they focused in the system deterioration.

Nielsen [64] presented a decision tool for operation and maintenance of offshore wind turbines. The tool uses influence diagrams, which are graphical representations of a decision tree based on Bayesian Networks, to assist the rational decision maker. The tool provides the expected utilities for decision alternatives which are useful to find the optimal strategy.

Nielsen et al. [65] used Bayesian networks to provide optimal decisions for repairing offshore wind turbines. The optimal deci- sion, made with Bayesian decision theory, was made such that the preventive maintenance effort is balanced against the costs of corrective maintenance.

Dinwoodie et al. [66] developed a combined operational and strategic decision support model for offshore wind operations to explore various operating scenarios and determine optimal oper- ating strategies and associated risks. They simulated the operating costs and lost revenue in MATLAB based on wind farm specifica- tions, climate and operating strategy. Then, Bayesian Belief Net- works and decision trees performed the decision analysis. They applied their model to a case study where the examination of different failure rates and alternative electricity price scenarios were considered.

Kougioumtzoglou et al. [67] developed a technique for pre- dicting offshore wind farms reliability using HAZID (Hazard Identification methods), FMECA (Failure Mode, Effects and Criti- cality Analysis) and Bayesian Networks. They presented a risk analysis methodology including the full-scale installation, the operation and maintenance, the personnel safety, the environ- mental impact, the asset integrity and the operation. Finally they found the hazards in the installation, operation and maintenance activities.

5.3. Applications

Chen and Hao [68] established the fault model of speed-up wind turbine gearbox. The conditional probability of the sub- nodes was defined by the conditional probability relationship of different nodes. On the other hand, the fault probability was found by the conditional independence of each node and by simplifying the probability distribution. The model was proved by a calcula- tion case on a test-platform showing that it improves the fault diagnosis and therefore the operation level of a wind turbine.

Plumley et al. [69] made an on-line condition monitoring sys- tem to diagnose wind turbine component conditions. Using a Dynamic Bayesian Network the system simulates the degradation of wind turbine components and calculates the cost of main- tenance strategies. The main application is to use it to reduce the cost of offshore operating wind turbines. They successfully tested

M. Borunda et al. / Renewable and Sustainable Energy Reviews 62 (2016) 32–45 37

the system with a gearbox oil testbed and the recognition of failure modes and anomalous sensor readings was validated.

Chen et al. [70] presented an approach to describe the rela- tionship between wind turbine failure main causes and symptoms with Bayesian networks. The aim was to reduce wind turbine operation and maintenance costs by detection, diagnosis and prognosis. They modeled their BN with results of a Venn diagram analysis, [71], and used 26 months SCADA data to train it.

Wang et al. [46] performed a wind turbine generator dynamic reliability test system based on feature recognition. They used the Bayesian network fault diagnosis method to construct the vibra- tion feature recognition system, of the fault signals vibration characteristics of the gearbox and the spindle, which are uncertain and fuzzy. The system, made of sensor technology, GPRS com- munication, PLC control system and artificial intelligence techni- ques, automatically controls and monitors the wind turbine gen- erator to ensure its safe and stable operation.

Dai et al. [72] studied the risk of collision between service vessels and offshore wind turbines. They established Bayesian networks with the relevant risk-influencing factors. They simu- lated seven collision scenarios and their consequences were ana- lyzed. They identified the critical values of force and energy for structural damage in each scenario as well as the critical vessel speeds.

Pan et al. [73] presented a reliability assessment method for wind turbine generators. They built a fault tree of wind turbine generator electrical components and transformed it to a Bayesian network. Using a Markov chain Monte Carlo inference they cal- culated the probabilistic distribution and reliability was given based on the probabilistic input.

Shuang [74] constructed a risk management model for wind power plants using Bayesian Networks.

Pattison et al. [75] constructed an intelligent, autonomous system using Random Forests (for condition monitoring), dynamic Bayesian networks (for reliability and maintenance modeling) and mimetic algorithms (for maintenance scheduling) together with one year database to automatically provide maintenance to an offshore wind farm consisting of over 100 turbines. The system automatically detects the faults, models and updates the turbines' survivability and creates a hierarchical, optimal program of maintenance actions focused to maximize the wind power generation.

Table 5 Applications of BNs in wind power, research done and things to do.

Wind energy

Highlights

Resource evaluation Weather data generation Weather forecasting Analysis of climatic variables Wind forecasting Estimation of long-term mean wind speed and wind power

Operation Support for operation, maintenance and repairing Prediction of wind farm reliability Maintenance of offshore wind farms

Applications Fault model of speed-up wind turbine gearbox Causes and symptoms of wind turbine failure Diagnosis of wind turbine's component conditions Optimization of maintenance strategies Reliability of wind turbine generator Risk collision between service vessels and offshore wind turbine Risk management of wind power plants Risk assessment in wind turbines

Li et al. [76] proposed a reliability and simulation model for a wind turbine system based on the Goal Tree, Success Tree and Master Logic Diagram (GTST-MLD) framework. They modeled the relationships among components and functions in a wind turbine system, and the impact of factors and mechanisms influencing the failure of the components. They integrated the model and used Monte Carlo simulations to compute the wind turbine system reliability. On the other hand, they used a Bayesian network of the wind turbine system to satisfactorily validate the numerical results.

Ashrafi et al. [77] provided a review of existing techniques for risk assessment in complex technology and focused in wind tur- bines. They proposed an effective framework to assess risk and reliability in a wind turbine through a BN considering structural, electrical and mechanical components interacting with human resources and natural, political, economic and social environ- mental factors.

5.4. Further steps

Given the extensive work in the field we would like to point out that another opportunity for the application of BNs to the field is in the wind turbine design for small scale as well as wind farms design with large wind turbines. BNs can also be used to avoid the grid code specification violation in the electric network and to smooth wind power fluctuations. Prediction of the wind resource for offshore wind power applications can also be inspected with these techniques. Table 5 summarizes the current state of the research done in wind power and new lines of inquiry.

6. Geothermal energy

We found that there is only one reported application of BNs in the field of geothermal energy, a fault diagnosis in a ground- source heat pump system.

6.1. Resource evaluation

Cai et al. [78] proposed a multi-source information fusion based fault diagnosis methodology for a ground-source heat pump sys- tem. The model is established by combining two Bayesian

Lacks

Wind assessment in offshore facilities

Load prediction Fault diagnosis Management Maintenance

Optimization of wind turbine design and wind farms for small and large scale Management of the fluctuations and failures in the electric network

s

M. Borunda et al. / Renewable and Sustainable Energy Reviews 62 (2016) 32–4538

networks. The BNs consider sensor data and observed information from people and they consist of two layers: the fault layer and the fault symptom layer. The model using only one sensor data is accurate for single fault but not for multiple-simultaneous faults. However, by including more observed information, the fault diagnostic accuracy can increase for a single fault and become better for multiple-simultaneous faults.

6.2. Further steps

One important contribution that could be done to this field is the use of BNs for the prediction of geothermal wells and the assessment for risk management in the construction of geother- mal power plants. Likewise, in geothermal geochemistry, BNs can help through the prediction of geothermal wells to support geo- thermal exploration and improve production efficiency. BNs can be integrated in the geochemical research to understand the nature of the fluid processes to infer reservoir temperature, identify fluid sources and to solve potential issues which may impact in the future utilization of the geothermal resource. Table 6 summarizes the current state of the research done in geothermal power and new lines of inquiry.

7. Hydroelectric energy

Hydroelectric energy is another widely used field for applica- tions of BNs. In this case the rainfall is considered as resource and it is an important issue in the operation of the vessel of a hydro- electric power plant, either with abundance or absence of water in the reservoir. In particular, some work concerning the precipita- tion forecast considering the rainfall-runoff model has been done. On the other hand, BNs have also been applied to the operation and decision support of reservoirs and hydropower equipment. Finally, BNs have also been used for fault diagnosis of hydroelectric power plants as it is shown below.

7.1. Resource evaluation

Cofiño et al. [79] applied Bayesian Networks to weather fore- casting and downscaling. In particular, they built a model for precipitation forecast and validated it with data from 90 days. They obtained the rainfall forecasts for 100 stations in Spain.

Table 6 Applications of BNs in geothermal power, research done and things to do.

Geothermal energy

Highlights Lacks

Resource evaluation

Prediction of geothermal wells Geothermal exploration

Operation Fault diagnosis in a ground-source heat pump system

Inference of reservoir temperature Load prediction Management Maintenance Planning of dispatch Monitoring and control of power fluctuations

Applications Risk management in the con- struction of geothermal power plants Identification of fluid sources Improve production efficiency

Garrote et al. [80] used a stochastic rainfall generator and a deterministic rainfall-runoff model together with a data set from Monte Carlo simulation to forecast floods. They successfully tested the method in Spain.

Petry et al. [81] proposed a stochastic rainfall model to generate daily values of rainfall at multiple locations. They coupled a sto- chastic rainfall model with a deterministic rainfall runoff model. A time series of 1000 years was generated as input for a rainfall runoff model of Germany. They studied the probability of disposal efficiency of the local reservoir system and the risk flood with Bayesian Networks.

Krekeler et al. [82] presented a technique based on Bayesian Networks and belief propagation to estimate and forecast stream flow based on flow at surrounding locations, rainfall, and groundwater levels. They considered stations as part of an array of nodes that communicate evidence of flow measurements to their neighbors. The probability density functions are generated based on the relationship between estimates of flow from the Watershed Assessment Model calibrated with data from the Santa Fe River Watershed from 1990 through 2008. They also evaluated the method's performance with flow measurements along the river.

Wang et al. [83] used Bayesian networks and Markov random analysis to estimate surface precipitation. They estimated the precipitation in seven stations in Qinghai Lake and compared the results with other surface precipitation methods. It turns out that their method accurately provides the relationship of precipitation among different points in the area and the simulations are more accurate than the ones provided by other methods.

Hellman et al. [84] used an ensemble of continuous Bayesian networks for rainfall prediction. They trained the individual Bayesian Network with a subset of data and a subset of attributes to identify important variables and relationships between them. They assembled the networks to represent nonlinear relationships. They successfully compared their method with rainfall prediction across the United States.

Botsis et al. [85] presented a study using a Bayesian Network to simulate the relationship between rainfall and runoff in a water- shed in North California. They successfully tested the model with daily rainfall and streamflow data series from drainage basin.

Madadgar and Moradkhani [86] developed a statistical drought conditions forecast model with BNs and historical data. They used the runoff data from January to June in Funnison River Basin to predict the runoff across the basin in July–December. The model also generates maps to show the runoff variation over the basin in the future which was successfully evaluated. They reported that their model estimates the 5–95% uncertainty bound of the accu- mulated runoff.

7.2. Operation

Garrote et al. [87] modeled the behavior of hydrologic basins during floods with Bayesian networks, in order to build decision models for the prediction and management of river floods in real time. Their approach was to represent hydrological processes with BNs. The BNs together with results from deterministic hydrologic simulation models are an alternative to rainfall-runoff models to estimate the flood risk.

Su et al. [88] presented a general situation assessment model based on Bayesian networks for hydropower equipment with a veracity of 95.2%. The model includes characterization, under- standing and assessment levels. The Bayesian networks consist of situation and event nodes according to their functions and the evidence of the event nodes comes from information acquired by sensors.

Mediero et al. [89] presented a decision maker model to sup- port reservoir operation during flash floods which was successfully

Table 7 Applications of BNs in hydropower, research done and things to do.

Hydroelectric energy

Highlights Lacks

Resource evaluation Weather forecasting Power generation predictionPrecipitation/rainfall

forecasting Flood forecasting Disposal efficiency Risk flood Study of the rainfall and runoff in a watershed Drought forecasting

Operation Behavior of hydrologic basins

Load forecasting

Assessment for hydropower equipment

Minimization of the start up/shut down costs

Decision support for reser- voir operation

Scheduling, planning of the facility Optimization of the dispatch

Applications Fault diagnosis Design of technology Integration with other energy sources

M. Borunda et al. / Renewable and Sustainable Energy Reviews 62 (2016) 32–45 39

validated in Spain. Their model uses BNs together with results of a rainfall-runoff model coupled to a reservoir operation model. The BN predicts the probabilistic outflow discharge and water level. The operation strategy is decided taking into account the prob- ability of maximum discharge of the reservoir and the risk of damage of the dam. They used two data sets of 4000 inflows hydrographs generated from a runoff and a reservoir management model, the first one teaches the BN and the other validates it.

Garrote et al. [90] presented a mixed approach based on the combination of deterministic physically based models and prob- abilistic data-driven models for flood forecasting. The real-time decision support is made through a Bayesian Network built from the results of a deterministic rainfall-runoff model. A Monte Carlo simulation was used to calibrate and validate the model. The model provides probabilistic discharge forecasts in real time using an uncertain quantitative precipitation forecast.

7.3. Applications

Zhang et al. [91] proposed a fault diagnosis model based on Bayesian networks. They considered multiple-fault, fuzzy fault symptoms and the mutual dependency of different operation management for a hydropower plant. They verified the effective- ness and accuracy of the model with the fault diagnosis for a speed governor in Fengman Hydropower Station.

7.4. Further steps

Even though BNs have been widely used in the hydropower field, there are still some remaining issues that could be explored with this tool. For instance, BNs could be used for optimization of the dispatch of hydroelectric generating units. BNs can also be applied for the minimization of the start up/shut down costs of the generating units. Finally BNs could be used to construct, schedule and plan different operation scenarios and look for the optimum one to increase generation and improve the environmental per- formance of hydropower facilities. BNs can be also employed to analyze data to conduct technology development and support future full-scale projects. Table 7 summarizes the current state of the research done in hydropower and new lines of inquiry.

8. Biomass

We found only two reported works in applications of BNs in Biomass. One of them is the fault detection for biomass boilers. The second one is used to classify the biomass made of weed-crop as it is shown below.

8.1. Resource evaluation

Bressan et al. [92] described the modeling of a biomass based weed-crop competitiveness classification process using Bayesian Networks. They used empirical data with collected in a corn-crop: the total density of weeds and the corresponding proportions of weeds biomass. They presented a set of 27 rules extracted from the Bayesian network classifier which categorizes the biomass of weeds.

8.2. Applications

Widarsson et al. [93] developed a method for decision support and fault detection, with Bayesian networks, in biomass-fuelled boilers. The BN gives support on preventive actions to reduce abnormal fouling from flue gases in superheaters. Finally they

validated the method and showed that it works well even when uncertainty exists.

8.3. Further steps

Besides the previous applications, biomass resources are pro- duced in cycles and therefore BNs can be applied to predict the inventory management of the production of biomass. This is useful to help the planning of different biomass resources and implement them. Likewise BNs can be used for decision making and control, and maintenance of bio-digestors since their operation depends on many meteorological variables such as temperature, atmo- spheric pressure and relative humidity. By predicting these meteorological variables, one can optimize the operation of bio- digestors. BNs can be used to deal with the data management required in biochemical processes to reduce costs and production times. Table 8 summarizes the current state of the research done in biomass and new lines of inquiry.

9. Energy storage

In energy storage we only found two applications reported, dealing with electrical and wind energy storage as described in the following two paragraphs.

9.1. Operation

He et al. [94] constructed an effective and efficient method for the estimation of the State of Health of Lithium-ion batteries based on Dynamic Bayesian Networks. They used the state of charge as hidden states and the terminal voltages as observations in the DBN. The training data was collected through Li-ion battery aging experiments.

9.2. Applications

Gibson and Patterson [95] developed a multi-agent system for the project “Intelligent Software Agents for Distribution, Man- agement and the Integration of Storage and Renewables”. The goal

Table 8 Applications of BNs to biomass, research done and things to do.

Biomass

Highlights Lacks

Resource evaluation

Classification of biomass based weed-crop

Prediction of biomass resources Planning of seasonal crops

Operation Decision making and control Management Maintenance Optimization

Applications Decision support and fault diagnosis in biomass-fuel- led boilers

Data management in bio- chemical processes to opti- mize production

Table 9 Applications of BNs in energy storage, research done and things to do.

Energy Storage

Highlights Lacks

Operation Evaluation of state of health of Li ion battery

Prediction of life-time batteries

Fault detection and diagnosis in large storage disposals Maintenance in large storage disposals

Applications Integration of wind energy generators with a storage system

Optimization of the cost-benefit for energy storage in large-scale systems Sizing of storage technologies Risk management Integration of energy storage devices into primary energy sources

M. Borunda et al. / Renewable and Sustainable Energy Reviews 62 (2016) 32–4540

of the project was to prove that wind power can be enhanced with an agent technology which integrates storage and improves management. The agent operates autonomously to gather utility SCADA data. Additionally, it forecasts local wind generation, recommends actions for capacitor bank operation and operates a storage system. Moreover, the agent detects abnormal conditions such as faults or missing sensor data using a Bayesian Belief Network.

9.3. Further steps

Even though nowadays there are a reduced number of appli- cations of BNs in energy storage, we consider that this is a new- born field with a high potential to be exploited to support the development of renewable energies. For instance, BNs could be used for optimization of the cost-benefit of the energy storage in large-scale systems. Bayesian Networks can also contribute to the development of new-generation batteries by predicting their life- time. BNs can be useful in sizing the available energy storage based on the weight, volume energy-density and the load requirements of each storage technology. Likewise, BNs could play a crucial role in the development of new storage technologies by analyzing the associated risk. Additionally, in the already installed facilities, BNs could be useful for fault detection and diagnosis and maintenance in large-scale storage availability, either electrical or thermal. Finally, BNs could be a valuable tool for the integration of energy storage devices into primary energy sources. Table 9 summarizes the current state of the research done in energy storage and new lines of inquiry.

10. Smart grids

As it is shown below BNs have been applied to deal with many problems of smart grids. Forecasting the power demand and generating capacity, management, control, operation and optimi- zation of the network, data analysis, risk evaluation, and power quality have been addressed with Bayesian Networks as it is described as follows.

10.1. Operation

Lehtila et al. [96] proposed a model to forecast the demand and required generating capacity of electricity using a Bayesian net- work. The BN was based on an influence diagram for the demand formation and supply decisions. They assumed that the decision process remains stable even in an unstable market. They suc- cessfully validated the model by reconstructing the demand and

supply profiles in Finland from 1971 to 1987. Their system can also be used for past and future extrapolations.

Rocha et al. [97] presented a decision support system to define the future power consumption of a region. The system forecasts the power demand, using a linear regression model, and it learns the influence patterns of the socio-economic and climatic factors on the power consumption with BNs.

Brevani et al. [98] built an intelligent agent-based control sys- tem using Bayesian networks for automatic generation control in an electric power system. The aim was to provide a control system to deal with the power fluctuation caused by a high penetration of wind farms, which negatively contributes to the power imbalance and frequency deviation. The system was tested on the 10- machine New England test power system and an experimental real time implementation was performed on the West Japan power system.

10.2. Applications

Bashar et al. [99] proposed a Decision Management System (DMS) based on Bayesian Belief Networks to provide smart deci- sions to the network management systems in order to reduce energy consumption, optimize the network performance and assure the quality of provided service. They found that it is pos- sible to achieve a 25% power saving per port with 3% increase in average delay.

Munteanu and Nemes [100] presented a method for quality power supply and risk evaluation in power networks, including renewable energy sources like wind generators and solar panels together with the main power network components. First, they constructed the BN structure for power interruption risk analysis. Then, they studied the correlation factor between solar and wind sources with data mining.

Munteanu et al. [101] used BNs to study the nodal power quality and interruption risk evaluation in power networks. They considered solar and wind sources and studied their correlation factors with data mining. They used BN to asses the nodal quality of the network including wind turbines and solar panels.

Tannahill and Jamshidi [102] constructed a bridge between a System of Systems (SoS) and Data Analytics. The SoS integrates independently operating, non-homogeneous systems. Data Ana- lytics reduces the size of Big Data and extracts information, builds knowledge from the data and develops a non-parametric model using different tools, among other, Bayesian networks. To validate their methodology they used data analytics to generate a model to

Table 10 Applications of BNs in Smart Grids, research done and things to do.

Smart grids

Highlights Lacks

Operation Power demand forecasting Control of intermittency sources

Forecasting of the generating capacity

Maintenance

Control to deal with power fluctuations in the operation Management

Applications Optimization of network performance

Grid optimization design

Quality assurance power Optimization of power dispatch Risk evaluation Optimization of the cost-benefit

Enhance the reliability, security and resiliency of the grid Sizing of hybrid systems Multi-objective optimization of hybrid systems Optimal introduction of storage devices

M. Borunda et al. / Renewable and Sustainable Energy Reviews 62 (2016) 32–45 41

forecast photovoltaic power production and to optimize a micro grid SoS.

10.3. Further steps

Even though the applications of BNs in smart grids have already swept many important issues in the field we can still point out opportunities. For instance, the design of the grid is an ela- borated issue whose degree of complexity increases as a function of the number and kind of the energy sources, and BNs can con- tribute to solve this task. Likewise, intermittency and non- controllability of renewable energies are inherent characteristics of renewable energy-based electricity generation systems and the inclusion of storage devices, to overcome them, becomes a com- plex task in which BNs can play a crucial role. Additionally, BNs can help one to optimize the energy source supply based on the pre- diction of the peak loads, the resource availability and the different costs associated to the production, investment, operation and maintenance of each energy source. Moreover, BNs could be used to optimize the cost-benefit for the distribution of energy in large- scale systems for the electrical market, either for local distribution or for integration in the electric grid. Furthermore, BNs can greatly contribute to enhance the reliability, security and resiliency of the grid taking into account critical infrastructure protection and highly constrained areas. An interesting recent topic in which BNs can contribute is the development of hybrid systems, where the sizing of the system is a complex task due to the large variety of available sources. Likewise BNs can be applied to multi-objective optimization of the hybrid systems. Table 10 summarizes the current state of the research done in Smart Grids and new lines of inquiry.

11. Energy assessment

11.1. Energy market

The energy market is a very interesting and useful topic to encourage and diversify the renewable energy resources for elec- tricity production. A current tool to approach the issues of this field is based on BNs. BNs have contributed to the field of renewable energy market mainly by forecasting electric loads, energy prices, acquisition/sale of energy, and searching for man- agement methods for energy policies and energy investment. In the next paragraphs we describe the works done in this area.

Teixeira et al. [103] constructed the fuzzy hidden Markov pre- dictor, which is a hybrid system made of fuzzy logic and Dynamic Bayesian Networks. The system forecasts monthly electric load. It was successfully compared with other forecast systems.

Sansom [104] investigated the electricity pool price trends for understanding the operation of the Australian national electricity market. He considered a variety of techniques to evaluate a seven day forecast of the prices. He found that all complex machine learning methods provide inferior accuracy forecasts compared to a weekly average method but the last one is computationally less expensive and more transparent to the user than any of the machine learning techniques. The Support Vector Machine and the BNs provided acceptable price forecasts for the region but the former performed better. However, the BNs provided a price forecast with confidence intervals for each half-hour. He also performed an investigation of international electricity markets finding that each market has different market structures, regula- tions, network topologies and ownership regimes. He concluded that his price forecasting techniques and results cannot be uni- versally applied without a detailed consideration of local conditions.

Santana et al. [105] made a decision support system, PREDICT, for the acquisition/sale of energy. The system provides load fore- cast for mid- and long term using artificial neural networks. It also employs BNs to study the correlations and analysis of the load with other variables like consumption, and climatic and social- economic conditions. They implemented the system in Brazil to help in the decision making process for acquisition/sale of energy at a future demand.

Cinar and Kayakutlu [106] proposed an scenario for energy policies using BNs models. They considered Turkey and its renewable energy resources, including domestic and imported ones, under different scenarios looking for sustainability. They took into account different schemes based on the economy sta- bility in Turkey. In a stable scenario they concluded that renewable investments overcome nuclear investment. On the other hand, they found that in an unstable scenario nuclear investment is better.

Shrivastava et al. [107] presented a load forecast methodology using Bayesian Belief Networks useful for power system opera- tional planning. They predicted maximum load growth per year relevant in the purchase and generation of energy, load switching, contract evaluation and infrastructure development.

Daim et al. [108] used Bayesian networks to construct scenarios of energy investment for Oregon. They constructed a framework for planning investment in energy considering initial costs, oper- ating costs, job intensity, perceived acceptability, ecological impact, environmental factors and efficiency. It was related to nuclear, hydro, solar, wind, geothermal and bioenergy in Oregon. The study is directly applied to energy policy-makers.

11.1.1. Further steps Even though many issues have been already addressed in this

field there are still areas of opportunity. Renewable energy mar- kets strongly depend on the fossil fuels price. Besides the forecast applications described above, Bayesian networks can be used to forecast fossil fuels prices. Likewise, BNs can contribute to find the dependence of the prices of renewable energies on the prices of hydrocarbons. Furthermore BNs can be useful for an optimization of the technologies involved in the renewable energy market to achieve a better use of the resources with an associated cost reduction. Additionally BNs can assist the planning of an appro- priate regulatory framework attracting financing. Moreover BNs can predict the low carbon prices to infer the best moment to

M. Borunda et al. / Renewable and Sustainable Energy Reviews 62 (2016) 32–4542

trigger investment in renewable energy technologies. BNs could forecast as well the energy market conditions to predict a short term economic dispatch subject to transmission and operational constraints. Table 11 summarizes the current state of the research done in the energy market and new lines of inquiry.

11.2. Energy efficiency

The goal of energy efficiency is to reduce the amount of energy required in a system. It reduces energy costs and carbon dioxide emissions, and slows down the rate at which energy resources are depleted. Energy efficiency, together with renewable energy, is the basis of a sustainable energy policy and its implementation is obviously needed in sustainable energy systems. Applications of BNs in this field have covered the prediction of energy consump- tion and the management of energy systems mainly in house- holds, greenhouses and smart buildings as it is shown in the fol- lowing paragraphs.

Hawarah et al. [109] showed a method to predict and diagnose the needed energy for a user service in a home automation system. The method is based on Bayesian networks and its goal is to avoid problems related to peak consumption.

Liu et al. [110] presented a DBN-based method to estimate the in-door occupancy in intelligent buildings using vision sensors to improve occupancy comfort and energy efficiency. The detection data from the sensor is the evidence nodes of the network and the true occupancy at time t is estimated by using the evidence prior to time t.

Shipworth [111] constructed and evaluated an occupant representation model, with BNs, to use in building energy models for energy efficiency.

Smith et al. [112] used BNs and neural networks for sensor data validation in buildings. The energy efficiency data of buildings is essential to optimize their energy use and efficiency. They found that temperature, humidity, pressure and flow can be predicted with a root-mean-square (RSME) of less than 10% with neural networks but, on the contrary, energy prediction shows 100% or even 200% RSME. Bayesian networks have RSME of less than 5% but take longer to train.

Espinoza-Huerta et al. [113] used BNs to successfully validate a Computational Fluid Dynamic model that determines gradients related to climate conditions inside greenhouses. BNs show the relationships between CO2 concentration and humidity with respect to temperature and quantify the relationships by infer- ences in the dependent probability distributions.

Carbonari et al. [114] designed an intelligent building energy management system (BEMS) based on a model for predictive control. The system predicts future building status to determine the optimal energy control policies to guarantee comfort at

Table 11 Applications of BNs in the energy market, research done and things to do.

Energy market

Highlights Lacks

Operation Electric load forecasting Price forecasting for hydrocarbons based on Renewable Energies

Forecasting of the electricity prices

Optimization of the renewable energy market

Analysis of sustainability for renewable markets

Prediction of energy market condi- tions for an optimal economic dispatch

Forecasting for energy pur- chasing and generation Planning of renewable energy investment

minimum operational cost. The system is used in Passeig de Gracia metro station in Barcelona to optimize the operation of the mechanical air supply ventilation and the DBNs forecast the expected temperature in the station.

Hernández et al. [115] constructed a model of energy demand in High-Tech greenhouse based on BNs. They considered inside and outside measured variables like temperature, humidity, radiation, CO2 concentration, rain and energy demand. They stu- died the distribution of energy demand in the greenhouse and characterized it by means of classifiers based on Bayesian net- works. The aim is to use the models in an energetic control system for the optimization of energy demand.

Hammer et al. [116] proposed a tool called User Trust Model, for automatic decision making in controlling devices to achieve energy saving in smart energy systems and smart environments. The model is based on Bayesian networks. Its initialization is done with empirical data and its integration is done in an office setting. They included a study showing the user's acceptance.

Morris et al. [117] used a BN complex system to implement an electricity reduction program in an Australian community. They considered data from 22 residential households' interviews to incorporate into the model the factors that the residential parti- cipants found important. The model illustrates the number of factors that influence the peak demand reduction and the complex interactions between them. The model successfully explains the peak demand reduction in the Australian community.

11.2.1. Further steps Even though some work has been done in this important field

there are plenty of opportunities for BNs to be explored here. One of the current goals of sustainable technology is addressed to develop smart communities and spread them. In order to achieve this, it is necessary to predict energy consumption and energy availability in each component of the community, either energy generators or energy loads such as households, buildings, clinics, hospitals, shopping malls, and to calculate the best overall energy balance in order to optimize the efficient use of the energy. Table 12 summarizes the current state of the research done in energy efficiency and new lines of inquiry.

12. Conclusions

In the reported literature we found 62 works dealing with the applications of Bayesian Networks in Renewable Energy from which 66% are published in journals and the rest are published in conference proceedings, reports, seminars and PhD thesis.

As Fig. 2 shows, most of the applications are devoted to wind and hydroelectric energy and the less applied and studied areas are geothermal, solar thermal and photovoltaic energies as well as biomass and energy storage. Forecast of wind, failure diagnosis in wind turbines and wind farms, and maintenance and risk man- agement in wind farms, onshore and offshore, have been studied in the wind energy field. Precipitation forecast with the inclusion of rainfall-runoff model, prediction of reservoirs level, operation of hydrologic basins, decision maker models to support reservoir operation and fault diagnosis in hydropower plants have been already addressed with BNs.

Dynamic Bayesian Networks, which address dynamic systems in a natural way, have been used in wind energy, energy storage, energy market and energy assessment. DBNs have been applied in wind forecast, and also operation, maintenance, and fault diag- nosis of onshore and offshore wind farms. In the energy market field DBNs have been employed to forecast energy demand. Finally DBNs have been very useful for energy efficiency purposes applied to households, intelligent buildings and greenhouses.

Table 12 Applications of BNs in energy efficiency, research done and things to do.

Energy efficiency

Highlights Lacks

Operation Prediction of the required energy in a home automation system Planning for the development of smart communities Estimation of in-door occupancy in intelligent building to improve comfort and efficiency Prediction of energy consumption Prediction of sensor data in buildings Prediction of energy availability Prediction of climate conditions inside greenhouses Optimization of the energy balance Energy management for smart buildings (successfully implemented in a metro station) Optimization of energy demand for greenhouses Energy saving in smart energy systems and environments Electricity reduction program

Fig. 2. Applications of Bayesian networks in renewable energy.

Fig. 3. Applications of BNs by area: (a), (b) and (c) show the utilization for resource evaluation, operation and application, respectively.

M. Borunda et al. / Renewable and Sustainable Energy Reviews 62 (2016) 32–45 43

Fig. 3 encompasses the use of BNs in Renewable Energy by area, in particular, Fig. 3(a), (b) and (c) refer to resource evaluation, operation, and applications, respectively.

The applications presented in this work show the usefulness of Bayesian Networks and Dynamic Bayesian Networks in Renewable

Energy. On the other hand, the open issues addressed in further steps, and summarized in the tables, at the end of each section, show the sea of opportunities to be yet explored and exploited by BNs and DBNs in the field, such as:

� Resource forecasting. � Power generation/demand forecasting. � Fault diagnosis. � Control. � Management. � Risk assessment/management. � Maintenance. � Decision support. � Data management. � Design/sizing/optimization. � Optimal operation. � Planning. � Integration among energy sources.

It is important to point out that Bayesian Networks can be easily implemented as a simple table look-up and are intrinsically fast. Additionally BNs can simply encode human knowledge and expertise, historical data or both, helping users to update models and increase the confidence in the correctness of the model. Likewise BNs support inference in any direction providing responses to any kind of query given some evidence and incor- porate, in a natural way, dynamic systems. These reasons strongly encourage the further research in the area.

Acknowledgments

Mónica Borunda wish to thank Consejo Nacional de Ciencia y Tecnología, CONACYT, support for her Catedra Research Position with ID 71557, and to Instituto de Investigaciones Eléctricas, IIE, for its hospitality. This work has been partially supported by PAPIIT- UNAM under the project IT100514. We thank Maximiliano Valdez González for his technical support in the network management.

References

[1] McCulloch W, Pitts W. A logical calculus of ideas immanent in nervous activity. Bull Math Biophys 1943;5(4):115–33.

[2] Haykin S. Neural networks: a comprehensive foundation. New York: Mac- millan; 1994.

[3] Lawrence J. Introduction to neural networks. Nevada City, CA: California Scientific Software Press; 1994.

[4] Holland JH. Adaptation in natural and artificial systems. Ann Arbor: Uni- versity of Michigan Press; 1975.

[5] Goldberg DE. Genetic algorithms in search, optimization and machine learning. Boston, MA: Addison-Wesley Longman Publishing; 1989.

[6] Michalewicz Z. Genetic algorithmsþdata structures¼evolution programs. New York, NY: Springer; 1996.

M. Borunda et al. / Renewable and Sustainable Energy Reviews 62 (2016) 32–4544

[7] Zadeh LA. Outline of a new approach to the analysis of complex systems and decision processes. IIE Trans Syst Man Cybern 1973;3:28–44.

[8] McNeill D, Freiberger P. Fuzzy logic: the revolutionary computer technology that is changing our world. New York, NY: Touchstone Rockefeller Center; 1993.

[9] Kalogirou SA. Artificial intelligence in energy and renewable energy systems. New York, NY: Nova Editor; 2006.

[10] Mellit A, Kalogirou SA. Artificial intelligence techniques for photovoltaic applications: a review. Prog Energy Combust Sci 2008;34:574–632.

[11] Mellit A, Kalogirou SA, Hontoria L, Shaari S. Artificial intelligence techniques for sizing photovoltaic systems: a review. Renew Sustain Energy Rev 2009;13 (2):406–19.

[12] Michaelides SC, Tymvios FS, Kalogirou SA. Artificial neural networks for meteorological variables pertained to energy and renewable energy appli- cations. In: Artificial intelligence in energy and renewable energy systems. New York, NY: Nova Editor; 2006. p. 47–82.

[13] Tapia A, Tapia G, Flores P. Application of control algorithms for wind speed. In: Artificial intelligence in energy and renewable energy systems. New York, NY: Nova Editor; 2006. p. 201–34.

[14] Hontoria L, Aguilera J, Almonacid F, Nofuentes G, Zufiria P. Artificial neural networks applied in pv systems and solar radiation. In: Artificial intelligence in energy and renewable energy systems. New York, NY: Nova Editor; 2006. p. 163–200.

[15] Lalot S. Artificial neural networks in solar thermal energy systems. In: Arti- ficial intelligence in energy and renewable energy systems. New York, NY: Nova Editor; 2006. p. 131–162.

[16] Kalogirou SA. Artificial neural networks in renewable energy systems - a review. Renew Sustain Energy Rev 2001;5(4):373–401.

[17] Kalogirou SA. Artificial neural networks and genetic algorithms for the optimisation of solar thermal systems. In: Artificial intelligence in energy and renewable energy systems. New York, NY: Nova Editor; 2006. p. 131–162.

[18] Razak JA, Sopian K, Nopiah ZM, Zaharim A, Ali Y. Optimal operational strategy for hybrid renewable energy system using genetic algorithms. In: 12th WSEAS international conference on applied mathematics; 2007.

[19] Suganthi L, Iniyan S, Samuel AA. Applications of fuzzy logic in renewable energy systems - a review. Renew Sustain Energy Rev 2015;48:585–607.

[20] Korb KB, Nicholson AE. Bayesian artificial intelligence. Boca Raton, London, New York: CRC Press; 2011.

[21] Chris JN, James RB. Inference in Bayesian networks. Nat Biotechnol 2006;1:51–3.

[22] Pearl J. Probabilistic reasoning in intelligent systems: networks of plausible inference. San Francisco, CA: Morgan Kaufmann Publishers; 1988.

[23] Charniak E, McDermott D. Introduction to artificial intelligence. Boston, MA: Addison-Wesley; 1989.

[24] Charniak E. Bayesian networks without tears. Artif Intell Mag 1991;4:50–63. [25] Hecherman D. Probabilistic similarity networks. Technical report. STAN-CS-

1316, Standford University; 1990. [26] Spiegelhalter D, Franklin R, Bull K. Assessment criticism and improvement of

imprecise subjective probabilities for a medical expert system. In: Fifth workshop on uncertainty in artificial intelligence; 1989. p. 335–342.

[27] Dean T. Coping with uncertainty in a control system for navigation and exploration. In: Ninth national conference on artificial intelligence; 1990. p. 1010–5.

[28] Charniak E, Goldman R. A semantics for probabilistic quantifier-free first- order languages with particular application to story understanding. In: Ele- venth international joint conference on artificial intelligence; 1989. p. 1074– 9.

[29] Goldman R. A probabilistic approach to language understanding. Technical report. CS-90-34, Brown University; 1990.

[30] Levitt T, Mullin J, Bindord T. Model-based influence diagrams for machine vision. In: Fifth workshop on uncertainty in artificial intelligence; 1989. p. 233–44.

[31] Hansson O, Mayer A. Heuristic search as evidential reasoning. In: Fifth workshop on uncertainty in artificial intelligence, 1989. p. 152–61.

[32] Kuikka S, Varis O. Uncertainties of climatic change impacts in Finnish watersheds: a Bayesian network analysis of expert knowledge. Boreal Environ Res 1989;2:109–28.

[33] Varis O. Belief networks for modelling and assessment of environmental change. Environmetrics 1995;6:439–44.

[34] Varis. O. A belief network approach to optimisation and parameter estima- tion: application to resource and environmental management. Artif Intell Mag 1998;101(1–2):135–63.

[35] Varis O, Kuikka S. A Bayesian approach to expert judgement elicitation with case studies on climate change impacts on surface waters. Clim Change 1997;37:539–63.

[36] Ames D, Neilson B. A Bayesian decision network engine for internet-based stakeholder decision making. In: ASCE world water and environmental resources congress conference; 2001.

[37] Borsuk M, Clemen R, Maguire L, Rechhow K. Stakeholder values and scien- tific modeling in the Neuse river watershed. Group Decis Negot 2001;10:355–73.

[38] Ibargüengoytia Pablo H, Delgadillo Miguel A, García Uriel A, Reyes Alberto. Viscosity virtual sensor to control combustion in fossil fuel power plants. Eng Appl Artif Intell 2013;29:2153–63.

[39] Ibargüengoytia Pablo H, Delgadillo Miguel A, García Uriel. Evaluating prob- abilistic models learned from data. In: Batyrshin Ildar, Sidorov Grigori,

editors. Advances in soft computing—MICAI 2011, Lecture notes in artificial intelligence, vol. 7095. Berlin, Heidelberg: Springer-Verlag; 2011. p. 95–106.

[40] Weber P, Medina-Oliva G, Simon C, Iung B. Overview on Bayesian networks applications for dependability, risk analysis and maintenance areas. Eng Appl Artif Intell 2012;25:671–82.

[41] Torres-Toledano JG, Sucar LE. Bayesian networks for reliability analysis of complex systems. In: Lecture notes in computer science, vol. 1484; 1998. p. 195–206.

[42] Cornalba C, Giudici P. Statistical models for operational risk management. Physica A 2004;338:166–72.

[43] Waeyenbergh G, Pintelon L. Maintenance concept development: a case study. Int J Prod Econ 2004;89(3):395–405.

[44] Weber P, Suhner M-C, Iung B. System approach-based Bayesian network to aid maintenance of manufacturing process. In: Proceedings of the sixth IFAC symposium on cost oriented automation, low cost automation; 2001. p. 33– 9.

[45] Ibargüengoytia Pablo H, Reyes Alberto. On-line diagnosis of a power gen- eration process using probabilistic models. In: 16th international conference on intelligent systems application to power systems, ISAP-2011 Hersonissos, Crete Greece. IEEE PES; 2011.

[46] Wang Ba, Wang Yb, Chen Xc. Research on wind turbine generator dynamic reliability test system based on feature recognition. Res J Appl Sci Eng Technol 2013;6(16):3065–71.

[47] Ibargüengoytia Pablo H, Enrique Sucar L, Vadera Sunil. Real time intelligent sensor validation. IEEE Trans Power Syst 2001;16(4):770–5.

[48] Ibargüengoytia PabloH, Vadera Sunil, Enrique Sucar L. A probabilistic model for information and sensor validation. Comput J 2006;49(January (1)):113– 26.

[49] Howard RA, Matheson JE. Influence diagrams. In: Readings on the principles and applications of decision analysis, vol. 2; 1981. p. 721–62.

[50] Ames D. Bayesian decision networks for watershed management [Ph.D. thesis]. Utah State University; 2002.

[51] Ames D, Neilson BT. A Bayesian decision network engine for internet-based stakeholder decision making; 2002.

[52] Murphy KP. Dynamic Bayesian networks: Representation, inference and learning [Ph.D. thesis]. University of California, Berkeley; 2002.

[53] Friedman N, Murphy K, Russell S. Learning the structure of dynamic prob- abilistic networks. In: Proceedings of the fourteenth conference on uncer- tainty in artificial intelligence98; 1998.

[54] Zweig G, Russell S. Compositional modeling with dpns. Technical report UCB/ CSD-97-970. Computer Science Division (EECS), University of California at Berkeley; 1997.

[55] Coleman A, Zalewski J. Intelligent fault detection and diagnostics in solar plants. In: Intelligent data acquisition and advanced computing systems; 2011. p. 948–53.

[56] Liu A, Liu Y, Zhang D, Cai B, Zheng C. Fault diagnosis for a solar assisted heat pump system under incomplete data and expert knowledge. Energy 2015;87 (C):41–8.

[57] Dong LA, Zhou WA, Zhang PB, Liu GC, Li WA. Short-term photovoltaic output forecast based on dynamic Bayesian network theory. Zhongguo Dianji Gongcheng Xuebao/Proc Chin Soc Electr Eng 2013;33(SUPPL):38–45.

[58] Oviedo D, Romero-Ternero MC, Hernández MD, Silvanes F, Carrasco A, Escudero JI. Multiple intelligences in a multiagent system applied to tele- control. Expert Syst Appl 2014;41(15):6688–700.

[59] Cano R, Sordo C, Gutiérrez JM. Applications of Bayesian networks in meteorology. In: Advances in Bayesian networks; 2004. p. 309–27.

[60] De la Torre-Gea G, Soto-Zarazúa GM, Guevara-González RG, Rico-García E. Bayesian networks for defining relationships among climate factors. Int J Phys Sci 2011;6(18):4412–8.

[61] Carta JA, Velázquez S, Matías JM. Use of Bayesian networks classifiers for long-term mean wind turbine energy output estimation at a potential wind energy conversion site. Energy Convers Manag 2011;52(2):1137–49.

[62] Ibargüengoytia Pablo H, Reyes Alberto, Romero Ines, Pech David, García Uriel, Enrique Sucar L, et al. Wind power forecasting using dynamic Bayesian models. In: Gelbokh A, editor. Advances in soft computing, Part II—MICAI 2014, Lecture notes in artificial intelligence, vol. 8857. Berlin, Heidelberg: Springer-Verlag; 2014.

[63] Zitrou A, Bedford T, Walls L. A model for supporting decisions regarding the operation and maintenance of offshore wind turbines. In: ESREL, Rhodes; 2010. p. 1401–8.

[64] Nielsen JJ. Bayesian networks as a decision tool for operation and main- tenance of offshore wind turbines. In: ASRANet: integrating structural ana- lysis, risk & reliability; 2010.

[65] Nielsen JJ, Sorensen JD. Risk based maintenance of offshore wind turbines using Bayesian networks. In: 6th EAWE PhD seminar on wind energy in Europe; 2010.

[66] Dinwoodie I, McMillan D, Revie M, Lazakis I, Dalgic Y. Development of a combined operational and strategic decision support model for offshore wind. Energy Proc 2013;35:157–66.

[67] Kougioumtzoglou MA, Lazakis I. Developing a risk analysis and decision making strategy for an offshore wind farm. In: 5th international symposium on ship operations, management and economics (SOME); 2015.

[68] Chen J, Hao G. Research on the fault diagnosis of wind turbine gearbox based on Bayesian networks. In: Proceedings of the sixth international conference on intelligent systems and knowledge engineering, vol. 3; 2011. p. 271–23.

M. Borunda et al. / Renewable and Sustainable Energy Reviews 62 (2016) 32–45 45

[69] Plumley CE, Wilson GK, Kenyon AD, Quail F, Zitrou A. Diagnostics and prognostics utilising dynamic Bayesian networks applied to a wind turbine gearbox. In: International conference on condition monitoring and machine failure prevention technologies; 2012.

[70] Tavner PJ, Feng Y, Song WW, Qiu Y, Chen B. Bayesian networks for wind turbine fault diagnosis. In: EWEA; 2012.

[71] Hameed Z, Hong YS, Cho YM, Ahn SH, Song CK. Condition monitoring and fault detection of wind turbines and related algorithms: a review. Renew Sustain Energy Rev 2009;13(1):1–39.

[72] Dai L, Ehlers S, Rausand M, Utne B. I. Risk of collision between service vessels and offshore wind turbines. Reliab Eng Syst Saf 2013;109:18–31.

[73] Pan HN, Qin M, Zhang J, Chang C, Lei P. Bayesian networks in electric reliability assessment of doubly-fed wind turbine generator. Appl Mech Mater 2014;494:1791–4.

[74] Shuang Y. Research of wind power plant risk management based on Bayesian network. Adv Mater Res 2014;3:587–90.

[75] Pattison D, Segovia García M, Xie W, Quail F, Review M, Whitfield RI, Irvine I. Intelligent integrated maintenance for wind power generation. Wind Energy 2016;19(3):547–62.

[76] Li YFA, Valla SA, Zio EAB. Reliability assessment of generic geared wind turbines by gtst-mld model and monte carlo simulation. Renew Energy 2015;83:222–33.

[77] Ashrafi M, Davoudpour H, Khodakarami V. Risk assessment of wind turbines: transition from pure mechanistic paradigm to modern complexity paradigm. Renew Sustain Energy Rev 2015;51:347–55.

[78] Cai B, Liu Y, Fan Q, Zhang Y, Liu Z, Yu S, et al. Multi-source information fusion based fault diagnosis of ground-source heat pump using Bayesian network. Appl Energy 2014;114:1–9.

[79] Cofiño AS, Cano R, Sordo C, Gutierrez JM. Bayesian networks for probabilistic weather prediction. In: Proceeding of the 15th European conference on artificial intelligence; 2002. p. 695–700.

[80] Garrote L, Molina M, Mediero L. Probabilistic forecasts using Bayesian net- works calibrated with deterministic rainfall-runoff models. In: Extreme hydrological events: new concepts for security, vol. 2; 2007. p. 173–83.

[81] Petry U, Hundecha Y, Pahlow M, Schumann A. Generation of severe flood scenarios by stochastic rainfall in combination with a rainfall runoff model. In: 4th international symposium on flood defence; 2008.

[82] Krekeler CR, Nagarajan K, Graham WD, Slatton KC. Stream flow estimation via belief propagation for sparsely instrumented watersheds. American Geophysical Union, Fall Meeting 2009; 2009.

[83] Wang HR, Ye LT, Xu SY, Feng QL, Jiang Y, Liu Q, et al. Bayesian networks precipitation model based on hidden Markov analysis and its application. Sci China Technol Serv 2010;53(2):539–47.

[84] Hellman S, McGovern A, Xue M. Learning ensembles of continuous Bayesian networks: an application to rainfall prediction. In: Conference on intelligent data understanding; 2012. p. 112–7.

[85] Botsis D, Latinopoulos P, Diamantars D. Investigation of the effect of inter- ception and evapotranspiration on the rainfall-runoff relationship using Bayesian networks. In: Proceedings of protection and restoration of the environment XI, Thessaloniki; 2012.

[86] Madadgar S, Moradkhani H. Spatio-temporal drought forecasting with Bayesian networks. J Hydrol 2014;512:134–46.

[87] Garrote L, Molina M, Blasco G. Application of Bayesian networks to real-time flood risk estimation. EGS - AGU - EUG Joint Assembly 9; 2003.

[88] Su Y, Zhao H, Su W-J, Xu Y. Situation assessment/diagnosis model based on Bayesian networks for hydropower equipment. Dongbei Daxue Xuebao/J Northeast Univ 2005;26(8):739–42.

[89] Mediero L, Garrote L, Martin-Carrasco F. A probabilistic model to support reservoir operation decisions during flash floods. Hydrol Sci J 2007;52 (3):523–37.

[90] Garrote L, Molina M, Blasco G. Learning Bayesian networks from determi- nistic rainfall-runoff models and monte carlo simulation. In: Practical hydroinformatics, computational intelligence and technological develop- ments in water applications. Part V; 2008. p. 375–88.

[91] Zhang X-D, Zhao H, Xie Y-M, Yin Z-Y. Bayesian network model for fault diagnosis of hydropower equipment. Dongbei Daxue Xuebao/J Northeast Univ 2006;27(3):276–9.

[92] Bressan GM, Oliveira VA, Hruschka ER, Nicoletti MC. Biomass based weed- crop competitiveness classification using Bayesian networks. In: Intelligent systems design and applications; 2007. p. 121–6.

[93] Windarsson B, Karlsson C, Dahlquist E. Bayesian network for decision sup- port on soot blowing superheaters in a biomass fuelled boiler. In: Probabil- istic methods applied to power systems, 2004; 2004. p. 212–7.

[94] He Z, Gao M, Ma G, Liu Y, Chen S. Online state-of-health estimation of lithium-ion batteries using dynamic Bayesian networks. J Power Sour 2014;267:576–83.

[95] Gibson GL, Patterson J. “Agents for integration of storage and renewables” project results. vol. 2, Atlanta, GA; 2012. p. 1440–4.

[96] Lehtila A, Silvennoinen P, Vira J. A belief network model for forecasting within the electricity sector. Technol Forecast Soc Change 1990;38(2):135– 50.

[97] Rocha CA, Santana AL, Frances CR, Bezerra U, Tupiassu A, Gato V, et al. Decision support in power systems based on load forecasting models and influence analysis of climatic and socio-economic factors. In: Proceedings of SPIE, vol. 6383, Wavelet applications in industrial processing IV.

[98] Bevrani H, Daneshfar F, Hiyama T. A new intelligent agent-based agc design with real-time application. IEEE Trans Syst Man Cybern Part C: Appl Rev 2012;42(6):994–1002.

[99] Bashar A, Parr GP, Il McClean S, Scotney BW, Subramanian M, Chaudhari SK, et al. Employing Bayesian belief networks for energy efficient network management. In: National communications commission conference; 2010. p. 1–5.

[100] Munteanu F, Nemes C. Belief networks utilization for nodal power quality and availability assessment. UPB Sci Bull Ser C: Electr Eng 2012;74(1):215–22.

[101] Munteanu F, Nemes C. Belief networks utilization for nodal power quality and availability assessment. Univ Politeh Bucur Bul Stiintific Ser C 2012;74.

[102] Tannahill BK, Jamshidi MM. System of systems and big data analytics— bridging the gap. Comput Electr Eng 2014;40(1):2–15.

[103] Teixeira MA, Zaverucha G. Fuzzy hidden Markov predictor in electric load forecasting. In: 2004 Proceedings of neural networks; 2004.

[104] Sansom D. Investigation into electricity pool price trends and forecasting for understanding the operation of the Australian national electricity market (nem) [Ph.D. thesis]. University of Queensland; 2006.

[105] Santana AL, Conde GB, Rego LP, Rocha CA, Cardoso LD, Costa JCW, et al. Predict—decision support system for load forecasting and inference: a new undertaking for Brazilian power suppliers. Electr Power Energy Syst 2012;38:33–45.

[106] Cinar D, Kayakutlu G. Scenario analysis using Bayesian networks: a case study in energy sector. Knowl Based Syst 2010;23(3):267–76.

[107] Shrivastava V, Misra RB. Development of Bayesian belief network model for electrical load demand. Int J Syst Assur Eng Manag 2010;1(2):170–7.

[108] Daim T, Kayakutlu G, Suharto Y, Bayram Y. Clean energy investment sce- narios using the Bayesian network. Int J Sustain Energy 2014;33(2):400–15.

[109] Hawarah L, Ploix S, Jacomino M. User behavior prediction in energy con- sumption in housing using Bayesian networks. In: 10th international con- ference artificial intelligence and soft computing, ICAISC 2010 (2); 2010. p. 372–9.

[110] Liu D, Guan X, Du Y, Zhao Q. Measuring indoor occupancy in intelligent buildings using the fusion of vision sensors. Meas Sci Technol 2013;24(7).

[111] Shipworth D. The vernacular architecture of household energy models. Per- spect Sci 2013;21(2):250–66.

[112] Smith MK, Castello CC, New JR. Generation of severe flood scenarios by sto- chastic rainfall in combination with a rainfall runoff model. IEEE Conf Publ 2013;1:305–8.

[113] Espinoza-Huerta TD, Ortíz-Vázquez IC, García-Manzo G, De la Torre-Gea GA. A multivariable computational fluid dynamics validation method based in Bayesian networks applied in a greenhouse. Int J Agric Innov Res 2014;4 (1):1473–2319.

[114] Carbonari A, Vaccarini M, Giretti A. Bayesian networks for supporting model based predictive control of smart buildings. In: Dynamic programming and Bayesian inference, concepts and applications; 2014. p. 3–39.

[115] Hernández C, Sagrado J, Rodríguez F, Moreno JC, Sánchez JA. Modeling of energy demand of a high-tech greenhouse in warm climate based on Baye- sian networks. Math Probl Eng 2015;1:1–11.

[116] Hammer S, Wissner M, Andre E. Trust-based decision-making for smart and adaptive environments. User Model User-Adapt Interact 2015;25(3):267–93.

[117] Morris P, Vine D, Buys L. Application of a Bayesian network complex system model to a successful community electricity demand reduction program. Energy 2015;84:63–74.