Scenario analysis using Bayesian networks: A case study in energy sector

Knowledge-Based Systems 23 (2010) 267–276

Contents lists available at ScienceDirect

Knowledge-Based Systems journal homepage: www.elsevier.com/locate/knosys

Scenario analysis using Bayesian networks: A case study in energy sector Didem Cinar, Gulgun Kayakutlu * Department of Industrial Engineering, Istanbul Technical University, Macka 34367, Istanbul, Turkey

a r t i c l e

i n f o

Article history: Received 28 July 2009 Received in revised form 20 January 2010 Accepted 23 January 2010 Available online 1 February 2010 Keywords: Bayesian networks Causal maps Scenario analysis Energy investments

a b s t r a c t This paper provides a general overview of creating scenarios for energy policies using Bayesian Network (BN) models. BN is a useful tool to analyze the complex structures, which allows observation of the current structure and basic consequences of any strategic change. This research will propose a decision model that will support the researchers in forecasting and scenario analysis fields. The proposed model will be implemented in a case study for Turkey. The choice of the case is based on complexities of a renewable energy resource rich country. Turkey is a heavy energy importer discussing new investments. Domestic resources could be evaluated under different scenarios aiming the sustainability. Achievements of this study will open a new vision for the decision makers in energy sector. Ó 2010 Published by Elsevier B.V.

1. Introduction Complexity of the decision making in energy sector is caused not only by multiple factors and processes [1], but also by variety of stake holders in the decision. Energy sustainability, stability and variety are considered to be vital for the economic development. Since energy is an inevitable input for all industries, the sustainable supply of energy resources becomes a necessary part of the national economical strategies. Availability of energy resources at reasonable cost and utilizing without causing negative social effects are essential strategies [2]. The high importance of energy investments cause preparation of future plans to be based on scenarios by using the alternative variables influential in decision. As Hobbs [3] stated scenario-based decision making crosses many domains and multiple perspectives. It is observed in literature survey that, until very recently, statistical methods were used to create scenarios [4]. Fluctuations in factor values and uncertainties cause fuzzy and stochastic analysis [5–7]. Knowledge systems are recently used in energy investment scenarios [8–10]. Furthermore, Bayes Networks are used mostly in risk analysis and classification [11–13]. The objective of this paper is analyzing the unstable and complex structure of energy sector by running an expert survey to find out the dependencies among effective factors and creating scenarios based on their opinions. This will give an opportunity to create scenarios independent of politics by using Casual Maps (CM) for analyzing the opinion poll and Bayes Network (BN) to create scenarios. The case application will be done in Turkey, a fossil energy

* Corresponding author. Tel.: +9 0532 212 77 34. E-mail address: [email protected] (G. Kayakutlu). 0950-7051/$ - see front matter Ó 2010 Published by Elsevier B.V. doi:10.1016/j.knosys.2010.01.009

importer country with unstable economic structure, where energy policies are deemed to be redesigned. Sustainable development of Turkish energy sector needs a change in the present energy production and consumption patterns. Investment is to be done in diversified energy resources and environmental concern is to be included in energy strategies [14]. The only alternative resource is recommended to be natural gas which has been growing rapidly [15]. Unfortunately this alternative caused the increase in import dependency [14]. Whereas, Turkey has an appropriate geographical location and weather conditions for extensive usage of renewable energy sources including hydropower, biomass, geothermal, solar and wind energy [16]. However, despite the reactions, nuclear energy is considered as a solution by Turkish government [17]. The case study in this research is constructed to respond the question of interactions of different factors effecting the renewable energy and nuclear energy investments. Hence, this study will contribute to knowledge system studies as well as decision makers in energy sector. In Sections 2 and 3 of this paper, Causal Maps (CM) and Bayesian Networks (BN) were explained and the choice of these tools is examined. In Section 4, CM and BN of Turkish energy sector was modelled and investment alternatives were examined under different economic and policy scenarios. Finally, concluding remarks and suggestions for further studies were given in the last section. 2. Causal maps Predictions of future events have an important role in decision making process. In forecasting, the main assumption is that the future will be much like the past. This causes a detailed analysis and clarification of the past in order to make accurate predictions [18].

268

D. Cinar, G. Kayakutlu / Knowledge-Based Systems 23 (2010) 267–276

The increasing uncertainty in the 21st century creates problems to capture the dynamic structure of events. Designing a graph, which represents causal relationships between events, is a way of arranging a circumstance for reasoning under uncertainty [19]. In the last few years, Causal Maps (CM) have been widely used to construct a framework and represent major factors, knowledge and conditions that influence decision making [20]. Causal Map (CM) is a visual representation of thinking about a subject. CMs are formed by nodes and arrows which imply believed causality. CM takes shape through interviews or through the analysis and coding of documents, so they represent the beliefs, values and expertise of decision makers relevant to the issue in hand [21]. CMs have three major parts: causal concept, causal connection and causal value. A causal concept represented by a node can be an attribute, issue, factor or variable. Causal connection is presented by an arrow and shows the direction of the connection. It depicts a cause–effect relation between two concepts. The concept at the tail of an arrow is taken to cause the concept at the head of the arrow. A causal connection can be either positive or negative. Positive causal connection implies the positive correlation between two variables connected by the arrow, when the variable at the tail of the arrow increases the variable at the head of the arrow increases. In the same way, the negative causal connection refers to negative correlation. Causal value is the strength of the causal connection. There are many different techniques used for determining the causal value. The technique used for finding the causal values of a certain CM is specified by the aim of the analysis [22]. CMs can provide us to look at the problem more extensively than other decision tools which consider causal relations, such as regression. CM has been widely used in international relations, administrative science, political science, sociology, policy analysis, organizational behaviour and management science [20,21,23,24].

3. Bayesian networks CMs are more illustrative than other analysis tools as regression and structural equation. Hence, they can supply missing information and details and bring the priorities and key factors into focus [21,23]. However there are some reasons to draw inferences inefficiently with causal maps. First reason is the inadequacy of modelling uncertainty. All variables have the same level of certainty in CMs. Problems with uncertain variables may have incomplete information or vaguely defined variables which are experienced to make inference. Second reason is the static representation of a problem. CMs cannot clarify the influence of changes on decision variables. There is a variety of methods used to make inferences from cognitive maps in literature; matrix algebra and network analytic methods, system dynamics, neural networks and BNs [20] are just some of them. These methods are valuable when the necessary data is available historically and the statistics can be run. Miao et al. [25,26], and Salmeron [27] have used a new approach of fuzzy cognitive map in order to model complicated and heterogeneous problems. They used fuzzy numbers to refer weights and concepts in cognitive map. The drawback of fuzzy cognitive map is being static in modelling the problem. Main advantage of BN is having a dynamic approach that is unavoidable to analyze the complex and unstable systems [21]. Bayesian networks are directed acyclic graph (DAG) which means there are no cycles. If there is a link between A and B (A ? B), we say that B is a child of A and A is a parent of B. In Bayesian networks, a link from node A to node B does not always imply causality. It implies a direct influence of A over B and the probability of B is conditioned on the value of A [2]. DAG represents the construction of causal dependence between nodes and gives the

qualitative reasoning in BN. Conditional Probabilistic Table (CPT) constitutes the quantitative part of BN with conditional probabilities of nodes. BN relies on the chain rule which is about the joint probability distribution of each variable. According to the chain rule, the marginal and conditional probabilities can be computed for each node of the network. Suppose that we have variablesX1, . . . , Xd, the joint probability of Xi is then:

PðX 1 ; . . . ; X d Þ ¼

d Y

PðX i jparentsðX i ÞÞ:

i¼1

When evidence received from external sources about possible states of a variable or a set of variables, the marginal and conditional probabilities of the variables can be computed by marginalizing over the joint. If some evidence is given over some variables, the probability of incident of some events is calculated as following:

PðUjeÞ ¼

PðU; eÞ ; PðeÞ

where U is the universe of variables X1, . . . , Xd [21]. BN has several advantages for making inferences. First, it is an effective method for data with missing values. Second, it enables us to look at the problem in a wide frame by presenting causal relations. Third, it combines the probabilistic and causal semantics so has an advantage to integrate the human knowledge and data. BM is a useful tool for modelling the uncertainty [22]. Recent studies have used CMs to provide qualitative interpretation of various decision problems. Qualitative techniques used to analyze CMs are useful in simple maps with few variables. When the map has a lot of variables and it is complex to analyze, there is a need to use a quantitative method to make inferences from CM. Bayesian Network (BN) is an artificial intelligence method which uses CM to make inferences for decision making [22]. BN helps to understand the problem structure graphical representation as well as breaking down the problem of representing the joint distribution of many variables into groups [28]. BNs can be used for several different purposes [29], including classification [30] clustering, forecasting [31,32], abductive reasoning (finding the diagnostic inference) and decision making. This paper focuses on decision making problem in energy sector. The literature covers various studies that use BN as a decision support tool in different areas. Trucco et al. [33] developed a BN modelling to analyze the risk factors in maritime transportation. Gupta and Kim [34] proposed an integrated model which combines BN and structural equation modelling to decision making in customer management. Aktas et al. [35] used BN to develop a decision support system for healthcare management. Ulengin et al. [36] proposed a decision support model based on BN for transportation policy decisions. Lauria and Duchessi [37] created a BN based decision support system to what-if analysis about information technology implementations. Sßahin et al. [24] used BN to analyze the structure of inflation in Turkey. Fenton and Neil [38] developed a hybrid approach which consists of BN and analytic hierarchical process with application to a safety assessment that was being used by a major transportation organization. Nadkarni and Shenoy [22] described the BN as a tool for making inferences from causal maps and constructed a BN for a product development decision. The results of BNs are visual, illustrative and easy to interpret for decision makers. 4. Case: Turkey 4.1. Energy policies in Turkey Energy generation, stability and efficiency are vital problems for developing countries. Extensive usage of fossil fuel by industries has caused considerable environmental problems. These problems which affect human health and welfare unfavourably can be

D. Cinar, G. Kayakutlu / Knowledge-Based Systems 23 (2010) 267–276

overcome by using alternative energy sources. In order to mitigate sustainability problems caused by fossil fuels, it is necessary to limit fossil energy consumption and encourage wider utilisation of green energy sources. Therefore, fossil energy production and consumption should be planned. For importing countries, like Turkey, this can be accomplished by determining the domestic sources and strategic energy policy [39]. In the energy literature, miscellaneous methods have been applied to decision problems. Chatzimouratidis and Pilavachi [40] used analytic hierarchy process to evaluate ten types of power plants available at present including fossil fuel, nuclear and renewable energy based power plants. Martinsen and Krey [41] improved a fuzzy optimization method to give a better representation of political decision processes in the energy economy and energy policy than a qualitative examination. Mulugetta et al. [42] assessed three different long-term scenarios and used the long-range energy alternative planning (LEAP) system to analyze and evaluate the technical and policy options of Thailand. Krukanont and Tezuka [43] developed a two-stage stochastic programming to identify the optimal allocation of investment under uncertainty and under different energy policies. Koskela et al. [44] used LCA (Life Cycle Assessment) methodology for three different future electricity supply scenarios based on domestic oil shale, imported natural gas, and imported nuclear power for Estonia in 2020 and compared to the situation in 2002. Mezher et al. [45] developed a pre-emptive goal programming model to the energy resource selection problem. Agrawal and Singh [46] used fuzzy goal programming to choice energy resources with respect to economic, environmental and technological factors. They used fuzzy numbers because of the uncertain nature of energy resource allocation problem. In our case, real data are available and probabilistic approach is able to catch the uncertainty of energy environment. On the other hand, deterministic models, such as system dynamics, goal programming, has some disadvantages to modelling the vagueness of energy problems. Despite the fact that Turkey is rich in variety of energy resources, a big portion of energy request is responded by fossil energy imports; furthermore, share of imported resources increases every year. In 1970 the share of the imports was 24.59%, which became 73.19% in 2006 [23]. Dependency in energy is the major problem of the Turkish energy sector. Fig. 1 shows the primary energy importation in Turkey from 1970 to 2006. Because of the vital role of energy in macro economy, energy planning is considered to be necessary to make better decisions, gain competitive advantage and avoid surprises. In this study Turk-

ish energy planning is taken as the application area because the installation of nuclear station in Mersin-Akkuyu and/or in Sinop, is one of the major agenda for government, public and non-governmental organizations. The major cause of establishing a nuclear plant is explained to be the reduction of importation. Nuclear energy is considered as a cure to generate energy more than the other alternative energies. However Chernobyl disaster is a frightening reason for public to be against nuclear power stations. The climate change reasoned by fossil fuels is concerned and countries are trying to find and evaluate alternative energy sources. Although Turkey has miscellaneous potential natural sources, renewable energy production is 10% in total energy consumption [23]. This study, using a BN model, explores the changes of imports and greenhouse emission of different investment strategies. Two strategies are analyzed: the first one is the nuclear energy investment supported by the current government, and the second one is the renewable energy as the scientists propose. Investment in both of the alternatives is not considered in this study. This study will explore how energy importation and greenhouse emission are affected when decision makers invest on renewable energy instead of nuclear energy. 4.2. Causal cognitive map There are many agents determining energy investments. In order to obtain an exclusive and detailed list of factors a cognitive map is used. Factors that affect energy investments are determined from literature [43,44,47] and validated by 15 energy experts. Energy investment is greatly affected by economic, social, technological and environmental factors which are investigated by a variety of researchers. Public benefits are given the priority in identifying significant influences. Gross domestic product (GDP), primary energy consumption, renewable energy production, fossil fuel production, GDP per capita, population, urbanization and industrialization are basic economic parameters for energy investments, to which greenhouse emission and energy import are included for public benefit. All together 11 driving forces are investigated to effect investments directly and indirectly. After having identified the factors, interactions between factors are scaled by asking the experts. 1, 0 and 1 are used for scaling. ‘‘1” refers to a negative effect meaning the increase in the row factor causes decrease in the column factor; whereas ‘‘1” refers to a positive effect meaning the increase in the row factor causes an increase in the column factor. ‘‘0” means there is no interaction between factors tested.

80000 70000 60000 50000 40000 30000 20000 10000 0 19 70 19 72 19 74 19 76 19 78 19 80 19 82 19 84 19 86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04 20 06

primary energy impor

269

Fig. 1. Primary energy importation in Turkey (tToe).

D. Cinar, G. Kayakutlu / Knowledge-Based Systems 23 (2010) 267–276

270 Table 1 Pairwise comparison matrix.

Renewable energy investment Nuclear energy investment Primary energy consumption Primary energy imports Renewable energy production Fossil fuel production GDP per capita Population Urbanization Industrialization Greenhouse emission

Renewable energy investment

Nuclear energy investment

Primary energy consumption

Primary energy imports

Renewable energy production

Fossil fuel production

GDP per capita

Population

Urbanization

Industrialization

Greenhouse emission

0

1

1

1

1

1

0

1

0

1

1

1

0

1

1

1

1

0

0

0

1

1

1

1

0

1

1

1

0

1

1

1

1

1 1

1 0

1 1

0 1

0 0

1 1

0 0

1 1

1 1

1 1

1 1

1 1 1 0 0 1

1 1 1 1 1 1

1 1 1 1 1 1

1 1 1 1 1 1

1 1 1 1 1 0

0 1 1 1 1 1

0 0 0 0 0 1

0 0 0 1 1 1

0 0 1 0 1 1

1 1 1 1 0 1

1 1 1 1 1 0

population

urbanization

GDP per capita

industrialization

renewable energy production

fossil fuel production

renewable energy investment

nuclear energy investment

primary energy import

greenhouse emission primary energy consumption

Fig. 2. Causal cognitive map for energy investment policies.

Experts who responded the survey is composed of academics in renewable energy, researchers of institutes like water resources and bio-energy production and even political experts designing regulation proposal. The whole pair wise comparison matrix is given in Table 1. The matrix is read from row to column. The resulting aggregated map of the matrix is shown in Fig. 2. Evaluation of experts results in the causal cognitive map shown in Fig. 2. Where, there are numerous relations between these 11 driving forces. Fossil fuel production, renewable energy production, energy import and greenhouse emission are the variables seem more central which means they are affected by the other variables more than others. In the same way, population, urbanization and industrialization are not affected by the other factors. Only GDP has direct impact on renewable and nuclear investments.

4.3. Bayesian network model Bayesian causal cognitive map for energy investment policy is shown in Fig. 3. The procedure that is used to transport CM to BN are the following [22]:  CM is a dependence map which guarantees that connected concepts are dependent, but it may display dependent variables as separated variables. On the other hand, BN is an independence map, which guarantees that separated variables are conditionally independent given other variables. In CMs, an arrow from A to B (A ? B) means that A causes a (positive or negative) change onB. However in BNs, A ? B means B is dependent on A that the probability of occurrence of a state in B, is dependent on the given state of A. In other words, A is relevant to B. At the same time, in CMs, A ? B ? C refers to the effect of A on B and

D. Cinar, G. Kayakutlu / Knowledge-Based Systems 23 (2010) 267–276

271

Fig. 3. Bayesian network for energy investment policies.

effects of B on C, so there is an indirect effect of A on C. In BNs, A is relevant to C but if we know the true state of B, knowledge of A is irrelevant to C. So A is conditionally independent of C given B. In BMs an absence of an arrow implies conditional independence between variables. For example, as shown in Fig. 2, an arrow is drawn from population to GDP. Because of the aim of making inferences about investment policies, the dependency of these variables can be disregarded as seen in Fig. 3.  Determining the cause–effect relations is another essential point to transform CM to BM. There are two types of reasoning in cause–effect relations: deductive and abductive. The relation is from cause to effect in deductive reasoning while it is from effect to cause in abductive reasoning. For example, in CM of energy investment policy, there is a direct link from renewable investment to renewable energy production. This is a relation from effect to cause, so it is deductive reasoning. If the arrow was in opposite direction (from renewable energy production to renewable investment), it would be abductive reasoning. The relations are investigated one by one if the direction of relation is proper.  The third procedure is identifying the direct and indirect relations between variables. For example urbanization effects both energy consumption and greenhouse emission. In fact, urbanization causes the greenhouse emission through the increase of energy consumption. So the direct effect from urbanization to greenhouse should be eliminated. This reduction can be seen in Fig. 3. Determining the direct and indirect relations helps us to identify the nature of relations. Also the complexity of causal model can be decreased and the representation of relations will be simple.  The last procedure is eliminating the cycles. As mentioned in Section 3, there should be no cycles in BNs. Because there is no loop in CM, Fig. 2, we do not need edit or delete any variable. Data, from 1970 to 2007, is collected from a variety of governmental and scientific information resources [48]. Renewable energy consumption, fossil fuel production, energy import, primary energy consumption collected from the Ministry of Energy and National Resources. Population, GDP, urbanization and industrialization are collected from the Turkish Central Bank. Finally the greenhouse emission data is from Turkish Statistical Institute. A graphical mapping package, Netica Software is used to design the

cognitive map, analyze the probabilistic values and make inferences. All concepts were converted to the increased rate of change to determine all the probabilistic dependencies. As an example energy consumption was converted into the rate of energy consumption change. After this revision, the related values are grouped into the equal three groups to construct the states of each variable. The probabilistic values were calculated by filtering. At the filtration, the experts judgments were considered to find out the impossible combinations of events and possible combinations that have not occur in the considered time period. Renewable energy investment and nuclear energy investment are considered as decision nodes. So they have two states as ‘‘no investment” and ‘‘investment”. If a decision node is the given condition of a state then expert judgments are used for determining the posterior marginal probabilities. Posterior probabilities can be derived by Bayes’ Formula as following:

PðBjAi ÞPðAi Þ PðAi jBÞ ¼ Pn ; j¼1 PðBjAj ÞPðAj Þ where union of Aj, j = 1, . . . , n, is the sample space and intersection of any Aj is empty set. Fig. 4 provides a simple example of Bayes’ rule. In Fig. 4, all five concepts (population, urbanization, GDP per capita, industrialization and primary energy consumption) have three states: low, medium and high. The change in population will be low, medium and high with probabilities 0.47, 0.39 and 0.14, respectively. The main consequence of this table is if the increased rate of change of population is low, urbanization is low, GDP per capita is low and industrialization is low, then the probability of the increased rate of change of primary energy consumption being low will be 1. After acquiring the posterior probabilities, the experts are asked to verify the probabilities. Some changes have done after the validation by experts. 4.4. Scenario analysis The initialised probability values are shown in Fig. 5. Both nuclear and renewable stations are not invested in this figure reflecting the current situation. Thus, for example, the probability that the greenhouse emission increase is between 6% and 14% is

272

D. Cinar, G. Kayakutlu / Knowledge-Based Systems 23 (2010) 267–276

Population

Urbanization

P(P) Low Medium High 0.47 0.39 0.14

P(U) Low Medium High 0.44 0.42 0.14

GDP per capita

Industrialization

P(I)

P(G) Low Medium High 0.22 0.50 0.28

Low Medium High 0.12 0.69 0.19

Primary energy consumption

P(E|GPUI) Low,Low,Low,Low Low,Low,Low,Medium Low,Low,Low,High … High,High,High,High

Low Medium High 1 0 0 0.6 0.4 0 0.5 0.5 0 0

0

1

Fig. 4. Example of conditional probability tables.

GDP_per_capita -0.3 to -0.05 22.0 50.0 -0.05 to 0.2 28.0 0.2 to 0.45 0.09 ± 0.19

population 47.0 0.01 to 0.015 0.015 to 0.019 39.0 0.019 to 0.069 14.0 0.0187 ± 0.012

fossil_fuel_production -0.1 to -0.01 4.81 -0.01 to 0.08 45.7 0.08 to 0.17 49.5 0.0752 ± 0.059

urbanization 44.0 0 to 0.01 0.01 to 0.03 42.0 0.03 to 0.05 14.0 0.0162 ± 0.013

primary_energy_consumption -0.07 to 0 12.0 0 to 0.06 46.6 0.06 to 0.12 41.3 0.047 ± 0.045

greenhouse_emission -0.1 to -0.02 6.13 -0.02 to 0.06 34.1 0.06 to 0.14 59.8 0.0629 ± 0.054

nuclear_generation investment 0 no investment 0

industrialization 12.0 -0.1 to 0 69.0 0 to 0.1 0.1 to 0.15 19.0 0.0522 ± 0.055

renewable_energy_production 65.9 -0.1 to -0.03 22.9 -0.03 to 0.11 11.1 0.11 to 0.4 -0.00529 ± 0.11

energy_import -0.2 to 0 26.6 0 to 0.1 46.3 0.1 to 0.4 27.0 0.0641 ± 0.14

renewable_investment investment 0 no investment 0

Fig. 5. Bayesian map in initialised state.

0.598, while the probability that the primary energy import increase is between 10% and 40% is 0.27. Greenhouse emission is influenced by the fossil fuel production, renewable energy production, nuclear generation and energy import. Energy import is affected by the fossil fuel production, renewable energy production, nuclear generation and primary energy consumption. In scenario analysis we will invest on either renewable or nuclear energy. Both investments cannot be done unless an efficient utilisation of nuclear investment is defined. It is assumed in this study that the investment cost is constant. The impacts of three scenarios on GDP and industrialization are explored. The first scenario is optimistic where, GDP and industrialization increase in the highest interval. Second scenario is stable,

based on the current trend. Increase of GDP, industrialization, population and urbanization are taken into account at the same level of 2007. Since the Turkish economy is fluctuating, the third and the last scenario is reserved for a pessimistic change. In this scenario, the influence of decrease in GDP and industrialization will be analyzed. In each scenario, two options (nuclear or renewable) are tried to find the most effective one. 4.5. Case results The first analysis is performed for optimistic scenario. The probabilistic results of nuclear investment under optimistic scenario

273

D. Cinar, G. Kayakutlu / Knowledge-Based Systems 23 (2010) 267–276

GDP_per_capita 0 -0.3 to -0.05 0 -0.05 to 0.2 100 0.2 to 0.45 0.325 ± 0.072

population 47.0 0.01 to 0.015 0.015 to 0.019 39.0 0.019 to 0.069 14.0 0.0187 ± 0.012

fossil_fuel_production -0.1 to -0.01 55.0 -0.01 to 0.08 35.0 0.08 to 0.17 10.0 -0.0055 ± 0.066

urbanization 44.0 0 to 0.01 0.01 to 0.03 42.0 0.03 to 0.05 14.0 0.0162 ± 0.013

primary_energy_consumption -0.07 to 0 0 0 to 0.06 0 0.06 to 0.12 100 0.09 ± 0.017

greenhouse_emission -0.1 to -0.02 25.3 -0.02 to 0.06 52.6 0.06 to 0.14 22.1 0.0174 ± 0.06

nuclear_generation investment 0 no investment 0

industrialization 0 -0.1 to 0 0 0 to 0.1 100 0.1 to 0.15 0.125 ± 0.014

renewable_energy_production 60.0 -0.1 to -0.03 -0.03 to 0.11 26.6 0.11 to 0.4 13.4 0.00581 ± 0.12

energy_import -0.2 to 0 33.1 0 to 0.1 54.3 0.1 to 0.4 12.6 0.0255 ± 0.12

renewable_investment investment 0 no investment 0

Fig. 6. Optimistic scenario with nuclear investment.

GDP_per_capita 0 -0.3 to -0.05 -0.05 to 0.2 0 0.2 to 0.45 100 0.325 ± 0.072

population 0.01 to 0.015 47.0 0.015 to 0.019 39.0 0.019 to 0.069 14.0 0.0187 ± 0.012

fossil_fuel_production -0.1 to -0.01 33.3 -0.01 to 0.08 53.4 0.08 to 0.17 13.3 0.017 ± 0.064

urbanization 0 to 0.01 44.0 0.01 to 0.03 42.0 0.03 to 0.05 14.0 0.0162 ± 0.013

primary_energy_consumption -0.07 to 0 0 0 to 0.06 0 0.06 to 0.12 100 0.09 ± 0.017

greenhouse_emission 31.7 -0.1 to -0.02 -0.02 to 0.06 45.9 22.4 0.06 to 0.14 0.0126 ± 0.063

nuclear_generation investment 0 no investment 0

industrialization -0.1 to 0 0 0 to 0.1 0 0.1 to 0.15 100 0.125 ± 0.014

renewable_energy_production 13.4 -0.1 to -0.03 26.6 -0.03 to 0.11 60.0 0.11 to 0.4 0.155 ± 0.14

energy_import -0.2 to 0 28.2 0 to 0.1 58.0 0.1 to 0.4 13.9 0.0355 ± 0.12

renewable_investment investment 0 no investment 0

Fig. 7. Optimistic scenario with renewable investment.

are shown in Fig. 6. While the increase in GDP per capita and industrialization is on the highest level, the probability that the increase of primary energy consumption to be between 6% and 12% is 1. If the nuclear power station is built in this scenario, renewable energy production and fossil fuel production decrease will have the highest probability. The investment on nuclear energy results in a decrease in the greenhouse emission and energy import (the probability of greenhouse decrease is 0.06 in Fig. 5 while 0.25 in Fig. 6). This result approves the reduction effect of nuclear energy on greenhouse and energy import. The probabilistic results of renewable investment under optimistic scenarios are shown in

Fig. 7. According to Fig. 6 and 7, renewable investment creates bigger probability on decrease of greenhouse effect than nuclear investment. However, the probability of reduction in imports is higher for nuclear investment than renewable investment. So we can infer that the amount of energy produced by nuclear power station is more than the energy generated by renewable energy foundations. In stable scenario, the level of 2007 is specified for the increase of GDP, industrialization, population and urbanization. So the future is assumed to continue with the same improvements. According to 2007 data, the rise of GDP is between 0.05 and 0.2. This

274

D. Cinar, G. Kayakutlu / Knowledge-Based Systems 23 (2010) 267–276

GDP_per_capita 0 -0.3 to -0.05 100 -0.05 to 0.2 0 0.2 to 0.45 0.075 ± 0.072

population 100 0.01 to 0.015 0 0.015 to 0.019 0 0.019 to 0.069 0.0125 ± 0.0014

fossil_fuel_production -0.1 to -0.01 62.5 -0.01 to 0.08 27.5 0.08 to 0.17 10.0 -0.0122 ± 0.066

urbanization 0 0 to 0.01 100 0.01 to 0.03 0 0.03 to 0.05 0.02 ± 0.0058

primary_energy_consumption -0.07 to 0 0 0 to 0.06 75.0 0.06 to 0.12 25.0 0.045 ± 0.031

greenhouse_emission -0.1 to -0.02 37.0 -0.02 to 0.06 50.2 0.06 to 0.14 12.8 0.000608 ± 0.058

nuclear_generation investment 0 no investment 0

industrialization 0 -0.1 to 0 100 0 to 0.1 0 0.1 to 0.15 0.05 ± 0.029

renewable_energy_production 90.0 -0.1 to -0.03 -0.03 to 0.11 6.65 0.11 to 0.4 3.35 -0.0473 ± 0.068

energy_import -0.2 to 0 64.2 0 to 0.1 32.7 0.1 to 0.4 3.15 -0.0399 ± 0.1

renewable_investment investment 0 no investment 0

Fig. 8. Stable scenario with nuclear investment.

GDP_per_capita 0 -0.3 to -0.05 100 -0.05 to 0.2 0 0.2 to 0.45 0.075 ± 0.072

population 100 0.01 to 0.015 0 0.015 to 0.019 0 0.019 to 0.069 0.0125 ± 0.0014

fossil_fuel_production -0.1 to -0.01 48.1 -0.01 to 0.08 31.0 0.08 to 0.17 21.0 0.0106 ± 0.075

urbanization 0 0 to 0.01 100 0.01 to 0.03 0 0.03 to 0.05 0.02 ± 0.0058

primary_energy_consumption -0.07 to 0 0 0 to 0.06 75.0 0.06 to 0.12 25.0 0.045 ± 0.031

greenhouse_emission -0.1 to -0.02 60.2 -0.02 to 0.06 30.1 0.06 to 0.14 9.68 -0.0204 ± 0.058

nuclear_generation investment 0 no investment 0

industrialization 0 -0.1 to 0 100 0 to 0.1 0 0.1 to 0.15 0.05 ± 0.029

renewable_energy_production 3.35 -0.1 to -0.03 -0.03 to 0.11 21.6 0.11 to 0.4 75.0 0.198 ± 0.13

energy_import -0.2 to 0 69.2 0 to 0.1 27.3 0.1 to 0.4 3.46 -0.0469 ± 0.1

renewable_investment investment 0 no investment 0

Fig. 9. Stable scenario with renewable investment.

means that the GDP can decrease to the lower limit of the range 5% or increase to the upper limit of 20%. For the population this range is between 1% and 15%. Changes in urbanization and industrialization are seen in Fig. 8. Under these circumstances the increase of primary energy consumption will be between 0 and 0.06 with probability 0.75. Figs. 8 and 9 show the effect of nuclear and renewable investments, respectively. The probabilities of the decrease in greenhouse effect and energy import are higher for renewable investment than nuclear investment. We can propose that the investment in renewable energies is more feasible than

nuclear investment according to the effects on greenhouse emission and energy import under the stable scenario. Finally the results of pessimistic scenario are shown in Figs. 10 and 11. Given that the GDP per capita and industrialization are decreasing, the probability that the primary energy consumption decrease is 0.896. Income and industrialization rate are main factors that affect the energy consumption. If nuclear investment is preferred, energy import will decrease with probability 0.781. This high probability is based on the decline in consumption. If the same amount is invested for renewable energy the probability

275

D. Cinar, G. Kayakutlu / Knowledge-Based Systems 23 (2010) 267–276

GDP_per_capita 100 -0.3 to -0.05 0 -0.05 to 0.2 0 0.2 to 0.45 -0.175 ± 0.072

population 47.0 0.01 to 0.015 0.015 to 0.019 39.0 0.019 to 0.069 14.0 0.0187 ± 0.012

fossil_fuel_production -0.1 to -0.01 74.0 -0.01 to 0.08 25.0 0.08 to 0.17 1.04 -0.0306 ± 0.049

urbanization 44.0 0 to 0.01 0.01 to 0.03 42.0 0.03 to 0.05 14.0 0.0162 ± 0.013

primary_energy_consumption -0.07 to 0 89.6 0 to 0.06 10.4 0.06 to 0.12 0 -0.0283 ± 0.028

greenhouse_emission -0.1 to -0.02 53.8 -0.02 to 0.06 39.0 0.06 to 0.14 7.23 -0.0172 ± 0.055

nuclear_generation investment 0 no investment 0

industrialization 100 -0.1 to 0 0 0 to 0.1 0 0.1 to 0.15 -0.05 ± 0.029

renewable_energy_production 91.0 -0.1 to -0.03 8.97 -0.03 to 0.11 0 0.11 to 0.4 -0.0556 ± 0.038

energy_import -0.2 to 0 78.1 0 to 0.1 21.9 0.1 to 0.4 0 -0.0671 ± 0.081

renewable_investment investment 0 no investment 0

Fig. 10. Pessimistic scenario with nuclear investment.

GDP_per_capita 100 -0.3 to -0.05 0 -0.05 to 0.2 0 0.2 to 0.45 -0.175 ± 0.072

population 47.0 0.01 to 0.015 0.015 to 0.019 39.0 0.019 to 0.069 14.0 0.0187 ± 0.012

fossil_fuel_production -0.1 to -0.01 50.3 -0.01 to 0.08 47.3 0.08 to 0.17 2.43 -0.00809 ± 0.056

urbanization 44.0 0 to 0.01 0.01 to 0.03 42.0 0.03 to 0.05 14.0 0.0162 ± 0.013

primary_energy_consumption -0.07 to 0 89.6 0 to 0.06 10.4 0.06 to 0.12 0 -0.0283 ± 0.028

greenhouse_emission 26.9 -0.1 to -0.02 -0.02 to 0.06 42.1 31.0 0.06 to 0.14 0.0232 ± 0.065

nuclear_generation investment 0 no investment 0

industrialization 100 -0.1 to 0 0 0 to 0.1 0 0.1 to 0.15 -0.05 ± 0.029

renewable_energy_production 44.8 -0.1 to -0.03 -0.03 to 0.11 46.9 0.11 to 0.4 8.28 0.0107 ± 0.097

energy_import -0.2 to 0 42.2 0 to 0.1 42.1 0.1 to 0.4 15.7 0.0181 ± 0.13

renewable_investment investment 0 no investment 0

Fig. 11. Pessimistic scenario with nuclear investment.

Table 2 Probability of falling in greenhouse emission and energy import under different scenarios.

Optimistic scenario Stable scenario Pessimistic scenario

Nuclear inv. Renewable inv. Nuclear inv. Renewable inv. Nuclear inv. Renewable inv.

Greenhouse emission

Energy import

0.25 0.31 0.37 0.60 0.54 0.26

0.33 0.28 0.64 0.69 0.78 0.42

for import is 0.42. This difference in results is caused by the structure of investments. Nuclear investment has been done only once, while renewable investments are iterative. So in the pessimistic scenario the construction of renewable facilities takes more time than the nuclear stations. 5. Conclusion Bayes Maps are effective in scenario creation of complex planning as in energy sector. Energy is considered to be a key player in the generation of wealth and a significant component in

276

D. Cinar, G. Kayakutlu / Knowledge-Based Systems 23 (2010) 267–276

economic development. Sustainable development demands a sustainable supply of energy sources. Turkey is taken as the case for implementing the proposed Bayes Mapping model since it has various energy resources but relying on imported fossil energy. In Turkey, alternative energy sources must be used more widely to be less dependent on foreign resources. This study contributes to literature by using CM and BNs in energy sector. A basic framework is designed for energy investment policies. Declining of greenhouse effect and energy imports are decision parameters in this study. Since the economic conditions are not stable in Turkey, different scenarios are held by BNs. The probability of intended states for decision variables are summarized in Table 2. These probabilities are considered to make a decision. In second scenario (stable scenario), renewable investment overcomes to nuclear investment. In pessimistic scenario nuclear investment seems to be better than renewable investment according to the decision variables (greenhouse emission and energy import). However in optimistic scenario there is a tie between these two alternatives. So we need to look up for different attributes to select the best one. The cost of implementation has been taken constant for both alternatives. An important improvement of the study should also include the foreign/domestic investment ratios and operating costs as two of the decision variables in these analyses. The third factor that should be considered in further studies is the cost of environmental safety. Since it is a very critical decision to be made by governmental authorities, the complexity of the map caused by adding those factors will be ignored. We have also observed that the dynamic structure of the macro economic factors analyzed, the results achieved by using BNs are suggested to be compared by the results of system dynamics models. Determination of priority of the influential factors by the energy experts and the decision makers will also make the results more realistic.

References [1] Y.P. Cai, G.H. Huang, Z.F. Yang, Q. Tan, Identification of optimal strategies for energy management systems planning under multiple uncertainties, Applied Energy 86 (4) (2009) 480–495. [2] X. Changliang, S. Zhanfeng, Wind energy in China: current scenario and future perspectives, Renewable and Sustainable Energy Reviews (8) (2009) 1966– 1974. [3] R.L. Hobbs, A scenario-directed computational framework to aid decisionmaking and systems development, Georgia Institute of Technology, 2005, 283 pp., AAT 3183205. [4] M. Kaliski, Possible scenarios of the European union energy policy versus stable deliveries of natural gas to central and eastern Europe based on the Russian energy policy, Rynek Energii 3 (2009) 2–12. [5] M. Goyal, R. Jha, Introduction of renewable energy certificate in the Indian scenario, Renewable and Sustainable Energy Reviews 13 (6–7) (2009) 1395– 1405. [6] K. Kowalski, S. Stagl, R. Madlener, I. Omann, Sustainable energy futures: methodological challenges in combining scenarios and participatory multicriteria analysis, European Journal of Operational Research 197 (3) (2009) 1063–1074. [7] S.A. Gabriel, J. Zhuang, R. Egging, Solving stochastic complementarity problems in energy market modeling using scenario reduction, European Journal of Operational Research 197 (3) (2009) 1028–1040. [8] E. Gnansounou, S. Pierre, A. Quintero, J. Dong, A. Lahlou, A multi-agent approach for planning activities in decentralized electricity markets, Knowledge Based Systems 20 (4) (2007) 406–418. [9] D. Iwashita, H. Mori, Risk quantification for ANN based short-term load forecasting, Electrical Engineering in Japan 166 (2) (2009) 54–62. [10] B. Wilke, J. Metzgar, K. Johnson, S. Semwal, B. Snyder, K.C. Yu, D. Neafus, Crossover Applications, VR’09: Proceedings of the 2009 IEEE Virtual Reality Conference, 2009. [11] C.-H. Lee, Improving classification performance using unlabeled data: Naive Bayesian case, Knowledge Based Systems 20 (3) (2007) 220–224. [12] J. Chen, H. Huang, F. Tian, S. Tian, A selective Bayes classifier for classifying incomplete data based on grain ratio, Knowledge Based Systems 21 (7) (2008) 530–534. [13] J. Xiao, C. He, X. Jiang, Structure identification of Bayesian classifiers based on GMDH, Knowledge Based Systems 22 (6) (2009) 461.

[14] H.S. Soyhan, Sustainable energy production and consumption in Turkey: a review, Renewable and Sustainable Energy Reviews 13 (2009) 1350–1360. [15] K. Kaygusuz, Energy situation, future developments, energy saving and energy efficiency in Turkey, Energy Sources 21 (1999) 405–416. [16] K. Kaygusuz, A. Sari, Renewable energy potential and utulization in Turkey, Energy Conversion and Management 44 (2003) 459–478. [17] A. Demirbasß, Energy balance, energy sources, energy policy, future developments and energy investments in Turkey, Energy Conversion and Management 42 (2001) 1239–1258. [18] B.L. Bowerman, R.T. O’Connell, A.B. Koehler, Forecasting, Time Series and Regression, Thomson Brooks/Cole, USA, 2005. [19] F.V. Jensen, Bayesian Networks and Decision Graphs, Springer-Verlag Inc., New York, 2001. [20] S. Nadkarni, P.P. Shenoy, A Bayesian network approach to making inferences in causal maps, European Journal of Operational Research 128 (2001) 479–498. [21] C. Eden, Analyzing cognitive maps to help structure issues or problems, European Journal of Operational Research 159 (2004) 673–686. [22] S. Nadkarni, P.P. Shenoy, A causal mapping approach to constructing Bayesian networks, Decision Support Systems 38 (2004) 259–281. [23] K. Siau, X. Tan, Improving the quality of conceptual modelling using cognitive mapping techniques, Data and Knowledge Engineering 55 (2005) 343–365. [24] S.Ö. S ß ahin, F. Ülengin, B. Ülengin, Using neural networks and cognitive mapping in scenario analysis: the case of Turkey’s inflation dynamics, European Journal of Operational Research 158 (2006) 124–145. [25] C.Y. Miao, A. Goh, Y. Miao, Z.H. Yang, Agent that models, reasons and makes decisions, Knowledge Based Systems 15 (2002) 203–211. [26] C. Miao, Q. Yang, H. Fang, A. Goh, A cognitive approach for agent-based personalized recommendation, Knowledge Based Systems 20 (2007) 397–405. [27] J.L. Salmeron, Augmented fuzzy cognitive maps for modelling LMS critical success factors, Knowledge Based Systems 22 (2009) 275–278. [28] E. Alpaydın, Introduction to Machine Learning, The MIT Press, Cambridge, London, England, 2004. [29] P. Larranaga, S. Moral, Probabilistic graphical models in artificial intelligence, Applied Soft Computing, Corrected Proof, 2009, . [30] K. Baumgartner, S. Ferrari, G. Palermo, Constructing Bayesian networks for criminal profiling from limited data, Knowledge Based Systems 21 (2008) 563–572. [31] S.P. Cheon, S. Kim, S.Y. Lee, C.B. Lee, Bayesian networks based rare event prediction with sensor data, Knowledge Based Systems 22 (2009) 336–343. [32] B. Min, J. Kim, C. Choe, H. Eom, R.I. McKay, A compound framework for sports results prediction: a football case study, Knowledge Based Systems 21 (2008) 551–562. [33] P. Trucco, E. Cagno, F. Ruggeri, O. Grande, A Bayesian belief network modelling of organisational factors in risk analysis: a case study in maritime transportation, Reliability Engineering and System Safety 93 (2008) 823–834. [34] S. Gupta, H.W. Kim, Linking structural equation modeling to Bayesian networks: decision support for customer retention in virtual communities, European Journal of Operational Research 190 (2008) 818–833. [35] E. Aktas, F. Ulengin, S.O. Sahin, A decision support system to improve the efficiency of resource allocation in healthcare management, Socio-Economic Planning Science 41 (2007) 130–146. [36] F. Ulengin, S. Onsel, Y.I. Topcu, E. Aktas, O. Kabak, An integrated transportation decision support system for transportation policy decisions: the case of Turkey, Transportation Research Part A 41 (2007) 80–97. [37] E.J.M. Lauria, P.J. Duchessi, A Bayesian belief network for IT implementation decision support, Decision Support Systems 42 (2006) 1573–1588. [38] N. Fenton, M. Neil, Making decisions: using Bayesian nets and MCDA, Knowledge Based Systems 14 (2001) 307–325. [39] V.Sß. Ediger, E. Kentel, Renewable energy potential as an alternative to fossil fuels in Turkey, Energy Conversion and Management 40 (1999) 743–755. [40] A.I. Chatzimouratidis, P.A. Pilavachi, Multicriteria evaluation of power plants impact on the living standard using the analytic hierarchy process, Energy Policy 36 (2008) 1074–1089. [41] D. Martinsen, V. Krey, Compromises in energy policy – using fuzzy optimization in an energy systems model, Energy Policy 36 (8) (2008) 2973– 2984. [42] Y. Mulugetta, N. Mantajit, T. Jackson, Power sector scenarios for Thailand: an explanatory analysis 2002–2022, Energy Policy 35 (2007) 3256–3269. [43] P. Krukanont, T. Tezuka, Implications of capacity expansion under uncertainty and value of information: the near-term energy planning of Japan, Energy 32 (10) (2007) 1809–1824. [44] S. Koskela, J. Seppala, A. Lipp, M.R. Hiltunen, E. Pold, S. Talve, Estonian electricity supply scenarios for 2020 and their environmental performance, Energy Policy 35 (2007) 3571–3582. [45] T. Mezher, R. Chedidy, W. Zahabi, Energy resource allocation using multiobjective goal programming: the case of Lebanon, Applied Energy 61 (1998) 175–192. [46] R.K. Agrawal, S.P. Singh, Energy allocations for cooking in UP households (India) a fuzzy multi-objective analysis, Energy Conversion and Management 42 (2001) 2139–2154. [47] M. Asıf, T. Muneer, Energy supply, its demand and security issues for developed and emerging economies, Renewable and Sustainable Energy Reviews 11 (2007) 1388–1413. [48] WEC-TMK (World Energy Council – Turkish National Committee), World and Turkey Energy Statistics, Ankara, Turkey, in Turkish, 2006.