Bayesian Networks based Policy Making in the Renewable

Energy Sector

Moldir Zholdasbayeva

1,*

, Vasilios Zarikas

1,2,†

and Stavros Poulopoulos

3,‡

Department of Mechanical and Aerospace Engineering, Nazarbayev University,

Kabanbay Batyr 53, Nur-Sultan 010000, Kazakhstan

General Department, Theory Division, University of Thessaly, Volos, Greece

Department of Chemical and Material Engineering, Nazarbayev University,

Kabanbay Batyr 53, Nur-Sultan 010000, Kazakhstan

Keywords: Bayesian Networks, Expert Models, Renewable Energy, Geothermal Energy, Hydro Energy.

Abstract: Extensive research on energy policy nowadays combines theory with advanced statistical tools such as Bayes-

ian networks for analysis and prediction. The majority of these studies are related to observe energy scenarios

in various economic or social conditions, but only a few of them target the renewable energy sector. Therefore,

it is crucial to design a method to understand the causal relationships between variables such as consumption,

greenhouse emissions, investment in renewables and investment in fossil fuels. This research paper aims to

present expert models using the capabilities of Bayesian networks in the renewable energy sector, considering

renewables in two countries: Germany and Italy. For this purpose, expert models are built in BayesiaLab with

supervised learning. An augmented naïve model is applied to quantitative data consisting of the consumption

rate of geothermal and hydro energy sectors. As a result, it is indicated that in the optimum case, geothermal

and hydro energy consumption will be increased in parallel with investment. It is found that, as oil price grows,

greenhouse emissions will decrease. The precision of the expert model is no less than 90%.

1 INTRODUCTION

Bayesian networks are widely being used in various

fields of study, namely in environmental (Martos et

al., 2016; Marcos et al., 2018; Ropero, Renooij and

Gaag, 2018), ecological (Barton et al., 2016; Corani

and Scanagatta, 2016; McLaughlin and Reckhow,

2017; Orun et al., 2018; Liu and Callies, 2019), sus-

tainable development (Keshtkar et al., 2013; Franco

et al., 2016), agricultural (Mukashema, Veldkamp

and Vrieling, 2014; Barton et al., 2016), mapping

(Landuyt et al., 2015; Gonzalez - Redin et al., 2016),

risk management (Gerstenberger et al., 2015; Tang et

al., 2016a, 2016b; Kabir and Papadopoulos, 2019),

reliability (Amrin, Zarikas and Spitas, 2018;

Kameshwar et al., 2019), medicine (Zarikas,

Papageorgiou and Regner, 2014; Zarikas et al., 2018)

and safety (Zarikas et al., 2013; Washington et al.,

2019). Further, the theory behind Bayesian networks

https://www.researchgate.net/profile/Moldir_Zholdasbayeva

†

https://research.nu.edu.kz/en/persons/vasileios-zarikas

‡

https://research.nu.edu.kz/en/persons/stavros-poulopoulos

can be applied in diverse research areas to analyze the

numerous data (Bapin and Zarikas, 2014; Amrin,

Zarikas and Spitas, 2018). It is a prognostic method

for conducting diagnostics and calculating probabili-

ties, which is useful for uncertain data (Conrady,

Jouffe and Elwert, 2014; Zarikas, 2014).

It is important to point out that the majority of

studies have been dedicated to the energy sector based

on the latest environmental report (Borunda et al.,

2016). However, with the vast volume of data, there

are a few studies related to the use of Bayesian

networks in the renewable energy sector (Res et al.,

2009; Cinar and Kayakutlu, 2010; Borunda et al.,

2016; Gambelli et al., 2017), particularly for the

energy policy cases (Kumar et al., 2010; Bhowmik et

al., 2017). Therefore, it is pivotal to develop a tech-

nique to predict future progress in this field for the

sources of energy such as geothermal energy, hydro

energy, bioenergy, solar energy, wind energy. For this

Zholdasbayeva, M., Zarikas, V. and Poulopoulos, S.

Bayesian Networks based Policy Making in the Renewable Energy Sector.

DOI: 10.5220/0008925004530462

In Proceedings of the 12th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2020) - Volume 2, pages 453-462

ISBN: 978-989-758-395-7; ISSN: 2184-433X

453

purpose, energy policies are important, so that they can

contribute to the understanding of the future market for

renewable energy production and consumption.

Case studies may be considered to be an effective

method to create energy policies (Kim et al., 2018).

For this, the algorithms in Bayesian networks can be

modified and built in relevance with structural re-

strictions of the system (Campos and Castellano,

2007; Pitchforth and Mengersen, 2013; Perreault and

Sheppard, 2019). Two approaches can be applied to

data analysis in the renewable energy sector such as

supervised and unsupervised learning. In case of un-

supervised learning, it is useful to determine the

causal relationships between variables, otherwise, su-

pervised learning allows referring to one variable at a

time (Conrady, Jouffe and Elwert, 2014). One of the

relevant studies using unsupervised learning in BNs

in this field is dedicated to forecasting the investment

in the renewable energy sector in Turkey for the years

1970 to 2007 (Cinar and Kayakutlu, 2010). Several

variables, which were expected to have even a slight

impact on energy investment, were chosen. Also,

gross domestic product, renewable energy produc-

tion, fossil fuel production, urbanization, and indus-

trialization were directly imported to Bayesian net-

work. Three scenarios were created (optimistic, stable

and pessimistic). As a result, in the optimistic sce-

nario investment in renewables grew with a gradual

decrease in greenhouse emissions. This leads to the

fact that with a high rate of industrialization process

and GDP the demand for renewables increases.

Another study in this area is related to the predic-

tion of the future market of biofuels using supervised

learning in Italy by 2030 (Gambelli et al., 2017). Two

scenarios were developed such as “best scenario” and

“worst scenario”. As an outcome, in the best scenario

biofuels would demonstrate the highest percentage of

market involvement in the near future. Nevertheless,

one requirement is needed: advanced technological

development and environmental policies should be

taken into action simultaneously.

The main aim of current research is to develop ex-

pert models (Tselykh, Tselykh and Barkovskii, 2018;

Jha, 2019)

for the renewable energy sector using a su-

pervised learning technique. Methodologies above

will be modified to create expert models, concerning

two energy sectors: hydro energy and geothermal en-

ergy. In addition, the application of Bayesian net-

works in the determination of the best scenarios for

geothermal energy shows only 2 percent of research

papers addressed this type of renewable, whilst for

hydro energy, it is 21 percent (Borunda et al., 2016).

Particularly, these studies will be concentrated on the

widely used energy source (hydro energy) and on the

least favourable one (geothermal energy). Factors as

GDP, fossil fuel and renewables consumption, green-

house emissions will be taken into account to verify

the results obtained from previous research.

In the following section, modified methods for

this research will be given and explained. K-Folds

analysis method will be discussed. In Section 3, re-

sults will be shown regarding the optimum and mini-

mum cases for the renewable energy sector with max-

imum and minimum consumption rate. In Section 4,

conclusions will be drawn regarding expert models

and the impact of selected variables on renewables.

2 METHODOLOGY

Data analysis for the identification of the optimum

and minimum cases for the renewable energy sector

is undertaken using BayesiaLab software (Conrady

and Jouffe, 2015). The optimum case is this with the

highest percentage of the increase in renewable con-

sumption and the minimum case is the less optimistic,

where a significant decrease in renewable consump-

tion will be observed by inserting evidence to a

model. Data on renewables is obtained from the offi-

cial site of OECD (Organisation for Economic Coop-

eration and Development). OECD is an organization

that provides a wide range of data on economics, wel-

fare, energy, and investment with open availability.

Specifically, data on import and export, the consump-

tion and the electricity production and price, the pro-

duction of greenhouse emissions is collected from the

dataset ‘Renewables balance’ and data on GDP is

taken from ‘Energy statistics’ for two countries (Ger-

many and Italy) for the years 1990 – 2017 (Organisa-

tion for Economic Cooperation and Development

[OECD], n.d.). The information on the investment for

both renewables and fossil fuels is extracted from

‘RD&D Budget’, whereas data on patents is from ‘Pa-

tent statistics’. Finally, data on gold and oil prices is

from ‘Main economic indicators’. Renewables such

as geothermal energy and hydro energy are used for

the preliminary analysis related to its full availability

and will be shown as separate models next sections.

2.1 Augmented Naïve Bayesian Model

To analyze those cases mentioned above, expert

models are created. The first stage for this is to

identify the discretization method and the learning

type. The supervised learning algorithm is used for

this research, considering that consumption of either

renewable type will be set as a target variable.

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

454

Before these steps will be conducted, it can be

mentioned that the discretization type is chosen to be

“Tree” for supervised learning for “Oil_Price (RI)”,

“TRenEnergy_import (ktoe)”, “Gold_ Price (US dol-

lars)”, “TRenEnergy_export (ktoe)”,“Geothermal_-

electricity (GWh)”, “GDP”, “Investment_renewables

(millions)”, “Investment_fossilfuels (millions)”, “En-

vironPolicy (%)”, “Electricity_Price (RI)”, “GhEmis-

sions_production (tonnes)”, “Geothermal_techpa-

tents”. A tree is considered to be the most commonly

used discretization methods for supervised learning.

The major process involves using the class infor-

mation of child nodes and applying a hierarchical dis-

cretization based on the correlation. The reason be-

hind applying this type of discretization lies in the op-

timization process of correlation between the target

variable (“Geothermal_consumption (ktoe)”) and

predictor variables.

The manual discretization process is applied to

“Geothermal_consumption (ktoe)” with generating

states by choosing R2-GenOpt discretization type

based on the regression model (Montgomery and

Runger, 2014):









/



(1)

It can be clearly shown from the above formula

that a sum due to regression is over a total sum of

squares, considering that discretization is chosen for

strengthening the connection between discrete and

continuous variables. This is called genetic optimiza-

tion.

It can be noted that all variables are continuous

while supervised learning requires at least one dis-

crete target variable. Therefore, using R2-GenOpt

converts a continuous variable to a discrete one (“Ge-

othermal_consumption (ktoe)”). Values for missing

parts of the data are generated by Missing Values Im-

putation, which gives the Structural EM method, ap-

plicable to a small set for data for the purpose of this

research (Conrady, Jouffe and Elwert, 2014). The

method of Structural Expectation Maximization is

based on finding the ‘most suitable’ estimate for the

missing part of the dataset by evaluating possible

structures for the parameter.

For the strength of each arc, Pearson’s correlation

method is used by showing the strongest and the least

strong connections. Thus, the structural analysis uses

Pearson’s correlation coefficients





,







,



/







∙



 (2)

for evaluating the differences between two nodes and

summation of the resultant values, which gives the

values with high precision and accuracy (Mu, Liu and

Wang, 2018).

Furthermore, a supervised learning procedure is

carried out by choosing the augmented naïve model.

It is applicable to analyze a small set of data. The

structure of the augmented naïve model is character-

ized by having similar properties as the naïve model,

but adding a higher precision and accuracy to the

model (Figure 1(b)). It is, therefore, achieved by cre-

ating new connections between the adjacent nodes

(Montgomery and Runger, 2014). In the naïve model,

nodes are considered to be independent of each other

without any correlation between neighboring nodes

(Figure 1(a)).

Figure 1: A simplistic naïve and augmented naïve models

(Y – target node; N1, N2, N3 – child nodes): a) naïve model;

b) augmented naïve model.

The augmented naïve model uses the famous

Bayes formula to identify joint probabilities not only

between dependent variables and one target variable,

but also correlations between several child nodes.











 | ∙ / (3)

2.2 K-Folds Analysis

In order to demonstrate the precision of each model

presented in these studies, k-folds analysis is exe-

cuted. K-Fold cross-validation is useful in machine

learning to evaluate the precision of a machine learn-

ing model on unseen data. It is a technique in which a

particular sample of a dataset is reserved on which

there is no need to train the data. Then, the model is

tested on this sample before finalizing it. Thus, a

small reserved sample is utilised to calculate how the

model is supposed to behave in general when used to

make predictions on data, but not used during the

Bayesian Networks based Policy Making in the Renewable Energy Sector

455

training of the model. The algorithm is characterised

by a single parameter “k” denoting the number of

groups that our sample is to be split into. This is the

reason the method it is called k-fold cross-validation.

3 RESULTS AND DISCUSSIONS

In this section, the method explained previously is ap-

plied to create expert models in the renewable energy

sector for two countries: Germany and Italy. Two

sources of energy are used in predicting the optimum

and minimum cases for consumption such as geother-

mal energy and hydro energy. The dependencies be-

tween child nodes are built by an automatic calcula-

tion of correlation using supervised learning.

3.1 Augmented Naïve Bayesian Model

for Renewables in Germany

In this subsection, two expert models are presented

for geothermal energy and hydro energy in Germany.

For both cases, consumption is considered to be a tar-

get variable with the manual discretization. In case of

geothermal energy, the discretization method is cho-

sen to be “Tree” for “Gold_Price (US dollars)”,

“TRenEnergy_export (ktoe)”, “Geothermal_electri-

city (GWh)”, “GDP”, “Investment_renewables (mil-

lions)”, “Investment_fossilfuels (millions)”, “Envi-

ronPolicy (%)”, “Electricity_Price (RI)”, “GhEmis-

sions_production (tonnes)”. The type of discretiza-

tion as “Perturbed Tree” is automatically generated

states for two variables such as “TRenEnergy_Import

(ktoe)” and “Oil_Price (RI)”, whereas R2-GenOpt is

chosen for “Geothermal_techpatents”.

In Figure 2(a), the model for geothermal energy is

presented using the supervised learning algorithm

with a structural coefficient of one. In Figure 2(b), it

is shown that the new connection between variables

“Geothermal_techpatents” and “Oil_Price (RI)” is

created by editing the structural coefficient to 0.5.

But, the relationship between “TRenEnergy_import

(ktoe)” and “Oil_Price (RI)” is deleted. The correla-

tion between variables are shown in the same figure

(Figure 2(b)) using Pearson’s correlation, where the

strongest relationship is bounded to be between “Ge-

othermal_consumption (ktoe)” and “Investment_re-

newables (millions)”, “Geothermal_consumption

(ktoe)” and “GDP”, “Geothermal_consumption

(ktoe)” and “Geothermal_electricity (GWh)”, “Geo-

thermal_consumption (ktoe)” and “Electricity_Price

(RI)”. The least strong connection is between “Geo-

thermal_consumption (ktoe)” and “Geothermal_-

techpatents”.

Figure 2: Supervised learning for geothermal energy in Ger-

many: a) with the structural coefficient of 1; b) with the

structural coefficient of 0.5 and with Pearson’s correlation.

After applying the augmented naïve model for var-

iables in the geothermal energy sector, it is important

to use the joint probability to predict the optimum and

minimum cases for geothermal energy consumption.

Firstly, the optimum case for geothermal energy is ob-

served in Figure 3. By setting the evidence for the op-

timum case of “Investment_renewables (millions)” to

100%, the same percentage is obtained for the opti-

mum state of “Geothermal_consumption (ktoe)”,

whereas the minimum state of “GhEmissions_produc-

tion (tonnes)” is shown to be 80%.

Furthermore, the minimum case for geothermal en-

ergy is shown in Figure 4. By setting the evidence for

the optimum state of “Investment_fossilfuels (mil-

lions)” and the minimum state of “Gold_Price (US dol-

lars)” to 100%, the same percentage is obtained for the

minimum state of “Geothermal_consumption (ktoe)”.

The optimum state of “GhEmissions_production

(tonnes)” is described by 95.65% increase.

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

456

Figure 3: Optimum case for geothermal energy consump-

tion in Germany.

Figure 4: Minimum case for geothermal energy consump-

tion in Germany.

Table 1: Occurrences, reliability and precision of expert

model for geothermal energy in Germany.

Occurrences

Value <=89.973 (23) >89.973 (5)

<=89.973 (22) 22 0

>89.973 (6) 1 5

Reliability

Value <=89.973 (23) >89.973 (5)

<=89.973 (22) 100.0000% 0.0000%

>89.973 (6) 16.6667% 83.3333%

Precision

Value <=89.973 (23) >89.973 (5)

<=89.973 (22) 95.6522% 0.0000%

>89.973 (6) 4.3478% 100.0000%

It can be said that the optimum case where the

consumption of geothermal energy will increase is the

state with increasing investment in renewable energy.

However, the minimum case is obtained by setting a

condition with the increasing investment in fossil

fuels and the decreasing price for gold. To verify this

expert model, k-folds analysis is undertaken as shown

in Table 1 with the precision approximately equals to

95.6%, which demonstrates a quite acceptable result

for further analysis.

A similar expert model is created for the hydro en-

ergy sector as shown in Figure 5. The discretization

type is chosen to be “Tree” for three variables such as

“Hydro_electricity (GWh)”, “GDP” and “GhEmis-

sions_production (tonnes)”. The remaining set of vari-

ables is discretized by choosing “Perturbed Tree”. It is

observed (Figure 5(b)) that the strongest correlation is

between “Hydro_consumption (ktoe)” and “Hydro

_electricity (GWh)”, “GDP” and “GhEmissions_-pro-

duction (tonnes)”. The less apparent connection is the

same as with geothermal energy. New relation-ships

are created between “TRenEnergy_Import (ktoe)” and

“Hydro_techpatents”, “TRenEnergy_-Export (ktoe)”

and “Oil_Price (RI)”, “Electricity_- Price (RI)” and

“Investment_fossilfuels (millions)”, “Oil_Price (RI)”

and “GDP”.

Figure 5: Supervised learning for hydro energy in Germany:

a) with the structural coefficient of 1; b) with the structural

coefficient of 0.5 and with Pearson’s correlation.

The optimum case for hydro energy is demons-

trated in Figure 6. The maximum increase of 58.33%

in “Hydro_consumption (ktoe)” is achieved by set-

ting the value for evidence for “Investment_renew-

ables (millions)” and “Hydro_techpatents” to 100%.

The minimum case for hydro energy consumption is

the same as for geothermal energy as shown in Figure 7.

Bayesian Networks based Policy Making in the Renewable Energy Sector

457

Figure 6: Optimum case for hydro energy consumption in

Germany.

Figure 7: Minimum case for hydro energy consumption in

Germany.

The optimum condition for hydro energy with the

increasing investment in renewables leads to the grad-

ual growth of hydro consumption.

At the same time, the minimum case is shown to

be quite similar to one shown with geothermal energy

case, which is explained by using the same dependent

variables. The precision (100%) of the model is des-

cribed in Table 2.

3.2 Augmented Naïve Bayesian Model

for Renewables in Italy

In this subsection, two more expert models are de-

signed for geothermal and hydro energy in Italy. A

target variable remains the same from previous analy-

sis. In terms of geothermal energy, the discretization

method is chosen to be “Tree”, except for “Geother-

mal_techpatents” R2-GenOpt is applied. The expert

model is considered to remain stable even by chang-

ing the structural coefficient to 0.5 (Figure 8(b)).

The strongest relationship is shown between

“Oil_Price (RI)” and “Coal_Price (RI)”, “Geother-

mal_consumption (ktoe)” and “Geothermal_electri-

city (GWh)”, “Geothermal_consumption (ktoe)” and

“TRenEnergy_Import (ktoe)”. The weakest connec-

tion is between “GhEmissions_production (tonnes)”

and “EnvironPolicy (%)”.

Table 2: Occurrences, reliability and precision of the model

for hydro energy in Germany.

Occurrences

Value <=1636.371(10) >1636.371 (18)

<=1636.371 (10) 10 0

>1636.371 (18) 0 18

Reliability

Value <=1636.371 (10) >1636.371 (18)

<=1636.371 (10) 100.0000% 0.0000%

>1636.371 (18) 0.0000% 100.0000%

Precision

Value <=1636.371 (10) >1636.371 (18)

<=1636.371 (10) 100.0000% 0.0000%

>1636.371 (18) 0.0000% 100.0000%

Figure 8: Supervised learning for geothermal energy in

Italy: a) with the structural coefficient of 1; b) with the

structural coefficient of 0.5 and with Pearson’s correlation.

By giving evidence (100%) to “Investment_re-

newables (millions)” and “Geothermal_techpatents”,

“Geothermal_consumption (ktoe)” is indicated at the

maximum state of 94.94% for the case with optimum

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

458

conditions. The optimum states for “Geothermal_-

electricity (GWh)” and “TRenEnergy_import (ktoe)”

are described by the same percentage as for “Geother-

mal_consumption (ktoe)”, explained by a high corre-

lation (Figure 9).

Figure 9: Optimum case for geothermal energy consump-

tion in Italy.

The minimum case is described by 60% of

probability of decreased geothermal consumption by

giving evidence to “Coal_Price (RI)” and

“Investment_fossilfuels (millions)”. Therefore,

“GhEmissions_production (tonnes)” is maximized,

explained by the opposite correlation with “Geo-

thermal_consumption (ktoe)” (Figure 10).

Figure 10: Minimum case for geothermal energy consump-

tion in Italy.

In addition, the growth in investment for renew-

ables has a slight effect on environmental policy,

whereas GDP increases. The precision of this expert

model created for the geothermal energy sector in It-

aly is 92.3%, which is considered to be relevant to

these studies (Table 3).

Further, the augmented naïve model has been ap-

plied to the hydro energy sector in Italy. The most ob-

vious connection is described between “Hydro_con-

sumption (ktoe)” and “Hydro_electricity (GWh)”,

whereas the weakest one is between “Hyd-ro_con-

sumption (ktoe)” and “EnvironPolicy (%)”.The rela-

tionships between “Coal_Price (RI)” and “Hyd-

ro_techpatent”, “Oil_Price (RI)” and “GDP” are crea-

ted by changing the structural coefficient as shown in

Figure 11(b).

Table 3: Occurrences, reliability and precision of the model

for geothermal energy in Italy.

Occurrences

Value <=4258.527 (13) >4258.527 (15)

<=4258.527 (12) 12 0

>4258.527 (16) 1 15

Reliability

Value <=4258.527 (13) >4258.527 (15)

<=4258.527 (12) 100.0000% 0.0000%

>4258.527 (16) 6.2500% 93.7500%

Precision

Value

<=4258.527

(13)

>4258.527 (15)

<=4258.527 (12) 92.3077% 0.0000%

>4258.527 (16) 7.6923% 100.0000%

Figure 11: Supervised learning for hydro energy in Italy: a)

with the structural coefficient of 1; b) with the structural co-

efficient of 0.5 and with Pearson’s correlation.

In the optimum case, adding evidence to “Invest-

ment_renewables (millions)” and “Hydro_techpa-

Bayesian Networks based Policy Making in the Renewable Energy Sector

459

tents” leads to 100% probability of the maximum state

of “Hydro_consumption (ktoe)”. It can be, therefore,

mentioned that the price for electricity reaches its max-

imum state for the optimum situation (Figure 12).

Figure 12: Optimum case for hydro energy consumption in

Italy.

In the minimum case, setting evidence to

“Coal_Price (RI)” and “Investment_fossilfuels (mil-

lions)” leads to 100% probability for minimum level

of hydro energy consumption (Figure 13).

Figure 13: Minimum case for hydro energy consumption in

Italy.

From figures, it is obvious that GDP has a clear

impact on the consumption rate of hydro energy. The

precision of the expert model for hydro energy in Italy

equals to 100 % according to Table 4.

Table 4: Occurrences, reliability and precision of the model

for hydro energy in Italy.

Occurrences

Value <=3648.495 (18) >3648.495 (10)

<=3648.495 (18) 18 0

>3648.495 (10) 0 10

Reliability

Value <=3648.495 (18) >3648.495 (10)

<=3648.495 (18) 100.0000% 0.0000%

>3648.495 (10) 0.0000% 100.0000%

Precision

Value <=3648.495 (18) >3648.495 (10)

<=3648.495 (18) 100.0000% 0.0000%

>3648.495 (10) 0.0000% 100.0000%

4 CONCLUSIONS

Bayesian networks have been a well-known tool used

in diverse areas of science and technology. However,

its usage could be extended in the renewable energy

policy sector using the expert model.

At this work, the renewable sources such as geo-

thermal energy and hydro energy were taken into con-

cern as one of the widespread and the least preferable

types of renewables, respectively. Two expert models

were created using the augmented naïve method. Ini-

tially, structural coefficient was equal to one, then in

order to increase the precision, it was taken as a half

of value. This resulted in new causal connections. It

was noticeable that editing the structural coefficient

might give some robustness to the system, however it

was suggested to use the range no less than 0.1 and no

more than 1.

From the analysis, it was shown that the consump-

tion of geothermal energy in Germany could be opti-

mized by the increasing investment in renewables,

which proves the previous research works. Green-

house emissions were decreased to 80% for the opti-

mized case. On the other hand, the minimum case

demonstrated that the increasing investment in fossil

fuels and the cheapest price for gold resulted in a sit-

uation with the minimized state for geothermal en-

ergy consumption.

In terms of hydro energy in Germany, it was only

a slight increase in hydro consumption as a response

for the growing number of technical patents and the

investment. The minimum case showed that similar

results as for geothermal energy source, which was

explained by using the same input variables.

In case of Italy, the increasing number of tech-

nical patents and the investment in geothermal energy

lead to a considerable increase in geothermal con-

sumption, whereas a gradual change in environmental

policy could be noticed. For the minimum case, the

evidence was set to a coal price, which resulted in the

worst scenario (minimum case) for this type of renew-

able. As for hydro energy consumption, it was indi-

cated that its optimum case was set by giving the max-

imum evidence to technical patents, whereas the min-

imum situation was involved the decrease in the coal

price and the increased emissions.

Therefore, from the obtained results and precision

data, it can be said that Bayesian Networks is a suita-

ble tool for data analysis in renewable energy policy

making. Methods in previous sections will be deve-

loped further as a small set of data only was utilized

during this research. Furthermore, it is crucial to ex-

tend this method applying to other sources of renew-

able energy such as solar, wind and bio.

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

460

REFERENCES

Amrin, A., Zarikas, V. and Spitas, C. (2018) ‘Reliability

analysis and functional design using Bayesian networks

generated automatically by an “Idea Algebra” frame-

work’, Reliability Engineering and System Safety. Else-

vier Ltd, 180 (July), pp. 211–225. doi: 10.1016/j. ress.

2018.07.020.

Bapin, Y. and Zarikas, V. (2014) ‘Probabilistic Method for

Estimation of Spinning Reserves in Multi-connected

Power Systems with Bayesian Network-based Re-

scheduling Algorithm’, Proceedings of the 11th Inter-

national Conference on Agents and Artificial Intelli-

gence, pp. 840–849. doi: 10.5220/ 0007577308400849.

Barton, D. N. et al. (2016) ‘Assessing ecosystem services

from multifunctional trees in pastures using Bayesian

belief networks’, Ecosystem Services, 18, pp. 165–174.

doi: 10.1016/j.ecoser.2016.03.002.

Barton, D. N. et al. (2016) ‘Diagnosing agrosilvopastoral

practices using Bayesian networks’, Agroforestry Sys-

tems, 91(2), pp. 325–334. doi: 10.1007/s10457-016-

9931-1.

Bhowmik, C. et al. (2017) ‘Optimal green energy planning

for sustainable development : A review’, Renewable

and Sustainable Energy Reviews. Elsevier Ltd, 71

(December 2016), pp. 796–813. doi: 10. 1016/ j.rser.

2016. 12.105.

Borunda, M. et al. (2016) ‘Bayesian networks in renewable

energy systems : A bibliographical survey’, Renewable

and Sustainable Energy Reviews. Elsevier, 62, pp. 32–

45. doi: 10.1016/j.rser.2016. 04.030.

Campos, L. M. De and Castellano, J. G. (2007) ‘Bayesian

network learning algorithms using structural re-

strictions’, 45, pp. 233–254. doi: 10.1016/ j.ijar. 2006.

06.009.

Cinar, D. and Kayakutlu, G. (2010) ‘Knowledge-Based

Systems Scenario analysis using Bayesian networks : A

case study in energy sector’, Knowledge-Based Sys-

tems. Elsevier B.V., 23(3), pp. 267–276. doi: 10.1016/

j.knosys. 2010.01.009.

Conrady, S., Jouffe, L. and Elwert, F. (2014) ‘Causality for

Policy Assessment and Impact Analysis Directed Acy-

clic Graphs and Bayesian Networks for Causal’, (No-

vember). doi: 10.13140/2.1. 2350. 1763.

Corani, G. and Scanagatta, M. (2016) ‘Environmental Mo-

delling & Software Air pollution prediction via multi-

label classification’, 80. doi: 10.1016/ j.envsoft.2016.

02.030.

Franco, C. et al. (2016) ‘Environmental Modelling & Soft-

ware A Bayesian Belief Network to assess rate of

changes in coral reef ecosystems’, Environmental Mo-

delling and Software. Elsevier Ltd, 80, pp. 132–142.

doi: 10.1016/j.envsoft.2016.02.029.

Gambelli, D. et al. (2017) ‘Third generation algae biofuels

in Italy by 2030

: A scenario analysis using Bayesian

networks’, 103(April 2016), pp. 165–178. doi:

10.1016/j.enpol.2017.01.013.

Gerstenberger, M. C. et al. (2015) ‘Bi-directional risk as-

sessment in carbon capture and storage with Bayesian

Networks’, International Journal of Greenhouse Gas

Control. Elsevier Ltd, 35, pp. 150–159. doi:

10.1016/j.ijggc. 2015. 01.010.

Gonzalez-Redin, J. et al. (2016) ‘Spatial Bayesian belief

networks as a planning decision tool for mapping eco-

system services trade-offs on forested landscapes’, En-

vironmental Research, 144, pp. 15-26. doi:

10.1016/j.envres.2015.11.009

Jha, S. K. (2019) ‘A comprehensive search for expert clas-

sification methods in disease diagnosis and prediction’,

(April 2018). doi: 10.1111 /exsy. 12343.

Kabir, S. and Papadopoulos, Y. (2019) ‘Applications of

Bayesian networks and Petri nets in safety, reliability,

and risk assessments : A review’, Safety Science. Else-

vier, 115(April 2018), pp. 154–175. doi: 10.1016/

j.ssci.2019.02.009.

Kameshwar, S. et al. (2019) ‘Probabilistic decision-support

framework for community resilience: Incorporating

multi-hazards, infrastructure interdependencies, and re-

silience goals in a Bayesian network’, Reliability Engi-

neering & System Safety, 191, 106568. doi:

10.1016/j.ress.2019.106568

Keshtkar, A. R. et al. (2013) ‘Application of Bayesian net-

works for sustainability assessment in catchment mod-

eling and management (Case study : The Hablehrood

river catchment)’, Ecological Modelling. Elsevier B.V.,

268, pp. 48–54. doi: 10.1016/ j.ecolmodel.2013.08.003.

Kim, J. et al. (2018) ‘Sustainable Technology Analysis of

Artificial Intelligence Using Bayesian and Social Net-

work Models’. doi: 10.3390/ su10010115.

Kumar, A. et al. (2010) ‘Renewable energy in India : Cur-

rent status and future potentials’, (October). doi:

10.1016/j.rser.2010.04.003.

Landuyt, D. et al. (2015) ‘Environmental Modelling &

Software a GIS plug-in for Bayesian belief networks :

Towards a transparent software framework to assess

and visualise uncertainties in ecosystem service map-

ping’, Environmental Modelling and Software. Elsevier

Ltd, 71, pp. 30–38. doi: 10.1016/ j.envsoft. 2015. 05.

002.

Liu, Z. and Callies, U. (2019) ‘Implications of using chem-

ical dispersants to combat oil spills in the German Bight

e Depiction by means of a Bayesian network *’, Envi-

ronmental Pollution. Elsevier Ltd, 248, pp. 609–620.

doi: 10.1016/j.envpol.2019. 02.063.

Marcos, D. et al. (2018) ‘Ecotoxicology and Environmental

Safety Assessing mercury pollution in Amazon River

tributaries using a Bayesian Network approach’, Eco-

toxicology and Environmental Safety. Elsevier Inc.,

166(June), pp. 354–358. doi: 10.1016/j.ecoenv. 2018.

09.099.

Martos, A. et al. (2016) ‘Towards successful environmental

performance of sustainable cities : Intervening sectors.

A review’, Renewable and Sustainable Energy Re-

views. Elsevier, 57, pp. 479–495. doi: 10.1016/ j.rser.

2015.12.095.

McLaughlin, D., and Reckhow, K. (2017) ‘A Bayesian net-

work assessment of macroinvertebrate responses to nu-

trients and other factors in streams of the Eastern Corn

Belt Plains, Ohio, USA’, Ecological Modelling, 345,

pp. 21-29. doi: 10.1016/j.ecolmodel.2016.12.004

Bayesian Networks based Policy Making in the Renewable Energy Sector

461

Montgomery, D. C. and Runger, G. C. (2014) Applied Sta-

tistics and Probability for Engineers. 6th edn. John

Wiley & Sons Inc.

Mu, Y., Liu, X. and Wang, L. (2018) ‘A Pearson's correla-

tion coefficient based decision tree and its para-llel im-

plementation’, Information Sciences. Elsevier Inc., 435,

pp. 40–58. doi: 10.1016/ j.ins.2017.12.059.

Mukashema, A., Veldkamp, A. and Vrieling, A. (2014) ‘In-

ternational Journal of Applied Earth Observation and

Geoinformation Automated high resolution mapping of

coffee in Rwanda using an expert Bayesian network’,

International Journal of Applied Earth Observations

and Geoinformation. Elsevier B.V., 33, pp. 331–340.

doi: 10.1016/ j.jag.2014.05.005.

Organisation for Economic Cooperation and Development.

(n.d). Retrieved from https:// stats. oecd .org/

Orun, A. et al. (2018) ‘Use of Bayesian inference method

to model vehicular air pollution in local urban areas’,

Transportation Research Part D. Elsevier, 63(May),

pp. 236–243. doi: 10.1016/ j.trd.2018.05.009.

Perreault, L. and Sheppard, J. (2019) ‘International Journal

of Approximate Reasoning Compact structures for con-

tinuous time Bayesian networks’, International Journal

of Approximate Reasoning. Elsevier Inc., 109, pp. 19–

41. doi: 10.1016/ j.ijar. 2019.03.005.

Pitchforth, J. and Mengersen, K. (2013) ‘Expert Systems

with Applications A proposed validation framework for

expert elicited Bayesian Networks’, Expert Systems

with Applications. Elsevier Ltd, 40(1), pp. 162–167.

doi: 10.1016/j.eswa. 2012. 07.026.

Res, C. et al. (2009) ‘Applying Bayesian modelling to as-

sess climate change effects on biofuel production’, 40,

pp. 249–260. doi: 10.3354/ cr00833.

Ropero, R. F., Renooij, S. and Gaag, L. C. Van Der (2018)

‘Discretizing environmental data for learning Bayesian-

network classifiers’, Ecological Modelling. Elsevier

B.V., 368, pp. 391–403. doi: 10.1016/ j.ecolmodel.

2017.12.015.

Tang, C. et al. (2016a) ‘Risk analysis of emergent water

pollution accidents based on a Bayesian Network’,

Journal of Environmental Management. Elsevier Ltd,

165, pp. 199–205. doi: 10.1016/ j.jenvman. 2015.

09.024.

Tang, C. et al. (2016b) ‘Risk forecasting of pollution acci-

dents based on an integrated Bayesian Network and wa-

ter quality model for the South to North Water Transfer

Project’, Ecological Engineering. Elsevier B.V., 96, pp.

109–116. doi: 10.1016/ j.ecoleng.2015.11.024.

Tselykh, A., Tselykh, L. and Barkovskii, S. (2018)

‘Knowledge discovery using maximization of the

spread of influence in an expert system’, (April 2017),

pp. 1–19. doi: 10.1111/exsy. 12312.

Washington, A. et al. (2019) ‘Adoption of a Bayesian Be-

lief Network for the System Safety Assessment of Re-

motely Piloted Aircraft Systems’, Safety Science, 118,

pp. 654-673. doi: 10.1016/j.ssci.2019.04.040

Zarikas, V. et al. (2013) ‘Statistical survey of elevator acci-

dents in Greece’,

Safety Science. Elsevier Ltd, 59, pp.

93–103. doi: 10.1016 /j.ssci .2013. 05.001.

Zarikas, V. (2014) ‘Modeling decisions under uncertainty

in adaptive user interfaces’, (May 2007). doi: 10.1007/

s10209-007-0072-1.

Zarikas, V. et al. (2018) ‘Medical Decision Support Tool

from a Fuzzy-Rules Driven Bayesian Network’,

2(Icaart), pp. 539–549.

Zarikas, V., Papageorgiou, E. and Regner, P. (2014) ‘Ar-

ticle approach for medical decision support’, 0(0). doi:

10.1111/exsy.12089.

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

462