=Paper=
{{Paper
|id=Vol-2834/Paper35.pdf
|storemode=property
|title=Bayesian Belief Networks as a Tool for Modeling Haz-ardous Natural Processes
|pdfUrl=https://ceur-ws.org/Vol-2834/Paper35.pdf
|volume=Vol-2834
|authors=Viktoriya N. Taran
}}
==Bayesian Belief Networks as a Tool for Modeling Haz-ardous Natural Processes==
https://ceur-ws.org/Vol-2834/Paper35.pdf
415
Bayesian Belief Networks as a Tool for Modeling Hazard-
ous Natural Processes*
Viktoriya N. Taran1[0000-0002-9124-0178]
1 V.I. Vernadsky Crimean Federal University, Simferopol, Russia
victoriya_yalta@ukr.net
Abstract. The article discusses dangerous natural processes that have disastrous
consequences, examples of such processes are given. The most significant factors
that affect the activation of complex hazardous natural processes with cata-
strophic consequences are analyzed and highlighted. A model was built to predict
the catastrophic consequences of dangerous natural processes using the Bayesian
belief network. The vertices of the Bayesian network and the levels on which
these vertices are located are determined. To determine the possible transitions
in the network, an expert assessment of the values of the selected indicators was
carried out. Based on expert assessments, the Bayesian network was trained. The
“Investments” factor was proposed as a controlling operation on the network. The
modeling and forecasting of possible scenarios for the development of complex
natural processes and their catastrophic consequences were carried out. As a re-
sult of the study, a decision support system was developed for the proposed model
of complex hazardous natural processes, the interface of the forecasting system,
modeling and decision support was designed and programmed.
Keywords: Bayesian Belief Network, Complex Natural Systems and Processes,
Modeling, Forecasting, Training Bayesian Network, Forecasting of Funds for
the Rebuilding of Business Objects
1 Introduction
Various natural processes taking place on the North Black Sea Coast have a high level
of unpredictability and uncertainty, which makes it difficult to study, model and fore-
cast them. The presence of a stochastic component in the course of complex natural
processes that have catastrophic consequences for the environment, the environment,
and man does not allow us to obtain reliable results using traditional methods and mod-
els.
The opportunity to foresee the activation of complex natural processes on the North
Black Sea Coast with catastrophic consequences allows by minimizing these conse-
quences by carrying out special protective measures, thereby protecting human life,
* Copyright 2021 for this paper by its authors. Use permitted under Creative Commons License
Attribution 4.0 International (CC BY 4.0).
416
which will also allow using the unique coast recreation territory for the rest of tourists.
Earthquakes, landslides, mudslides, natural fires, hurricanes, processing of the coastal
zone of the sea, snow avalanches, extreme air temperatures [1] cause significant dam-
age to human economic activity, do not allow the recreational territories to be fully and
qualitatively used.
The northern Black Sea coast has a unique geographical location and a special med-
ical climate and is also a recreational and well-known tourist region. Also, on the coast,
there is a dense network of roads, and the population density is more than 2 times higher
than the average for the Northern Black Sea region. Consequently, complex natural
processes with catastrophic consequences pose a significant threat, destroying roads,
buildings, and structures, causing great harm to their operation and threatening in some
cases human life.
Constant changes in geological, climatic, ecological and technogenic processes neg-
atively affect the economic objects and complex engineering structures, on which hu-
man life and health depend [3]. Complex hazardous natural processes that have cata-
strophic consequences, are complex random in nature, depend on many factors, contain
a large proportion of uncertainty, and therefore it is very difficult to predict.
The task of studying the dynamics of complex hazardous natural processes and their
modeling, as well as estimating the amount of money needed for different areas to pre-
vent or cover material damage from the destruction of economic facilities, is a difficult
task. It is due to the need to develop new approaches, methods, models, algorithms,
intelligent systems that can increase the efficiency of the process of developing and
making management decisions in terms of risk and various types of economic and en-
vironmental uncertainties.
Studies on modeling and forecasting complex hazardous natural processes in the
Northern Black Sea region often have a narrow focus on cartographic research and ge-
ologist's expert assessment, which is based on observations, monitoring and shows the
limited use of modern information technologies used in intelligent decision-making
systems, and also indicates the absence systematic approach to solving the problem.
The use of Bayesian belief networks will allow us to solve the task by obtaining prob-
abilistic estimates of possible variants of the course of the events and processes under
consideration.
A model based on the Bayesian belief network allows you to combine both statistical
data and expert assumptions about the nature of behavior and the relationships between
elements [2,3]. Bayesian networks are one of the representations of knowledge bases
with uncertainty [4]
The purpose of this study is to build a Bayesian belief network form the basis of an
analysis of factors and indicators as a model, which characterizing the occurrence of
complex natural phenomena and processes on the North Black Sea Coast.
This information system will contain modules of expert assessments and recommen-
dations on preventing or forestall and minimizing losses from the catastrophic conse-
quences of complex natural processes.
417
2 Construction of the Bayesian belief network for modeling and
forecasting complex natural processes
2.1 The Bayesian Belief Network
Bayesian networks (BNs), also known as Bayesian belief networks or Bayes nets, are
a kind of probabilistic graphical model that has become very popular to practitioners
mainly due to the powerful probability theory involved, which makes them able to deal
with a wide range of problems. BNs have barely been used for Environmental Science
and their potential is, as yet, largely unexploited [5].
Bayesian networks (BNs) are widely used as one of the most effective models in
bioinformatics, artificial intelligence, text analysis, medical diagnosis, etc. Learning the
structure of BNs from data can be viewed as an optimization problem [6].
Because a Bayesian network is a complete model for the variables and their relation-
ships, it can be used to answer probabilistic queries about them. For example, the net-
work can be used to find out updated knowledge of the state of a subset of variables
when other variables (the evidence variables) are observed. This process of computing
the posterior distribution of variables given evidence is called probabilistic inference.
A Bayesian network can thus be considered a mechanism for automatically applying
Bayes’ theorem to complex problems. In the application of Bayesian networks, most of
the work is related to probabilistic inferences. Any variable updating in any node of
Bayesian networks might result in the evidence propagation across the Bayesian net-
works [7].
Bayesian networks are one of the representations of knowledge bases with uncer-
tainty [4]. The Bayesian network was mainly employed as a statistical scheme for prob-
abilistic forecasting that can represent the cause-effect relationships between the varia-
bles [8]. Bayesian networks have become a commonly used tool for inferring the struc-
ture of gene regulatory networks from gene expression data. In this framework, genes
are mapping to nodes of a graph, and Bayesian techniques are used to determine a set
of edges that best explain the data, that is, to infer the underlying structure of the net-
work [9].
The Bayesian Belief Network is an oriented acyclic graph whose vertices are discrete
random variables X with a finite number of states, and U-edges are cause-effect rela-
tions between them, characterized by a table of unconditional probabilities of transi-
tions from one state to another under the influence of perturbations. So, the Bayesian
network is a pair (G, P), where G = is a directed acyclic graph on a finite set X
whose elements are connected by a set of oriented edges U, and P is the set of condi-
tional probability distributions [10,11].
Under uncertainty, the basis for decision-making with the help of the Bayesian belief
network is the calculation of the probabilities of transition strategies from one to the
other state of the system [12]. Uncertainty is uncovered by calculating the probabilities
of vertex states based on available information about the value of other vertices of the
network, thanks to this message, the system proceeds to the next state [13].
418
2.2 Building the Bayesian Belief Network
As the factors having a strong influence on the course of complex natural processes
of the North Black Sea Coast, accompanied by catastrophic consequences, the follow-
ing should be noted:
· precipitation (it makes sense to consider the amount of precipitation for the hydro-
geological year, ie from September of the previous year to August of the current year);
· solar activity (catastrophic natural phenomena directly depend on solar activity and
its 11-year cycle [13]);
· seismic activity;
· level of landslide activity during the previous period; invested funds in strengthen-
ing slopes, structures and roads [1,12].
It also makes sense to consider the intermediate and resulting indicators that will be
located on the following levels of the Bayes confidence network:
· time intervals during which a catastrophic event is possible;
· the level of natural risk in probabilistic interpretation, i.e. probability of the occur-
rence of a particular natural process or phenomenon;
· the final level of monetary investment to eliminate the catastrophic consequences
of events that have already occurred and phenomena [1,12].
So, we construct a Bayesian belief network for modeling and forecasting, based on
the factors considered, as well as for predicting the level of possible consequences,
which will make it possible to formulate recommendations on investing funds for pro-
tective and fortification measures.
We build a Bayesian belief network on the considered factors for modeling and fore-
casting, as well as for predicting the level of possible consequences. This will make it
possible to formulate recommendations on investing money for protective and fortifi-
cation measures.
In constructing the Bayesian belief network, we arrange the primary (initial) factors
on the upper (first) level, i.e. we will place precipitation, solar activity, seismic activity,
landslide activity and invested funds at the top level of the network. Each of these fac-
tors has characterized by an unconditional probability, obtained from statistical data as
a result of long-term observations. Define the qualitative values of the factors in the
form: Small, Average and Catastrophic (catastrophically many). The exception is the
top "Invested funds", for which we choose the values: "In_Full", "Average", "Small".
Thus, the Bayesian belief network for modeling the catastrophic natural processes
of the North Black Sea Coast has presented (see Ошибка! Источник ссылки не
найден.) (using the shareware program Netica).
At the second level, we place one vertex: "Natural Risks", which will take on values
similar to natural factors. At the third level - "Time limits", which will take values: two
days, a week, a month and will not happen at all. As the resultant indicator, take the
"Final_Amount_of_Money", i.e. the total amount spent on preventing and eliminating
catastrophic consequences.
419
Precipitations Solar_Activity Seismic_Activity LandSlade
Small 33.0 Small 33.0 Small 33.0 Small 33.0
Average 34.0 Average 34.0 Average 34.0 Average 34.0
Catastrophic 33.0 Catastrophic 33.0 Catastrophic 33.0 Catastrophic 33.0
Invested_Funds
In Full 33.0
Average 34.0
Small 33.0
Natural_Risks
Small 55.6
Average 26.2
Catastrophic 18.1
Time_terms
Two Days 15.0
Week 13.6
Month 27.2
Do Not 44.3
Final_Amount_of_Money
Small 50.9
Average 29.6
Catastrophic 19.5
Fig. 1. The initial state of the Bayes confidence network for modeling complex natural pro-
cesses that have catastrophic consequences on the southern coast of the Crimea
To train the network, fill in the tables of conditional probabilities with the help of ex-
perts for each vertex, except for the vertices of the first level. For example, the top
"Natural Risks" has 35 rows of conditional probabilities (for each top-level factor) (see
Ошибка! Источник ссылки не найден.).
Fig. 2. Training Bayesian network (conditional probabilities for the top "Natural_Risks")
420
2.3 Simulation of complex natural processes using the constructed Bayesian
belief network
Based on the developed Bayesian network of beliefs, we will build a forecast of the
possible development of complex natural processes on the North Black Sea Coast. First,
we consider catastrophic changes in only one factor, i.e. the selected factor takes the
value "Catastrophic" with a probability of 100%, and the rest - the values corresponding
to the average statistical observations (The following Table 1).
Table 1. Results of Modeling with the Bayesian Belief Network.
Factor values as Natural_Risks
Catastrophic
Small Average Catastrophic
(Catastrophic=100%)
Precipitation 32.4% 29.7% 37.9%
Solar_Activity 42.2% 25.6% 32.1%
Seismic_Activity 50.2% 27.8% 21.9%
Landslide_Activity 55.5% 26.3% 18.2%
All factors 4.0% 6.0% 90.0%
Factor values as Time_terms
Catastrophic
2_Day Week Month Do_Not
(Catastrophic=100%)
Precipitation 25.1% 17.4% 23.8% 33.7%
Solar_Activity 26.1% 16.0% 24.8% 37.6%
Seismic_Activity 17.0% 14.4% 26.6% 42.0%
Landslide_Activity 15.0% 13.6% 27.1% 44.3%
All factors 64.6% 20.0% 9.4% 6.0%
Factor values as Final_Amount_of_Money
Catastrophic
Small Average Catastrophic
(Catastrophic=100%)
Precipitation 39.4% 31.3% 29.3%
Solar_Activity 43.6% 30.4% 26.1%
Seismic_Activity 48.5% 30.1% 21.4%
Landslide_Activity 50.9% 29.6% 19.5%
All factors 10.5% 19.5% 70.0%
We will determine the consequences of these changes. In the next step, we consider
what happens to the vertex values that are at the lower levels, if all factors simultane-
ously take on catastrophic values, and the top “Invested Funds” - “Small”.
As can be seen from the table, catastrophic changes of only one factor do not always
lead to catastrophic destruction, which requires an immediate investment of funds to
421
eliminate them. But at the same time, the probability of activation of any complex nat-
ural processes increases substantially, and as a result, the level of total cash flows in-
creases.
If we assume that at the same time, the average values for all the top-level factors
are significantly exceeded, the resulting indicators make a sharp jump in the direction
of not only the activation of processes and investments but at the same time, there is
practically no time left for preventive and strengthening measures. The result of the
simulation is shown (see Ошибка! Источник ссылки не найден.).
Precipitations Solar_Activity Seismic_Activity LandSlade
Small 0 Small 0 Small 0 Small 0
Average 0 Average 0 Average 0 Average 0
Catastrophic 100 Catastrophic 100 Catastrophic 100 Catastrophic 100
Invested_Funds
In Full 0
Average 0
Small 100
Natural_Risks
Small 4.00
Average 6.00
Catastrophic 90.0
Time_terms
Two Days 64.6
Week 20.0
Month 9.40
Do Not 6.00
Final_Amount_of_Money
Small 10.5
Average 19.5
Catastrophic 69.9
Fig. 3. Simulation of complex natural processes.
Based on the considered options for changing factors, it can be concluded that preven-
tive measures about strengthening measures will help reduce overall costs by reducing
the catastrophic consequences of the devastation resulting from the activation of com-
plex natural processes
3 Modeling with the FCLSSCC decision support system based
on the Bayesian Belief Network
To simulate catastrophic natural processes, including landslide processes that cause the
greatest damage to human economic activities and threaten his life, the decision support
system FCLSSCC (see Ошибка! Источник ссылки не найден.) was developed,
which builds a long-term forecast based on classical statistical methods based on re-
gression analysis, integration of analogs, autoregression, methods of taking into ac-
count group arguments, etc., as well as a short-term forecast based on a Bayesian net-
work of beliefs. In this case, the system also allows you to make a short-term forecast
based only on one of the selected factors listed. If, for example, there was a sharp cata-
strophic increase in this factor with the other values remaining unchanged.
422
This decision support system allows you to simulate the resulting values in the form
of probabilities of the onset of alternative scenarios for the development of the process
and its catastrophic consequences. For example, we choose the factor of precipitation
and its value - “not active, but more than three consecutive days”, seismic activity
(tremors) - “strong and short”, solar activity “high”, Landslides of the previous period
- “a lot”, and previous investments in reinforcing activities were made small.
Fig. 4. Modeling with the FCLSSCC decision support system based on the Bayesian confi-
dence network when introducing the values of the first-level factors.
If the choice of parameter values is made, then you can click on the corresponding
transition and get the result, which will be presented in the form of the probability of
occurrence of each of the proposed scenarios for the selected criteria or result indica-
tors. We calculate the estimates of three criteria: the probability of manifestation of
natural risks (complex natural processes - landslides with catastrophic consequences),
the time during which these processes will be and the total amount of expenditures are
possible. The estimates obtained are shown (see Ошибка! Источник ссылки не
найден.).
Thus, we have the most likely outcome of "Activation of natural phenomena"
Threatening = 55%, and Catastrophic = 9%; "Times_Risk" show “Two_Days” = 41%
activation will occur; "Amount_Invested" Catastrophic = 33% or Average = 37%. The
presented calculations show the need to invest advance funds in reinforcement
measures.
423
a)
b)
c)
Fig. 5. Results of modeling using the Bayesian belief network: a) calculation of the possible
level of active landslides; b) the time interval for the expected activation of landslides; c) the
probability of total investing funds to eliminate catastrophic consequences
4 Conclusions
So, to take into account the uncertainty and randomly manifested risks of activation of
complex natural processes, it is possible to use Bayes confidence networks in modeling
and forecasting catastrophic consequences of natural processes [12].
The constructed Bayes confidence network and its training by filling in the tables of
conditional probabilities by experts and the introduction of numerical values of factors
for determining the unconditional (a priori) probabilities of the factors under consider-
ation made it possible to obtain a forecast concerning the occurrence of catastrophic
natural processes. The analysis showed that the use of the Bayesian belief network in
modeling under conditions of uncertainty is becoming increasingly popular and meets
the objectives. The Bayesian network is also useful when predicting processes of vari-
ous origins, including complex hazardous natural processes, which allows one to take
424
into account the structural and statistical uncertainties of the phenomena studied [1].
Thus, when using the decision support system and creating a single technical bureau
for information processing and forecasting, it is possible to envisage the number and
nature (i.e. consequences) of the activation of complex hazardous natural processes; the
time during which they can occur, as well as optimize the costs of implementing
measures to overcome the catastrophic consequences of exogenous processes taking
place in the mountainous regions of the Northern Black Sea Region [14].
References
1. Taran, V.N.: Modeling of natural catastrophic processes of the Southern Coast of Crimea
with the help of the Bayes network. Auditorium. No. 3 (11), 47-54 (2016).
2. Kharitonov, N.A., Zolotin, A.A., Tulupyev, A.L.: Software Implementation of Algorithms
for Maintaining Consistency in Algebraic Bayesian Networks. In: 2017 IEEE Proceedings of
the XXI International Conference on Soft Computing and Measurements SCM`2018, 19-22 (2018)
3. Goriainov, S.V., Kuznetsov, A.Yu., Tushkanov, E.V., Kuznetsova, O.V., Romanova, E.B.:
Analysis of Bayesian mathematical systems for pattern recognition for identifying objects
on hyperspectral images. In: 2017 IEEE Proceedings of the XXI International Conference
on Soft Computing and Measurements SCM`2018, 69-72 (2018).
4. Zolotin, A.A., Tulupyev, A.L.: Matrix-vector algorithms for global a posteriori output in
algebraic Bayesian networks. In: 2017 IEEE Proceedings of the XXI International Confer-
ence on Soft Computing and Measurements SCM`2018, 45-48 (2018).
5. Aguilera, P.A., Fernández, R., Fernández, A., Rumí, R., Salmerón, A.: Bayesian Networks
in Environmental Modelling. Environmental Modelling & Software. No 12. 1376-1388.
(2011). DOI: 10.1016/j.envsoft.2011.06.004
6. Khanteymoori, A.R., Olyaee, M.-H., Abbaszadeh, O., Valian, M.: A novel method for
Bayesian networks structure learning based on Breeding Swarm algorithm. Methodologies
and Application, Volume 22, Issue 9, 3049–3060 (2018).
7. Jianguo Ding Probabilistic Inferences in Bayesian Networks URL:
https://arxiv.org/pdf/1011.0935.pdf
8. Shin Ji Yae, Ajmal Muhammad, Yoo Jiyoung, Kim Tae-Woong.: A Bayesian Network-
Based Probabilistic Framework for Drought Forecasting and Outlook. Advances in Meteor-
ology. Volume 2016, (2016). Article ID 9472605, http://dx.doi.org/10.1155/2016/9472605
9. Bauer, S., Robinson, P.: Bayesian Networks for Modeling and Inferring Gene Regulatory
Networks. Handbook of Research on Computational Methodologies in Gene Regulatory
Networks. (2010). DOI: 10.4018/978-1-60566-685-3.ch003.
10. Terentyev, A.N., Bidyuk, P.I.: A method of probabilistic inference in Bayesian networks
based on training data. Cybernetics and Systems Analysis, No. 3. 93-99 (2007).
11. Bidyuk, P.I., Klimenko, O.M., Shekhter, D.V.: Principles of construction and application of
the Bayesian network. Information technologies, systems analysis and management. No 5.
14-25. (2005).
12. Taran, V.N.: Bayesian networks for modeling complex systems In: 2017 IEEE II Interna-
tional Conference on Control in Technical Systems (CTS), pp. 240 - 243. (2017) DOI:
10.1109/CTSYS.2017.8109535
13. Neapolitan, R.E.: Learning Bayesian Networks. Pearson Prentice Hall. (2003).
14. Taran, V.N.: Modeling complex (hazardous) natural processes using the Bayesian trust net-
work. Caspian journal: management and high technology, No 2 (46). 90-100 (2019).
DOI: 10.21672/2074-1707.2019.46.2.090-100