Application of Bayesian Networks in Energetics
Stanislav Chren
Faculty of Informatics
Masaryk University
chren@mail.muni.cz
2. 10. 2014
(FI MU) PV226 Lasaris 2. 10. 2014 1 / 18
Motivation
Real world prediction and diagnostic analysis suffer from many
challenges
Incomplete input data
Uncertain data sources
Complex non-linear relationship between variables
(FI MU) PV226 Lasaris 2. 10. 2014 2 / 18
Outline
1 Introduction to Bayesian networks
Definition
Construction
Analysis
Pros and Cons
2 Bayesian networks in energetics
Demo
(FI MU) PV226 Lasaris 2. 10. 2014 3 / 18
Definition
Informal definition
Framework for graphically representing the logical relationships
between variables and capturing the uncertainty in the dependency
between these variables using conditional probabilities
(FI MU) PV226 Lasaris 2. 10. 2014 4 / 18
Probability tables
(FI MU) PV226 Lasaris 2. 10. 2014 5 / 18
Probability tables
(FI MU) PV226 Lasaris 2. 10. 2014 6 / 18
Probability tables
(FI MU) PV226 Lasaris 2. 10. 2014 7 / 18
Analysis I
Unconditional (a priori) node probabilities
(FI MU) PV226 Lasaris 2. 10. 2014 8 / 18
Analysis I
Unconditional (a priori) node probabilities
(FI MU) PV226 Lasaris 2. 10. 2014 9 / 18
Analysis II
Sensitivity analysis
Identification of nodes with the highest impact on the target node
(FI MU) PV226 Lasaris 2. 10. 2014 10 / 18
Analysis II
Sensitivity analysis
Identification of nodes with the highest impact on the target node
(FI MU) PV226 Lasaris 2. 10. 2014 11 / 18
Analysis III
Inference and propagation of evidence
(FI MU) PV226 Lasaris 2. 10. 2014 12 / 18
Analysis III
Inference and propagation of evidence
(FI MU) PV226 Lasaris 2. 10. 2014 13 / 18
Analysis III
Inference and propagation of evidence
(FI MU) PV226 Lasaris 2. 10. 2014 14 / 18
Analysis III
Inference and propagation of evidence
(FI MU) PV226 Lasaris 2. 10. 2014 15 / 18
Applications and Variations
Domains of BN application
Medical diagnostics
Language understanding
Weather forecasting
Legal arguments
Software reliability analysis
...
Modifications and extensions of BN
Dynamic BN
OOBN
Influence diagrams
(FI MU) PV226 Lasaris 2. 10. 2014 16 / 18
Pros and Cons
Advantages of Bayesian networks
Uncertainty handling
Complex relationships between variables
Robust model
Intuitive parameters
Variety of analysis options
Disadvantages of Bayesian networks
Support for continuous variables
Construction of topology
(FI MU) PV226 Lasaris 2. 10. 2014 17 / 18
Bayesian Networks in Energetics
DEMO
(FI MU) PV226 Lasaris 2. 10. 2014 18 / 18