Parameter estimation in models of dynamic systems is a common preliminary procedure in control, monitoring, fault detection and diagnosis. A range of tools to fulfill this task has been constantly expanded with new additions to meet the increasing demands of modern technological processes. This article proposes the use of Bayesian networks to estimate the parameters of linear dynamic systems described by a state-space model. The proposed approach is used for system identification of two simulated dynamic systems. The parameters are estimated using a learning function for Bayesian networks from Bayes Net Toolbox for Matlab. The results indicate that Bayesian networks can be used as a system identification tool that can compete with conventional methods. However, the approach requires further research aiming to increase convergence of estimates and eliminate numerical problems.
Paper
Full text
Parameter Estimation in Linear Dynamic Systems using Bayesian networks
Semantic Scholar · Engineering · 2019
Abstract
Parameter estimation in models of dynamic systems is a common preliminary procedure in control, monitoring, fault detection and diagnosis. A range of tools to fulfill this task has been constantly expanded with new additions to meet the increasing demands of modern technological processes. This article proposes the use of Bayesian networks to estimate the parameters of linear dynamic systems described by a state-space model. The proposed approach is used for system identification of two simulated dynamic systems. The parameters are estimated using a learning function for Bayesian networks from Bayes Net Toolbox for Matlab. The results indicate that Bayesian networks can be used as a system identification tool that can compete with conventional methods. However, the approach requires further research aiming to increase convergence of estimates and eliminate numerical problems.