Reliability Prediction Based on Multi-State Dynamic Bayesian Network Structure Learning for Complex System With Time-Lagged Correlation

Abstract

Due to the complicated interaction in complex systems, correlations between modules are nonlinear and relate to time intervals. In this paper, a novel correlation analysis method based on cross correlation analysis and using two parameter matrixes to describe correlation with time-lag appropriately in multistate complex systems is proposed. Then the judgement criterion of independent, simultaneous correlation and time-lagged correlation is given. To structure the model to describe the time-lagged correlation, the Time-lagged Dynamic Bayesian Network (TDBN) is built, meanwhile the modelling and calculation method are provided. According to the parameter comparation among time-lagged correlation analysis and Maximum Likelihood Estimation (MLE), the correlation judgment is deduced. To make sure the time-lagged correlation analysis is compatible with structure learning methods used currently, the data preprocessing method, corresponding model transformation method and restoration method are proposed. Through the analysis of the data of dismantled smart electronic meters, the correlation estimated method, the TDBN model, and the data progressing method proposed for electronic system in this paper are proved. It also proposed ideas for system reliability modeling in other industrial scenarios.

Keywords

• Reliability prediction
• dynamic Bayesian network
• redundant system
• cross-correlation analysis
• train control system