Parameter learning for the belief rule base system in the residual life probability prediction of metalized film capacitor

Abstract

The Inertial Confinement Fusion (ICF) laser device consists of thousands of Metalized Film Capacitors (MFC). The Belief Rule Base (BRB) system has shown privileges in reflecting complex system dynamics. However, the BRB system requires the referenced values of each attribute to be limited. The traditional BRB learning and training approaches are no longer applicable since the referenced values of the attributes in the BRB system are pre-determined. A parameter learning approach is proposed with three strategies and each strategy is designed for one specific scenario. Strategy I (for Scenario I) is designed when only the training dataset is selectable. Strategy II (for Scenario II) is designed when new referenced values are predictable yet there is only one scale in the conclusion part. Strategy III (for Scenario III) is designed when new referenced values are predictable and there are multiple scales in the conclusion part. The Differential Evolution (DE) algorithm is used as the optimization engine to identify the key referenced values. A case is studied to validate the efficiency of the proposed parameter learning approach with multiple referenced values. The comparative results show that the parameter learning approach performs best in Scenario III.