Interpretable hierarchical belief rule base expert system for complex system modeling

Abstract

Belief rule base (BRB) expert system has been widely used in complex system modeling. Readability and consistency are the primary characteristics and advantages of BRB. However, these two characteristics may be lost in BRB establishment and optimization. Aiming at this problem, a novel hierarchical BRB (H-BRB) expert system is proposed. To quantify the interpretability of BRB, the definition of interpretability of BRB is proposed from the readability and the consistency. Two strategies are proposed to construct the H-BRB expert system with good readability. A new optimization method is proposed to guarantee the consistency of H-BRB, including a new optimization mechanism, a new objective function with interpretability constraints, and a modified optimization algorithm. With all the above, an H-BRB expert system with good readability and consistency can be established. A case study of the health state assessment of the aerospace relay is implemented to verify the effectiveness of the proposed method.