Measurement and Optimization of Rule Consistency in a Belief Rule Base System.

As a rule-based expert system, good interpretability is one of the prominent advantages and characteristics of the belief rule base (BRB) system, and consistency is one of the important indicators of interpretability. However, the modeling of the BRB system in current research and applications usually focuses only on accuracy but ignores rule consistency. In this paper, the concept and connotation of consistency in the BRB system are first discussed. On this basis, a dynamic measurement formula of rule consistency is designed to determine whether a pair of rules is consistent and thus identify inconsistent rules. Then, an optimization process of inconsistent rules is proposed, in which expert experience or optimization algorithms can be utilized as the optimization tool. In particular, a conflict degree-based sorting method is presented to determine the optimization sequence of inconsistent rules. The proposed measurement and optimization process of rule consistency in the BRB system is validated by two cases, namely, pipeline leakage detection and remaining useful life estimation. The experimental results show that the process can effectively identify and eliminate inconsistencies in the BRB system and further ensures the effectiveness and interpretability of BRB systems in practical applications.