A new approximate belief rule base expert system for complex system modelling

Expert knowledge is the foundation of the interpretability of belief rule base (BRB) expert system. However, the rule explosion problem and weak extendability of BRB limit the utilization of expert knowledge. To solve this problem, a new approximate belief rule with single attributes is proposed, with which a new expert system named as ABRB is constructed. In the new rule, the correlation among attributes is discounted by the independency factor. To illustrate the similar modelling ability of ABRB and BRB, the universal approximation ability of ABRB is proved theoretically. In the proposed ABRB, the key components, such as attributes, referential values, and the frame of discernment, can be extended to guarantee its effectiveness in the long-term practice. A case study of the Lithium-ion power battery is conducted to verify the effectiveness of the proposed model. • The feature correlation in proposed approximate belief rules is handled by discounted weights. • Expert knowledge can be introduced into the proposed ABRB more easily. • The key components of the proposed ABRB can be well extended.