Composite Activation Belief Rule Base Model for Complex Systems Based on Multi-Dimensional Reduction

As a model based on expert knowledge, belief rule base (BRB) is widely used in model construction of complex systems. However, due to the complexity and uncertainty of complex systems, a large number of rules are often generated when complex models are constructed using the initial BRB. This results in rule base expansion and diminished modeling capability. Therefore, a multi-dimensional rule evaluation framework is first developed in this study. The BRB is effectively screened and optimized, resulting in a significant reduction in the total number of rules and a decrease in model complexity. Furthermore, to ensure the completeness and accuracy of the simplified BRB, a composite rule activation method is introduced for model inference, and a novel evidential reasoning (ER) calculation procedure is designed. Ultimately, a projection covariance matrix adaptation evolution strategy (P-CMA-ES)-based optimization algorithm is proposed to improve the remaining parameters of the model. The optimal model structure is selected using the Akaike information criterion (AIC). In this paper, through case study, the implementation of this method is demonstrated, and the effectiveness of the model in practical applications is validated.