As a widely used rule-based system, one of the strongest advantages of belief rule-based (BRB) systems is good interpretability. Currently, most studies on BRB modeling focus on improving accuracy only but overlook interpretability. In this study, the definition of interpretability in a BRB system is first discussed, and the evaluation indicators are designed for measurement. On this basis, a single-objective optimization framework for the interpretability and accuracy trade-off in the modeling of BRB systems is proposed. The differential evolution (DE) algorithm is utilized as the optimization tool, and a new encoding method of chromosomes is proposed to realize the coding of rules with different structures. Two case studies are tested to validate the efficiency and feasibility of the proposed optimization framework for BRB modeling in both prediction and classification. The results show that the proposed optimization framework can ensure the interpretability and accuracy of BRB systems simultaneously, which is conducive to promoting the application of BRB systems in fields that require high interpretability and accuracy.

作为一种广泛使用的基于规则的系统，基于信念规则（BRB）系统的最大优势之一是良好的可解释性。目前，大多数关于 BRB 建模的研究只关注提高准确性，而忽视了可解释性。在本研究中，首先讨论了 BRB 系统中可解释性的定义，并设计了评估指标进行测量。在此基础上，提出了一种用于BRB系统建模中可解释性和准确性权衡的单目标优化框架。以差分进化（DE）算法为优化工具，提出一种新的染色体编码方法，实现不同结构规则的编码。测试了两个案例研究，以验证所提出的 BRB 建模优化框架在预测和分类方面的效率和可行性。结果表明，所提出的优化框架可以同时保证BRB系统的可解释性和准确性，有利于促进BRB系统在对可解释性和准确性要求较高的领域的应用。