Abstract:  In view of the current study on the belief rule base (BRB), the prerequisite attribute, belief degree and the size of the structure of the rule base are given by experts, which will make BRB be limited in expert knowledge and could lead to the parameters of BRB inaccurate. This paper puts forward an improved differential evolution algorithm (IDE) to optimize parameters of BRB. In IDE, the mutation strategy is randomly selected to maintain the diversity of population, and a simple local search is used to balance the global and local search ability of DE. Finally, experiments are carried out with tipping paper permeability test data taken by a Chinese cigarette factory. The experimental results show that the optimization of BRB of the proposed method is simple and effective.

Key words: Belief rule base, differential evolution algorithm, parameter optimization, local search

摘要： 在当前对置信规则库(BRB)的研究中，前提属性、输出置信度和规则库大小等参数都由专家给出，会使BRB参数受专家知识的局限而不准确，提出了一种基于改进的差分进化算法(IDE)的对置信规则库的参数进行优化的方法。在IDE中，自适应选择变异策略来提高执行效率，且在变异部分加入扰动，并加入了局部搜索来平衡差分进化全局与局部搜索的能力。最后，使用中国某烟厂采取的水松纸透气度检测数据进行了实验验证。实验结果表明，所提出的方法对BRB的优化是有效的。

关键词: 置信规则库, 差分进化算法, 参数优化, 局部搜索