Abstract: To the problem of undesirable inference efficiency in extended belief rule base (EBRB) with large number of rules, this paper introduces BK-tree data structure and proposes a structure optimization framework based on BK-tree. Firstly, the index with tree structure of EBRB is made by the metric distance between the belief rules in the metric space. By setting the threshold, reducing the number of search rules and activating the key rules, the reasoning efficiency of the EBRB system is improved. Finally, simulation experiments on a nonlinear function, a practical pipeline leak detection problem and multiple classification data sets are conducted to validate the performance of the optimization framework combined with EBRB system. The experimental results show the proposed method can be used to optimize the reasoning efficiency and decision accuracy of the EBRB system.

Key words: extended belief rule base (EBRB), evidential reasoning (ER), BK-tree, optimization framework

摘要： 针对扩展置信规则库（extended belief rule base, EBRB）系统在规则数较多时推理效率不理想的问题，引入BK树数据结构，提出了一种基于BK树的结构优化框架。首先根据置信规则在度量空间中彼此的距离建立EBRB的树形索引结构，然后通过设置阈值减少EBRB系统推理时搜索规则的数量，并激活关键规则，最终达到提高EBRB系统推理效率的目的。以非线性函数拟合、输油管道泄露仿真实验及分类数据集的对比实验，验证结构优化框架在EBRB系统中的有效性，实验结果表明，所提框架能够优化EBRB系统推理效率并提高决策准确性。

关键词: 扩展置信规则库（EBRB）, 证据推理（ER）, BK树, 优化框架