基于主成分分析的置信规则库结构学习方法

Structure learning for belief rule base using principal component analysis

为了解决前提属性过多时置信规则库规模的组合爆炸问题,提出了基于主成分分析的置信规则库结构学习方法.首先将前提属性转化为新的空间中的若干个主成分,再利用载荷矩阵反推出对于各主成分贡献较大的关键前提属性.以某装甲装备体系综合能力评估作为示例分析,对比研究了在单方案和多方案条件下结构学习方法与RIMER方法,验证了本文提出的结构学习方法的有效性.示例分析结果显示本文提出的结构学习方法可大幅约减置信规则库的规模,与RIMER方法的计算结果一致,并且具有较强的鲁棒性.

A structure learning approach is proposed using the principal component analysis （PCA） in order to downsize the belief rule base （BRB）. First, we select the principal components （PCs） from the attributes, and then identify the attributes with bigger contributions to the PCs using the loading matrix. A numerical case study to evaluate the comprehensive capability for an armored system of systems is analyzed under both single input and multiple inputs scenarios. The efficiency of the proposed approach is validated by the results of the case study： the BRB is significantly downsized and the consistency between the proposed approach and RIMER is preserved. Besides, the robustness of the structure learning approach is further verified.