基于平衡决策树相关向量机的模拟电路多类故障诊断方法

Multi-Class Fault Diagnosis Approach Based on Balanced Decision Tree Relevance Vector Machine

针对模拟电路的多类故障诊断问题,提出一种基于平衡决策树(Balanced Decision Tree,BDT)相关向量机(Relevance Vector Machine,RVM)的故障诊断方法;综合考虑类内紧密性和类间分散性,建立了一种类的可分性度量方法,并根据不同类划分的可分度大小优化确定BDT结构,通过在BDT各个决策节点构造RVM二类分类器实现RVM的多类分类;故障诊断实验结果表明:该方法在训练效率、诊断效率和诊断精度等方面的综合性能优于已有的RVM多类故障诊断方法。

Aiming at the problem of multi-class fault diagnosis for analog circuits,an approach based on balanced decision tree（BDT） relevance vector machine（RVM）was proposed.The class tightness and inter-class dispersion were considered synthetically to establish the separability measure.The distribution of BDT nodes was optimized according to the established separability measure,and then the structure of BDT was determined.RVMs were used to make binary classification at every node,which made multi-class classification realized.The experimental results show that the proposed approach has higher training efficiency,diagnostic efficiency and accuracy than the existing RVM approaches for multi-fault diagnosis.