基于有序分割的支持向量机多分类方法

SVM Multiclass Classification Method Based on Ordered Dividing

摘要:

针对模拟电路多故障分类中存在拒识区域,分类效果不理想的问题,提出基于有序分割最佳偏二叉树结构的模拟电路故障诊断方法；首先分析多分类中分割次序对结果的影响,通过样本类中心欧式距离对故障分割进行排序,以此构造偏二叉树结构；然后对采样数据进行主成份分析降维压缩,最后采用支持向量机作为分类器,对故障数据进行分类；该方法可以使样本类间距离最大,提高了支持向量机模拟电路故障分类的效率和准确度；故障诊断结果表明文章提出的诊断方法在精度提高的情况下,所需训练时间和支持向量数目大大减少,更具稀疏性。

Abstract:

In allusion to unclassifiable region existed in analog circuit fault diagnosis which causes classification unsatisfactory, a new analog circuit fault diagnosis method of optimal binary tree based on ordered dividing is presented. Firstly, the effect of dividing order on multiclass classification results is analyzed. Best diagnosable fault component set is selected based on Euclidean distance of component set center, and binary tree formation is built in the ground of the order. Secondly, sampled data dimension of analog circuit is reduced by principal component analysis. Finally, SVM is taken as the classifier, which classifies the fault data. This method can make distance of fault component set maximum, and improve the accuracy and efficiency of analog circuit fault diagnosis. The result indicates that this method need less SVs and training time with high accuracy and sparser.