全样本支持矢量数据描述模拟电路故障分类-地球信息科学学报.pdfVIP

20 12 年 8 月 电 工 技 术 学 报 Vol.27 No. 8 第 27 卷第 8 期 TRAN SACTION S OF CHINA ELECTROTECHNICAL SOCIETY Aug. 20 12 全样本支持矢量数据描述模拟电路故障分类 李传亮 王友仁 罗慧 崔江 （南京航空航天大学自动化学院 南京 2 100 16） 摘要 传统支持矢量数据描述 （Support Vector Domain Description, SVDD）分类规则完 忽略非支持矢量所包含的样本信息，为此本文提出了一种 样本 SVDD 分类方法，并应用到模 拟电路故障诊断中。该方法以贝叶斯理论和分类器模糊融合思想为基础，利用核密度估计得到 的类条件概率密度和先验概率的乘积对 SVDD 相对距离进行加权。实验结果表明，与SVM 扩 展的多分类器相比，本文方法能够有效提高模拟电路故障诊断的准确率，且 样本 SVDD 分类 模型对参数变化具有较强的 健性。 关键词：模拟电路 故障诊断 多类分类 支持矢量数据描述 核密度估计 中图分类号：TP277 Analog Circuit Fault Classification Based on All Samples Support Vector Data Description L i Chuanliang Wang Youren L uo H ui Cui J iang （Nanj ing University of Aeronautics and Astronautic Nanj ing 2 100 16 China ） Abstract The information included in the non-support vectors is completely ignored for the classification rules of the classical support vector domain description(SVDD) ，so an all samples SVDD method is proposed in this paper, and it is applied to analog circuit fault diagnosis. The new method is based on Bayes theory and classifier fuzzy fusion strategy . The relative distances of this classifier are weighted by the product of prior probability value and conditional probability value, which are calculated by kernel density estimation . The simulation results show that, compared with the multi-class SVM classifiers, the introduced method improves the fault diagnosis accuracy of analog circuit . Moreover, the all samples SVDD classifier is robust against the changes of classifier parameter . Keywords ：Analog circuit, fault diagnosis, multi-class classification, support vector data description, kernel density estimation