基于改进粒子群算法优化SVM的变压器故障诊断  

Transformer fault diagnosis based on improved particle swarm optimization SVM

作　　者：石延辉[1] 杨洋 廖毅[1] 洪乐洲[1] 廖名洋 逄帅 SHI Yanhui;YANG Yang;LIAO Yi;HONG Lezhou;LIAO Mingyang;PANG Shuai(Guangzhou Bureau,China Southern Power Grid Extra-High Voltage Transmission Company,Guangzhou510663,China;School of Electrical Engineering and Automation,Wuhan University,Wuhan 430072,China)

机构地区：[1]中国南方电网超高压输电公司广州局,广东广州510663 [2]武汉大学电气与自动化学院,湖北武汉430072

出　　处：《武汉大学学报（工学版）》2023年第10期1238-1244,共7页Engineering Journal of Wuhan University

基　　金：南方电网有限公司科技项目(编号:CGYKJXM20180229)。

摘　　要：为提高变压器的运行可靠性,提出一种基于核主成分分析(kernel principal component analysis,KPCA)和改进粒子群优化(improved particle swarm optimization,IPSO)算法优化支持向量机(support vector machines, SVM)参数的变压器故障诊断方法。首先建立变压器状态评估的基础指标体系,并采用KPCA对基础指标进行关键参量的提取,得到降维后的评价指标,然后将其输入SVM训练诊断模型,最后采用IPSO对SVM进行参数优化。仿真结果表明,该方法对变压器故障诊断的准确率达到95.42%,相比KPCA-PSO-SVM以及KPCA-SVM诊断模型,准确率有显著提升。In order to improve the reliability of transformer operation,this paper proposes a transformer fault diagnosis method based on kernel principal component analysis(KPCA)and improved particle swarm optimization(IPSO)to optimize support vector machines(SVM)parameters.This method first establishes the basic index system for transformer condition evaluation,then uses KPCA to extract the key parameters of the basic index to obtain the dimensionality-reduced evaluation index,and then inputs it into the SVM training diagnostic model,and finally uses IPSO to optimize the parameters of the SVM.The simulation results show that the accuracy of this method for transformer fault diagnosis reaches 95.42%,which is significantly improved compared to the KPCA-PSO-SVM and KPCA-SVM diagnosis models.

关 键 词：变压器 故障诊断 核主成分分析 粒子群算法 支持向量机 

分 类 号：TM411[电气工程—电器] TP183[自动化与计算机技术—控制理论与控制工程]