分频段能量谱及马氏聚类算法在开关柜局部放电模式识别中的应用  

Application of Signals Separated Band Energy Spectrum and Mahalanobis Clustering Algorithm for Switchgear Partial Discharge Pattern Recognition

作　　者：陈攀[1,2] 姚陈果[1] 廖瑞金[1] 陈昱 米彦[1] 

机构地区：[1]重庆大学输配电装备及系统安全与新技术国家重点实验室,重庆400030 [2]国网重庆市电力公司江北供电分公司,重庆410047 [3]南方电网中山市供电公司,中山528400

出　　处：《高电压技术》2015年第10期3332-3341,共10页High Voltage Engineering

基　　金：重庆市科技攻关计划(院士专项)(CSCT2010BC3003);国家重点实验室自主研究课题(2007DA10512709104);重庆市电力公司科学技术项目(2011渝电科技18#)~~

摘　　要：针对传统超高频(UHF)检测法对系统采样频率、数据处理存储要求严苛及成本较高的缺点,融合混频降频、功率检波技术,研制出一套较经济的高压开关柜超高频在线监测系统。采用该系统对沿面放电、气隙放电、针–板放电和自由微粒放电共4种典型缺陷模型的局部放电信号进行监测,获得信号在不同频段的能量谱;并在此基础上提出了将局部放电的分频段能量作为特征向量、利用马氏距离算法进行聚类分析的高压开关柜局部放电模式识别方法。研究结果表明:不同放电类型在频域内的能量分布差异较大,相同的放电类型在频域内的能量分布差异较小;不同放电类型在能量比重三角图中表现出显著的区域聚集特性。该局部放电模式识别方法对4种典型缺陷放电类型的识别准确率达到99.125%。In order to overcome defects of traditional ultra high frequency（UHF） detection such as high cost, strict re- quirements on system sampling rate and data storage, we developed an economical UHF on-line monitoring system for HV switchgear by incorporating down-mixing with power detection technology. We used the proposed system to monitor the partial discharge （PD） signals of four typical defect models, namely surface discharge, air gap discharge, needle-plate discharge and free particle discharge, and obtained their energy spectrum in different frequency bands. Then, taking the sub-band energy of PD signals as eigenvectors, we proposed a PD pattern recognition method for HV switchgear by using Mahalanobis distance as clustering analysis tool.The results show that,the energy distributions of different PD models are more significantly different than the same PD models, and that different types of PD apparently present a feature of re- gional clustering in energy triangle plot. Moreover, by using the proposed PD recognition method, the accurate rate of PD recognition for the four typical PD defects is 99.125%.

关 键 词：局部放电 超高频法 能量分布 区域聚集 马氏距离 模式识别 

分 类 号：TM591[电气工程—电器]