采用小波变换奇异值分解方法的局部放电模式识别  

Partial Discharge Pattern Recognition Using Discrete Wavelet Transform and Singular Value Decomposition

作　　者：唐炬[1] 李伟[1] 欧阳有鹏[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

出　　处：《高电压技术》2010年第7期1686-1691,共6页High Voltage Engineering

基　　金：国家重点基础研究发展计划(973计划)(2009CB724506)~~

摘　　要：为提高电缆模拟缺陷的正确识别率,针对110kV高压XLPE电缆附件出现的绝缘缺陷以及产生的局部放电特点,设计了4种电缆中间接头内部典型的绝缘缺陷物理模型,对获取的大量甚高频局部放电信号数据,用离散小波变换的奇异值分解方法进行缺陷类型辨识。该方法首先对单次局部放电信号进行离散小波变换(DWT),得到各尺度小波分解系数,用基于Birge-Massart阈值策略提取各尺度系数的有效极大值,形成极大值的矩阵可以减少冗余数据和噪声的影响,再对小波变换值矩阵进行奇异值分解,提取奇异值作为特征量,最后采用人工神经网络分类器进行模式识别,识别结果表明该方法效果良好。Four kinds of typical artificial defect models of 110 kV cable joint and partial discharge （PD） detection system were established. PD sample data caused by typical artificial defects in XLPE cable were gotten by the very-high frequency and high speeds sampling systems. The wavelet transform-singular value decomposition method was applied to partial discharge pattern recognition. First,partial discharge data were processed by discrete wavelet transform （DWT）,the great value wavelet decomposition coefficients at all levels were extracted on the basis of Birge-Massart strategy so as to form the great value coefficient matrix and to reduce data redundancy and noise interference. Then,the singular value decomposition（SVD） on the coefficient matrix was implemented to extract effective features. Finally,the artificial neural network classfier was adopted so that high recognition probability is achieved.Results demonstrate the method is effective.

关 键 词：局部放电 小波变换 奇异值分解 模式识别 神经网络 Birge-Massart策略 

分 类 号：TM83[电气工程—高电压与绝缘技术] TM855