基于最小二乘支持向量机的改进型GIS局部放电识别方法  

An Improved Approach to Recognize Partial Discharge in GIS Based on Minimum Least Square-Support Vector Machine

作　　者：王天健[1] 吴振升[1] 王晖[1] 刘栋[1] 

机构地区：[1]北京交通大学电气学院,北京市海淀区100044

出　　处：《电网技术》2011年第11期178-182,共5页Power System Technology

基　　金：铁道部科技研究开发计划项目(2007J007)~~

摘　　要：利用最小二乘支持向量机(least square-support vectormachine,LS-SVM)的方法识别气体绝缘组合电器局部放电的类型。在信号的快速分类后利用相位分布的局部放电特征谱图的特征参数作为LS-SVM识别放电类型的依据;信号快速分类处理部分主要包括信号时间-频率特性提取部分和模糊C-均值聚类2大部分,它们把信号的时间-频率点群分为由若干具有相似信号组成的信号子群。仿真实验表明该方法可有效地应对设备情况复杂的场合且有效回避传统神经网络识别受初始值影响较大、维数过高等一系列问题。The approach of minimum least square-support vector machine （LS-SVM） is used to recognize the type of partial discharge （PD） occurred in gas insulated switchgear （GIS）. After rapid classification of signals, the characteristic parameters of spectrogram of PD characteristics based on phase distribution is used as the foundation to recognize PD type by LS-SVM. The fast classification processing of signals mainly includes two parts： the extraction of time-frequency characteristic of signals and fuzzy C-means clustering, and they divide the group of time-frequency points into several signal subgroups consisting of similar signals. Results of simulation tests show that the proposed method can effectively cope with complex occasions of equipment status and can effectively evade the defects of traditional neural network such as neural recognition network is greatly affected by initial value, too high dimensions of neural network, and so on.

关 键 词：气体绝缘组合电器 等效时频法 模糊C-均值聚类法 最小二乘支持向量机 

分 类 号：TM835[电气工程—高电压与绝缘技术]