基于统计不相关最优鉴别矢量集的GIS局部放电模式识别

PD Pattern Recognition Based on Optimal Sets of Statistical Uncorrelated Discriminant Vectors in GIS

针对气体绝缘金属封闭开关设备(GIS)局部放电及其缺陷特点,设计了4种典型的GIS缺陷模型,采用甚高频高速采集大量局部放电样本,构造了局部放电灰度谱图。文中将主分量分析-线性鉴别方法应用于局部放电模式识别,即首先进行主分量分析,将数据从超高维空间降至低维空间,再提取统计不相关的最优鉴别矢量集,最后采用最小距离分类器进行模式识别。识别结果表明该方法对GIS各类模拟缺陷的正确识别率高,效果良好。

In view of the PD （partial discharge） characteristics and their defects, four kinds of GIS （gas insulator switchgear） defect models are designed and GIS gray intensity images built based on numerous samples collected by the very-high frequency and high speed system. A PCA-FDA （principal component analysis-Fisher discriminant analysis） algorithm based on PD images is proposed for PD pattern recognition. To begin with, the principal component analysis is introduced to condense the PD images from a hyper-high dimension to a low one, and then the optimal sets of statistically uncorrelated discriminant vectors are extracted. Finally, the minimum distance classifier is adopted for pattern recognition. The identified results indicate that the proposed method can effectively improve the discrimination of the four kinds of defects in GIS.