采用小波分析与神经网络技术的局部放电统计识别方法

CLASSIFICATION OF PARTIAL DISCHARGE DISTRIBUTION PATTERNS USING WAVELET TRANSFORM AND NEURAL NETWORK

小波变换技术适合于局部放电(PD)的检测与识别,但所用小波或提取的特征量不合适时,识别效果不理想。因此,测量了6种典型绝缘缺陷模型产生的144组PD脉冲数据,用基数B样条导数小波成功地从这些实测数据的极强电磁干扰中检测出了PDs,得到了相应的三维统计分布图与各种二维统计分布图,提出了一种新的小波分解方法处理这些PDs对应的统计分布图,并在小波变换域定义了一组与传统方法完全不同的新特征量来识别这些PDs。 以新定义的特征量组成输入矢量,用基于BP算法的前馈型神经网络,对6种典型PDs及加上4种混合PDs组成的共10种PDs进行了识别测试,识别效果远远优于现有方法水平。

Wavelet transform(WT) can be used for detecting and recognizing parital discharges (PDs). Two key points for this purpose are choosing right wavelet and extracting proper features from the WT coefficeinces. 144 sets of PD pulse data are obtained from 6 types of insulation defect models. Using the Cardinal B-spline derivative wavelets all the PDs are picked up successfully from strong EM interference background encountered in the measurement field, and 2D and 3D statistical distribution patterns of these PDs are constructed. Then new WT ways to decompose distribution patterns are presented, new features in WT-domain to identify PDs are defined, and the feature extraction methods are developed. Taking these features as fingerprint parameters 10 types of distribution patterns (6 original SPDs and 4 mixed SPDs) are recognized correctly by a feed forward neural network using the BP algorithm. The recognition results are much more superior to others reported up to now.