PD Pattern Recognition of XLPE Cable Based on parameter optimal Support Vector Machine Algorithm

Abstract

The result of partial discharge(PD) pattern recognition is influenced by the convergence rate of artificial neural network and the parameter selection of support vector machine(SVM). M-ary classification theory is introduced to extend the generalization and learning ability of SVM algorithm into multi-class classifier. The improved genetic algorithm (GA) is used to optimize the penalty factors, slack variables and kernel function parameters of each sub-classifier. Then, the optimal parameter SVM classification model is constructed. The PD simulation experiment of Cross-linked polyethylene cable is carried out. The four fractal dimensions which represent the intrinsic fractal features of the gray image are extracted. It can be used as discharge fingerprint in the process of PD pattern recognition. The optimized SVM, un-optimized SVM and Radial Basis Function (RBF) neural network are used as a pattern classifier to complete the defect classification. The optimized SVM, un-optimized SVM and Radial Basis Function (RBF) neural network are used as a pattern classifier to complete the defect classification. The results show that the accuracy of defects recognition is higher than 95%when use parameter optimal SVM as the classifier. Whether the parameters are optimized or not, the recognition result obtained by using SVM as the classifier is better than the RBF neural network.