Classification of Partial Discharge Sources in XLPE Cables by Artificial Neural Networks and Support Vector Machine

Abstract

Classification of partial discharge (PD) patterns is a significant tool in identifying the type of defects in cables. Development of reliable classifiers to identify various defects in the cable insulation is of vital importance in assessing the condition of cables in service. This paper proposes the development of Artificial Neural Network (ANN) based classifiers and Support Vector Machine (SVM) classifier for identification of cable defects such as voids, metal particle in the insulation, high potential metal tip, semiconductor layer tip, metals in the insulation and insulation incision. PD measurements are done on 11 kV XLPE cables with defects and wavelet based de-noising method is applied to abstract the PD pulses. Various PRPD features are extracted and used for training the ANN and SVM based models in MATLAB environment. The performance of SVM classifier and ANN based back propagation neural network classifier are analyzed for various types of defects. Classification accuracy of each models are analyzed and feasibility of optimum models for classification of cable defects is presented.