Partial Discharge Pattern Classification based on Deep Learning for Defect Identification in MV Cable Terminations

Abstract

This paper presents a method based-on Convolutional Neural Network (CNN) to identify the partial discharge (PD) sources located in medium voltage (MV) cable terminations. Five different defect types due to improper workmanship are artificially fabricated in 36 kV cross-linked polyethylene (XLPE) cable terminations. Phase-resolved Partial Discharge (PRPD) patterns of the aforementioned defects are recorded for 30 seconds (1500 cycles of the input voltage) to generate one pattern. Different cable sets are utilized in the training and testing steps to test the robustness of the model, and 600 patterns are extracted for each cable group. Every PRPD data is converted into a Red-Green-Blue (RGB) image to be inputted to the proposed CNN algorithm. The algorithm is used for both feature extraction and classification steps. Effects of hyperparameters on the performance of the method are evaluated and optimized.