Partial Discharge Data Augmentation of High Voltage Cables based on the Variable Noise Superposition and Generative Adversarial Network

Abstract

Partial Discharge (PD) pattern recognition is one of the most important steps of PD based condition monitoring and diagnosis of High Voltage (HV) cables. Although different types of pattern recognition methods, e.g. Support Vector Machine (SVM), Back-propagation Neural Network (BPNN) and Deep Learning, have been developed and applied to PD pattern recognition, limited training samples is one of the most important factors which restricts the PD pattern recognition accuracy. To overcome the challenge two PD data augmentation methods, based on the Variable Noise Superposition (VNS) and Generative Adversarial Network (GAN), are presented in the paper, which are evaluated with 1500 sets of experimental PD data and three pattern recognition methods, Support Vector Machine (SVM), Logical Regression (LR) and Random Forest (RF). The results show that PD pattern recognition accuracy, based on the VNS data augmentation method, is improved by 0.99%, evaluated with SVM, and is improved by 0.96%, evaluated with RF; PD pattern recognition accuracy, based on the GAN data augmentation method, is improved by 0.52%, evaluated with SVM, and is improved by 1.72%, evaluated with LR. Both two methods, VNS and GAN, are effective for PD data augmentation, which are applicable for PD pattern recognition of HV cables and other HV apparatuses, especially for pattern recognition methods which requires large volume of training samples.