A GAN-based Method for the Enhancement of Phase-Resolved Partial Discharge Map Data

Abstract

Partial discharge (PD) is one of the most important indicators of an impending failure in electrical power system components. The advantages of PD are: first, active maintenance at the very early stage; second, prevent the expensive interruptions in the supply of power; third, reduce the risk of catastrophic failures. The patterns of Phase-resolved partial discharge (PRPD) demonstrate the behaviors of PD. However, the challenges of recognizing the patterns of PRPD remain. Data inadequacy, noise in collected data, as well as differences introduced by different types of detection equipment, could inevitably lead to the deterioration of accuracy. Hence, this paper proposes a GAN (Generative adversarial network)-based method for the PRPD data enhancement. Specifically, we first build our method based on the auxiliary classification adversarial neural network (Auxiliary Classifier GAN, ACGAN) and train the model with four classic PD datasets. Next, we adopt three classic convolutional neural network models LeNet-5, AlexNet, and VGG-16, to validate our method. The results show that our proposed ACGAN-based PD data enhancement method generates a large amount of high-quality data, improving the recognition accuracy effectively. Thus, our method provides an effective solution for this task.