SVAE-WGAN Data Augmentation Method for Partial Discharge Fault Detection of 5G based Power Intelligence System
Liwen Qin, Xiaoyong Yu, Lifang Wu, Qin Bin · 2023
Partial discharge is one of the most popular fault in power system, and the popularity of 5 G makes it possible to analyze partial discharge faults using artificial intelligence algorithms. Low frequency and high collection cost of partial discharge faults lead to the problems of small sample size and imbalance sample, which bring challenges to partial discharge fault detection of 5G based power intelligence system. Hence, in this paper, an data augmentation method that named SVAEWGAN is proposed, which based on stacked variational autocoding and Wasserstein generative adversarial network for partial discharge fault samples in transmission lines to generate fault samples and improve the diversity of samples. Firstly, the stack variational autoencoder is used to extract the depth features of the fault waveform. Secondly, a partial discharge fault sample generation model is constructed by combining stacked variational autocoding and Wasserstein generative adversarial network, and a partial discharge fault detection framework based on the data augmentation method is proposed. Finally, the performance of partial discharge fault detection algorithm based on LightGBM and Bi-LSTM-Attention before and after using data augmentation method is compared and analyzed based on VSB fault detection dataset by selecting accuracy and F1 score, etc. Experiments show that the data augmentation method can effectively improve the performance of partial discharge fault detection, which proves the effectiveness of the proposed method.