부산대학교 도서관

Accurate and rapid fault diagnosis is crucial for the safe and stable operation of distribution networks. Although artificial intelligence gains popularity in fault diagnosis for distribution networks, the imbalance in the proportion of fault data and normal operational data hampers accurate diagnosis. To address this problem, a fault data enhancement model for distribution networks using the Wasserstein generative adversarial network with gradient penalty (WGAN-GP) is proposed. WGAN-GP uses the Wasserstein distance as the training target and introduces a gradient penalty mechanism to make the training process more controllable and efficient. Compared to the GAN-based method, WGAN-GP reduces the average relative error for nine dimensions in the standard deviation from 18.30% to 5.30% and in the mean from 4.03% to 0.48%, Compared with the probabilistic statistical based method, WGAN-GP greatly reduces the Wasserstein distance between real and generated data. Experimental results reveal that WGAN-GP can generate higher-quality data and solve the problem of data imbalance in power fault diagnosis.