GAN-Based Data Augmentation Technique for Various Transmission Line Fault Data

Transmission line fault data plays an important role in power system reliability analysis and fault prediction. However, real fault data is not enough because transmission line faults do not frequently happen. To obtain various fault data, we propose a generative adversarial network (GAN)-based data augmentation technique. The proposed technique consists of three steps. i) it generates fault data using the wasserstein GAN with gradient penalty (WGAN-GP) model. ii) the generated data is filtered through an isolation forest (IF) algorithm, and iii) the filtered data is evaluated for its quality through KL-divergence. We visually showed that the proposed technique&quots data generation performance in terms of data diversity. It is also confirmed that the generated data is closer to the real fault data than the simulated data.