Exploring GAN and Their Derivatives in Power Data Generation: A Review
Zixuan Zhu, Xinqian Xia, Zikai Wang, Junjie Chen, Zelin Guo, Yiyan Li · 2023
In the continuously evolving and increasingly complex environment of power systems, utilizing Generative Adversarial Networks (GAN) to process and optimize multidimensional electric power data has grad-ually become a hot research topic. This paper explores the applications and challenges of utilizing GAN for large-scale electric power data processing, highlighting their ability to generate synthetic data amidst challenges like cost, equipment limitations, and data privacy in the power domain. While GAN showcase significant potential in enhancing time-series and image data sam-ples, they exhibit drawbacks, such as complex training and uncontrollable results. Therefore, various improved models, such as Conditional GAN, Wasserstein GAN, Wasserstein GAN-gradient penalty, etc. have been derived to address these issues, and are elaborated upon in the latter part of the review. Ultimately, this paper summarizes the current application challenges and future research directions of GAN in electric power systems.