Abstract: With the rapid development of power systems, how to fully utilize the massive, multi-source and multi-dimensional power data is one of the important technical issues faced by the power industry currently. Compared with the traditional machine learning algorithms, deep learning has superior performance in dimensionality reduction, non-linear fitting and feature extraction. Among them, the generative adversarial network (GAN) has advantages in power data enhancement and generation. In this paper, the fundamental theory of GAN is introduced, and the advantages and disadvantages of GAN are analyzed. From the perspective of the network structure and the objective function, four GAN derivative models widely used in power systems are introduced respectively. On this basis, the applications of GAN in power systems are reviewed in detail, and the GAN models in each application scenario and its characteristics are discussed. Finally, the problems to be solved for further application of GAN in power systems are summarized, and deeper and broader applications of GAN are presented.