A Comprehensive guide to Generative Adversarial Networks (GANs) and application to individual electricity demand

Abstract

Generative Adversarial Networks (GANs) have become a popular research topic in many research fields in the last decade. The fragmented body of knowledge on GANs drives researchers to a trial-and-error procedure while choosing the right GAN for a given task. This study provides a comprehensive guide on GANs, where we start with considering problems such as nonconvergence, mode collapse, vanishing gradient, and unstable training. We then compare various GANs concerning the application perspective, their outputs, and evaluation metrics. In addition, we suggest using a new evaluation metric to identify the best candidate of GANs for time series tasks, the so-called average coverage error (ACE). Finally, we discuss the application of the ACE using the electricity consumption of an individual industry customer and display significant selection criteria. The ACE is beneficial in identifying the best GANs by comparing outcomes and reduces the computational cost of searching related to the evaluation metrics.