Medium and Long Term Scenario Generation Method Based on Autoencoder and Generation Adversarial Network

Abstract

With the proposal of carbon peak and carbon neutrality target, developing renewable energy power generation technology is imperative, accurate simulation for renewable energy power characteristics provides important guarantee for reliable operation of power system. In view of the difficulty in simulating the sequentiality and uncertainty of medium and long-term output scenarios, this paper proposes a medium and long-term scenario generation method based on stacked autoencoder (SAE) and improved conditional generative adversarial networks(CWGAN_GP). Firstly, the ability of automatic feature extraction based on SAE to extract hidden features efficiently, which reduces dimensions for high dimensional series. Secondly, feature weighted K-means clustering method and CWGAN_GP are used to generate typical medium and long term scenarios. Finally, a case is used to verify the effectiveness and serviceability of the proposed method for the simulation that has medium and long-term scenarios with sequentiality and uncertainty.