A Generative Adversarial Imitation Learning-based Unit Commitment Strategy with Renewable Distributed Generators

Abstract

With the integration of large-scale renewable distributed generators (RDGs), the uncertainties and complexity of the security-constrained unit commitment (SCUC) problem have increased significantly. Traditional model-driven methods struggle with computational speed and the need for high-precision modeling, while reinforcement learning (RL) approaches require manually defined reward functions. To address these issues, this paper proposes a novel SCUC strategy based on Generative Adversarial Imitation Learning (GAIL). The proposed strategy allows for the direct learning of the optimal SCUC policy under the guidance of an established expert system. To enhance the quality of the scheduling strategies generated by the generator network, this paper introduces the loss function from the proximal policy optimization (PPO) algorithm. The effectiveness of the proposed method is demonstrated through a simulation case study of a provincial power grid in China.