A Novel Graph Reinforcement Learning Approach for Stochastic Dynamic Economic Dispatch under High Penetration of Renewable Energy

Abstract

Due to the fast development of new-type power systems, the power grid is facing increasing uncertainties brought by high penetration of distributed generations. How to improve the decision quality of economic dispatch under such conditions becomes a very crucial task. Therefore, a novel graph reinforcement learning (GRL) approach for dynamic economic dispatch under high penetration of renewable energy is proposed. Compared with other reinforcement learning method, a novel graph-based representation of system state is adopted. Thus, the implicit correlations of uncertainties considering system topology can be more effectively captured. As a fully model-free method, the proposed methodology does not rely on the explicit models of physical system and uncertainty distributions. The asymptotic optimal policy can be obtained by continuous interaction with the environment. Case simulations illustrate that the proposed method achieves a better performance in terms of optimality and efficiency compared with existing learning methods.