Dynamic Economic Dispatch Considering Demand Response Based on Reinforcement Learning

Abstract

With the explosive growth of various participants and information in smart grid, data-driven methods such as reinforcement learning are getting increasing attention for solving problems concerning power system operation and management. In this paper, a dynamic economic dispatch method based on deep deterministic policy gradient (DDPG) algorithm is designed to minimize total operation cost of multi-period economic dispatch. The model for multi-period economic dispatch considering demand response is firstly established. To transform it into a reinforcement learning problem, the model is then reconstructed as a sequential decision-making process, with state, action and reward defined accordingly. A modified DDPG algorithm is introduced to solve the decision-making problem. Finally, case study based on a modified IEEE 14-bus system validates that the proposed method can obtain a satisfactory dispatch schedule which can approximate the effect of optimization solvers near real-time with robustness.