Optimal scheduling for charging and discharging of electric vehicles based on deep reinforcement learning

The growing scale of electric vehicles (EVs) brings continuous challenges to the energy trading market. In the process of grid-connected charging of EVs, disorderly charging behavior of a large number of EVs will have a substantial impact on the grid load. Aiming to solve the problem of optimal scheduling for charging and discharging of EVs, this paper first establishes a model for the charging and discharging scheduling of EVs involving the grid, charging equipment, and EVs. Then, the established scheduling model is described as a partially observable Markov decision process (POMDP) in the multi-agent environment. This paper proposes an optimization objective that comprehensively considers various factors such as the cost of charging and discharging EVs, grid load stability, and user usage requirements. Finally, this paper introduces the long short-term memory enhanced multi-agent deep deterministic policy gra dient (LEMADDPG) algorithm to obtain the optimal scheduling strategy of EVs. Simulation results prove that the proposed LEMADDPG algorithm can obtain the fastest convergence speed, the smallest fluctuation and the highest cumulative reward compared with traditional deep deterministic policy gradient and DQN algorithms.