Reinforcement Learning-based Scheduling Method for Photovoltaic Power Integrated Electric Vehicle Charging Stations

Due to environmental changes and energy crises, the widespread adoption of electric vehicles has become an inevitable trend. However, optimizing the planning of charging stations is crucial due to the limited charging power and diverse charging behaviors of EV owners. In this paper, we first analyze the charging data from a charging station in Chongqing, China, and establish the charging simulation environment. Following this, we modeled the problem as a Markov decision process. Then, a reward function concerning the satisfaction level of EV owners and the profit of the charging station is proposed. Ultimately, this reward function was applied to the EV charging optimization problem using the importance-weighted actor-learner architecture (IMPALA) algorithm. After convergence, the average reward in one episode of the IMPALA algorithm outperformed the genetic algorithm (GA) by 43.7%.