A Q-Learning Based Charging Scheduling Scheme for Electric Vehicles

This paper presents a Q-learning algorithm based dynamic charging scheduling scheme which intent to optimize the operation benefit for electric vehicles. The method imitates the charging station operator's illation and decision procedure which similar to solving a reinforcement learning problem. The scheduling problem involved is focusing on the bidirectional interaction between the vehicle and the grid, including the grid-to-vehicle charging and the vehicle-to-grid (V2G) electricity returning. Regarding the dynamic characteristics of the electricity market, the scheme has included the time-of-use electricity rates as a core parameter to establish the reward tables which is necessary for learning. Furthermore, several simulations were conducted which demonstrates the day-long optimal vehicle charging decisions under the proposed scheme. Favorable expansibility and maintainability can be achieved in this Q-Iearning framework.