Optimal Charging Control for Electric Vehicle Fleet Using Mean Field Game Theory

Optimizing the charging protocol for large-scale electric vehicles is complex and computationally costly. Therefore, this paper proposes an advanced approach using machine learning-assisted mean field game theory to handle this issue. Mean field game theory simplifies the communication between electric vehicle owners and the grid operator by considering the probabilistic flow of the electric vehicle group. While analyzing stochastic dynamics environments like coordinating a fleet of electric vehicles is challenging analytically, the numerical methods often approximate the solution of the mean field mathematical model consisting of nonlinear partial differential equations Hamilton-Jacobi-Bellman and Fokker-Planck equations. Thus, the actor-critic reinforcement learning algorithm can be used to enhance the performance of electric vehicle charging strategies. The proposed method aims to stabilize the grid demands and prevent overloading by refining the charging actions of electric vehicles. Furthermore, the grid search hyperparameter optimization approach is utilized to improve the efficiency of the training process. The kernel estimator is implemented to verify the proposed approach. The heterogeneity of electric vehicles is implemented within the model by considering different charging efficiency and energy consumption rates to reflect reality. A numerical example will be presented to demonstrate the efficacy of the proposed approach.