Fictitious Self-Play for Vehicle-to-Grid Game with Imperfect Information

The vehicle-to-grid (V2G) technique, which enables the bidirectional power exchange between electric vehicles (EVs) and power grid, becomes promising in current smart grid research. In this paper, a game theoretic model is proposed to study the interaction among EVs in a V2G system with V2G technique incorporated. This V2G game is a game with imperfect information in which each EV does not any private information of other EVs. To find the Nash equilibrium of this game, a machine learning-based algorithm is proposed based on fictitious self-play. Our simulation results show that the proposed algorithm can approximately converge to the Nash equilibrium of the game under imperfect information. This demonstrates the efficacy of the proposed algorithm in solving the V2G game. A pre-training approach is also proposed to accelerate the convergence of the algorithm by using the historical data from the interactions of EVs in the game.