Scenario-aware electric vehicle energy control with enhanced vehicle-to-grid capability: A multi-task reinforcement learning approach

Vehicle-to-grid (V2G) technology offers an innovative solution for integrating a range-extended electric vehicle (REEV) into the power grid. However, the associated energy management challenges require urgent attention. Traditional stationary-state energy management strategies often reserve excessive state of charge (SOC) due to inaccurate estimations of users' driving distances and difficulties in capturing their driving patterns. Given the strong coupling between the stationary-state and running-state scenarios, this paper proposes an integrated energy management strategy (I-EMS) based on a multi-task deep reinforcement learning (M-DRL) algorithm, with a gating mechanism to switch REEV energy control modes. The M-DRL agent dynamically adjusts the reserved SOC to optimize the V2G participation, while also considering the battery aging costs. Additionally, a hybrid approach combining Markov and Monte Carlo simulations is employed to model user commuting patterns, and comprehensive experiment results show that the I-EMS reduces operational costs by up to 19.2 % compared to the decoupled energy control system.