Research on Multi-objective Charging Scheduling Strategy for Electric Vehicles: A Deep Reinforcement Learning Approach

Electric vehicles (EVs), with their energy-saving, eco-friendly, and efficiency-driven characteristics, have emerged as a crucial component of the smart grid. However, their large-scale assimilation into the power grid poses challenges to the stability of grid operations. It is key issue to ensure the benefits of EV users while maintaining grid stability. Adopting a user-oriented perspective, this paper comprehensively considers the utilities of EV users and the grid. The optimization objective is to minimize both user costs and grid load curve variance. We formulate the multi-objective EV charging scheduling problem as a Markov decision process (MDP). To determine the optimal charging strategy, we introduce a real-time scheduling framework employing the soft actor-critic (SAC) algorithm. The multi-objective optimization goal is achieved through the configuration of three reward components. In the proposed framework, an actor network guides the EV user’s charging strategy, while two Q-networks address the issue of Q-value overestimation and two target Q-networks ensure training stability. By utilizing this proposed approach, EV users can guarantee charging satisfaction, adhere to constraints, maximize their utilities, and contribute to power grid peak shaving. Simulation results validate the efficacy of our proposed method.