A Novel Deep Reinforcement Learning Approach for Microgrids Scheduling Considering Demand Response and Vehicle-to-Grid Systems

In response to the complexity and uncertainty brought by the surge of distributed energy and electric vehicle clusters in microgrid scheduling and operation, this paper combines probabilistic demand response (DR) modeling with vehicle to vehicle (V2G) technology to effectively smooth the load curve, reduce operating costs, and improve system flexibility and reliability. On this basis, this paper proposes a PERA-A3C optimization framework that enhances the exploration efficiency and convergence stability of the algorithm by prioritizing experience replay and adaptive entropy regularization. Experimental verification has confirmed the excellent performance of PERA-A3C in terms of cumulative rewards and convergence indicators, reducing daily operating costs by approximately 6%.