Optimal scheduling of the virtual power plant with electric vehicles using Dueling DDQN

With the application of Vehicle-to-Grid technology, optimizing the scheduling of EV charging stations and surrounding distributed renewable energy generation as a Virtual Power Plant (VPP) helps improve the economic efficiency and satisfaction of EV users’ charging and discharging, while also increasing the utilization of distributed renewable energy and smoothing grid load fluctuations. However, the overall charging and discharging load characteristics of EV charging stations are an aggregation of the random behaviors of numerous individual EV users, making it difficult to accurately describe them with mathematical models, posing a challenge for traditional centralized optimization scheduling. This paper proposes a VPP scheduling framework based on the Dueling Double Deep QNetwork, aiming to maximize the total benefit of VPP operations, including minimizing the total charging cost for EV users and penalties for unmet charging demands. Through interactive evaluation with the regional power grid environment, the framework continuously learns and updates action strategies. Simulation results show that this method enhances the benefits for EV users, coordinates EV charging and discharging loads with distributed renewable energy generation, achieves peak shaving and valley filling of the overall VPP load, and improves the overall operational performance of the VPP.