Research on Online Optimal Dispatch Strategy of Virtual Power Plants Based on Deep Reinforcement Learning

To ensure the long-term stability of virtual power plant (VPP) online dispatch, this research proposes a strategy that integrates optimization models with Lyapunov optimization and deep reinforcement learning techniques. The proposed approach features a two-layer optimization model that separates the temporal aspects, addressing the complexity of dispatchable resources within the VPP. In the outer layer, a deep reinforcement learning algorithm optimizes the temporal dispatch of VPP resources, including thermal power units, energy storage devices, and flexible loads. The inner layer focuses on optimizing remaining resources, such as wind power and photovoltaic equipment. The model, once trained with historical VPP data, will enhance online dispatch based on real-time operational data. Analysis indicates that this method effectively manages all VPP resources while maintaining dispatch reliability.