Optimization Strategy for Virtual Power Plants Participating in Peak Shaving Market Based on Multi-agent Reinforcement Learning

To address the challenge of service selection due to the incomplete information, such as the participation of multi-adjustable resources in the virtual power plant and the competition restriction relationship, a multi-agent game strategy for virtual power plants is proposed based on multi-agent reinforcement learning and Bayesian game. Firstly, a virtual power plant model containing multiple adjustable resources is established according to different resource characteristics. Secondly, a Bayesian game model for the optimal dispatch of virtual power plants participating in the peak regulation market is established, considering environment and economy coordination. The output of virtual power plant resources participating in peak regulation market is optimized, thereby confirming the existence of a Nash equilibrium solution. Finally, in the incomplete information market environment, the multi-agent reinforcement learning algorithm combined with Bayesian game is used to solve the above model, and the multi-agent optimal dispatching scheme for virtual power plants under multi-objective can be obtained. Simulation analysis indicates that, under the game model of controllable resources with incomplete information, the optimal dispatching scheme for virtual power plants can effectively improve economic efficiency and reduce carbon emissions. This strategy achieves a win-win outcome in terms of both economic and environmental multi-objective optimization.