Application of Improved Reinforcement Learning Technology for Real Time Operation and Scheduling Optimization of Virtual Power Plant

This paper introduces an improved deep Q-network (MDQN) algorithm, utilizing deep reinforcement learning, to tackle the real-time dispatch problem in virtual power plants. Extensive comparative experiments have consistently shown that the MDQN algorithm surpasses existing algorithms, such as DDPG, SAC, and TD3, in terms of both training efficiency and performance. Notably, the MDQN algorithm strictly adheres to operational constraints within the action space, ensuring the feasibility of scheduling plans during real-time operations. The experimental findings underline the promising potential of the MDQN algorithm in effectively addressing real-time optimization and dispatch challenges encountered in virtual power plants. These results contribute to the existing research in the field and hold significant implications for practical applications in the virtual power plant domain.