Multi-Agent Reinforcement Learning for Active Voltage Control on Multi-Hybrid Microgrid Interconnection System

Under the new trend of decarbonization, a large number of renewable distributed generations are integrated into the distribution network to form a multi-hybrid microgrid interconnection system. Aiming at the problem of bus voltage fluctuation caused by the uncertainty of source and load in hybrid microgrid, a voltage stability control method based on multi-agent reinforcement learning is proposed in this paper. The distributed power and energy storage devices in the interconnected system of multi-hybrid microgrid are used to alleviate power congestion and improve voltage quality. The bus voltage stability control problem is transformed into a Markov decision game process, and the reward function is designed according to the voltage stability. The voltage stability control model is constructed with the framework of centralized training and decentralized execution. This method does not require accurate power flow modeling, and constantly updates the neural network parameters through the continuous interaction between the agent and the environment to control voltage dynamic stability. Finally, the effectiveness of the proposed method is verified by numerical simulation, and the characteristics of different multi-agent reinforcement learning algorithms are compared and summarized.