摘要
we introduce a novel approach, called cooperative distributed Q-learning, building upon our prior distributed Q-learning methods, aimed at achieving autonomous frequency synchronization within an isolated AC microgrid (MG). Within the framework of multi-agent reinforcement learning, each agent's objective is to collaboratively determine an optimal policy aligned with the central performance function by exchanging local learning parameters across a communication network. To realize this goal, the Q-function of each individual agent will append a mixing term containing the average Q-function of neighboring agents received via communication networks. Simulation results demonstrate that the inclusion of this mixing term in each individual Q-function can indeed enhance the convergence speed for plug-and-play operations within isolated AC MGs.