Multi-energy Management of Interconnected Multi-microgrid System Using Multi-agent Deep Reinforcement Learning

The multi-directional flow of energy in a multi-microgrid (MMG) system and different dispatching needs multiple sources time location hinder the optimal operation coordination between microgrids. We propose an approach to centrally train all agents achieve coordinated control through individual attention mechanism with deep dense neural network for reinforcement learning. novel allow each agent attend specific information that is most relevant its reward. When training complete, proposed can construct decisions manage within MMG fully decentralized manner. Using only local information, coordinate internal allocations microgrids external multilateral multi-energy interactions among interconnected enhance operational economy voltage stability. Comparative results demonstrate cost achieved by at 71.1% lower than obtained other multi-agent learning approaches.