Multi-agent reinforcement learning energy management method for microgrids

作者：Jun Wan, Jiawei Pan, Fang Bi · 发表于：2025 IEEE 3rd International Conference on Sensors, Electronics and Computer Engineering (ICSECE) · 年份：2025 · DOI：10.1109/icsece65727.2025.11257028

Aiming at the problems of complex distributed collaboration and strong dynamic uncertainty in microgrid energy management, this paper proposes a multi-agent reinforcement learning energy management method for microgrids. Firstly, a microgrid system and agent communication modeling covering grid-connected and island operation characteristics are constructed to clarify the unified optimization goal of energy management. Secondly, a multi-agent reinforcement learning method based on a centralized training and decentralized execution framework is designed, using MADDPG as the main line, and introducing MASAC and QMIX methods for comparative verification. At the same time, the training efficiency is improved through a differential reward mechanism. Subsequently, a Simulink and PyMARL joint simulation platform is built, and two scenarios of normal operation and abnormal disturbance are set up to systematically evaluate the performance of the proposed method in terms of revenue improvement, energy optimization and abnormal recovery capability. Finally, the research results are summarized and the future development direction of fusion prediction collaborative optimization and large model-assisted decision-making is prospected.