@article{Pu2026, 
author = {Xiaowei Pu and Hao Xiao and Wei Pei and Yanhong Yang and Li Ma and Tengfei Ma and Shuai Zhang},
title = {Optimal Energy Management of Networked Multi-energy Microgrids Based on Improved Multi-agent Federated Reinforcement Learning},
year = {2026},
journal = {CSEE Journal of Power and Energy Systems},
volume = {12},
number = {5},
pages = {2209-2220},
keywords = {Energy management, FA-MADDPG method, federated reinforcement learning, networked multi-energy microgrids},
url = {https://www.sciopen.com/article/10.17775/CSEEJPES.2023.05920},
doi = {10.17775/CSEEJPES.2023.05920},
abstract = {The energy management of the networked multi-energy microgrids (MEMGs) contributes to promoting the consumption of renewable energy and balancing the different interests of multi-agent MEMG systems. A novel energy management method of MEMG system based on improved multi-agent federated reinforcement learning is proposed. First, this study establishes the optimization model of MEMG system considering multi-party economic benefits in P2P trading market. To solve the problems of gradient instability and low learning efficiency in traditional federated reinforcement learning with FedAvg federation strategy, a novel Federated Attention-based Multi-agent Deep Deterministic policy gradient (FA-MADDPG) framework is proposed for real-time scheduling optimization of MEMG system. In addition, this study proposes an agent exploration mechanism based on directional noise to achieve the SoC balance of distributed energy storage before and after scheduling. Finally, a numerical example is given to verify the effectiveness of the proposed method. Compared with MADDPG and federated MADDPG method, the proposed method has better convergence and energy scheduling ability, and the economic cost is reduced by 13.57% and 10.22% respectively.}
}