Resilient Multi-Agent Reinforcement Algorithm for Distributed Energy Optimization in Smart Microgrids

作者：Malati Basnet, Ching Joo Jie · 发表于：2025 International Conference on Intelligent Innovations in Engineering and Technology (ICIIET) · 年份：2025 · DOI：10.1109/iciiet65921.2025.11378467

The growing complexity of smart microgrids, characterized by distributed renewable energy resources, dynamic load demands, and decentralized control, necessitates intelligent and adaptive energy management solutions. Traditional centralized optimization approaches often suffer from scalability issues, communication overhead, and vulnerability to single-point failures. To address these challenges, this paper proposes a Resilient Multi-Agent Reinforcement Learning Algorithm (RMARL) for distributed energy optimization in smart microgrids. The RMARL framework employs autonomous agents representing individual energy units-such as generators, storage systems, and consumers-that collaboratively learn optimal policies through decentralized reinforcement learning. Each agent utilizes local state information and limited peer communication to balance energy generation, consumption, and storage, thereby enhancing overall grid efficiency and resilience. The algorithm integrates a resilience-aware reward function to ensure stability under uncertain conditions such as renewable intermittency, communication loss, and sudden demand surges. Experimental results from simulation studies on IEEE 33-bus and modified microgrid test systems demonstrate that RMARL achieves faster convergence, improved energy efficiency by 12.6%, and enhanced fault tolerance compared to conventional Q-learning and deep deterministic policy gradient (DDPG)-based methods. Furthermore, the distributed nature of RMARL ens...