Reinforcement Learning Based Energy Trading Strategy for Decentralized Microgrids

作者：Divya Negi, Mehul Manu, Pankaj Kumar, Shweta Kushwaha, Rudramani Bhutia, Arpit Goel · 发表于：2026 1st International Conference on AI, Data Science, Cyber Security and Smart Manufacturing for Sustainable Development (ICADCS) · 年份：2026 · DOI：10.1109/icadcs70036.2026.11583197

In the case of decentralised microgrids prone to renewable variability, this study aims to offer an effective and privacy-compliant energy trading service since conventional coordinating methods are not necessarily stable or scalable. Improving peer-to-peer was the driving factor behind this study's development and evaluation of a multi-agent reinforcement learning system based on federated clustering. utilising real-world datasets from India, experimental research was carried out to train numerous microgrid agents utilising an FC-MASAC framework that incorporates federated aggregation, clustering, and forecasting. The proposed model minimized the total trading cost to 12150 in contrast to 15230 of MADDPG and 13780 of MASAC and minimized the energy imbalance to 9.8 kWh. It reached a higher convergence rate on 240 episodes and a higher local energy usage of 79.6 with a minimised grid dependency of 20.4. These findings suggest that the proposed framework is more efficient in terms of trading, stability in coordination and scalability, and it is viable solution to decentralised energy management of smart grids.