Federated Multi-Agent Reinforcement Learning for Incentive-Based DRS over Blockchain enabled Microgrids

The paper investigates the integration of multiagent reinforcement learning (MARL) and federated learning for smart energy management systems focusing on Sustainable Development Goals (SDG) 7 and 11 which are Affordable and Clean Energy, and Sustainable Cities and Communities. We present a demand response system (DRS) within microgrids that is incentive-based with solar energy as the main renewable resource. Our model incorporates the use of federated learning to improve data privacy and efficiency, while multi-agent reinforcement learning optimizes the process of energy trading, load balancing, and grid stability. Three algorithms-Deep Deterministic Policy Gradient (DDPG), Multi-Agent Deep Deterministic Policy Gradient (MADDPG), and Asynchronous Advantage Actor-Critic (A3C)-are compared in terms of their performance for this purpose. The study also investigates how direct load control (DLC) programs along with demand bidding can be utilized as incentives that encourage participation in DRS. The experimental results show that merging MARL with federated learning in microgrids leads to improved energy efficiency, cost reduction, as well as sustainable energy consumption behaviours. These outcomes support broader targets to establish strong, durable energy systems in smart cities under the United Nations’ SDGs.