A Reinforcement Learning Framework for Decentralized Decision-Making in Smart Energy Systems

Abstract

The complexity of modern smart grids decentralized energy systems and renewable energy sources has increased, requiring advanced energy management solutions. The paper presents a framework for reinforcement learning for decentralized energy management in smart grids. Based on production, consumption, and storage dynamics, the proposed model adapts unit costs to the individual prosumers’ energy strategies. Meanwhile, external production operators (EPOs) have dynamically adjusted pricing in response to energy shortages and surpluses throughout the system. Through simulation, the framework demonstrates that actors with different energy profiles can independently design an optimized strategy, reducing the need for external energy supplies and stabilizing costs throughout the system. The research demonstrated the scalability and robustness of decentralized learning in energy management efficiency and adaptation and contributed to the development of smart grids.