Energy Purchase Optimization for Microgrid Systems Using Deep-Q-Learning

The increasing integration of renewable energy sources into microgrid systems has introduced challenges due to the intermittent and unpredictable nature of renewable energy generation. To address these challenges, this paper proposes a reinforcement learning (RL)-based framework for the optimal utilization of available energy resources and information in microgrids, aiming to minimize energy purchase costs from the power grid while ensuring efficient energy management. The designed system incorporates real-time renewable energy production forecasts, time-varying market prices, and battery storage characteristics, enabling dynamic decision-making for energy procurement and storage. A Deep Q-Network (DQN) is employed to learn optimal strategies by interacting with a simulated microgrid environment. The proposed approach achieved a 14.91% reduction in total energy costs compared to a rule-based baseline, along with a 10.74% reduction in grid energy usage during peak periods. We anticipate that this model will be useful in household microgrid applications.