Abstract
The uncertainties introduced by distributed energy resources bring significant challenges to the real-time operation of microgrids, particularly in managing the fluctuating generation and load profiles required for optimal energy management system (EMS) performance. This study proposes a deep reinforcement learning-based framework to achieve real-time scheduling of battery storage within a grid-connected microgrid. The control problem is formulated as a Markov Decision Process (MDP), a Double Deep Q-Network (DDQN) agent is trained to generate optimal charging and discharging actions based on observed system states. Simulation results demonstrate that the proposed EMS controller can effectively derive optimal control policies, ensuring system stability, cost efficiency, and energy selfsufficiency.