Adaptive Deep Q-Network Model for Intelligent Energy Storage Management in Microgrids

Abstract

In light of growing concerns over dependability, the integration of renewable energy, and the rising penetration of DG and active distribution networks, a new paradigm in power systems is taking shape: the SG. Intelligent Energy Storage Management is crucial in microgrids because microgrids are integrated power distribution systems that function as small-scale networks linked to the main grid. These networks comprise loads, renewable energy sources, and DG units that work together as one controllable entity. In order to forecast the SoH of a battery utilising, characteristics obtained from voltage, current, and temperature during charge-discharge cycles, this research presents a data-driven framework that emphasises appropriate data preparation techniques and a methodology. To handle storage operations, it build a RL method based on DQN to choose the best charging and discharging actions under operational restrictions by estimating the action-value function. The results of the simulation show that the suggested approach controls storage units nearly optimally, with the DQN model reaching an accuracy of 95.22%. These results validate the strategy's ability to improve operational efficiency and decision-making. In sum, the research highlights reinforcement learning's capacity to pave the way for contemporary microgrid systems to implement adaptable, dependable, and environmentally friendly energy management solutions.