Deep Reinforcement Learning-Based Voltage Optimization Control for Microgrid Energy Storage Systems

The integration of high-penetration renewable energy sources poses significant challenges to the secure and stable operation of microgrids. Energy storage systems are critical resources for mitigating voltage fluctuations, yet their lifespan is closely related to the frequency and depth of charge/discharge cycles. To address the coordination between grid stability and storage economic protection, this paper proposes a voltage optimization control method based on Dynamically Evolving Deep Reinforcement Learning (DEDRL). A voltage control model is established that accounts for storage life degradation costs. These costs are integrated into the optimization objective. The DEDRL algorithm introduces a population evolution mechanism for the critic network within the actor-critic framework. This mechanism coordinates adaptive crossover, directional mutation, elitist selection, and local evolution. Through their coordination, the algorithm achieves a balance between global exploration and local exploitation. It also alleviates the tendency of gradient-descent methods to converge to local optima. Simulation results demonstrate that DEDRL significantly improves the average reward compared to baseline algorithms. The system node voltage deviation is reduced by 37.79% after energy storage integration.