Stochastic Scheduling and Learning-Based Control of Energy Storage in Active Distribution Systems

This paper presents a modeling and control framework for energy storage in active distribution networks. A stochastic optimal dispatch problem is first formulated and solved to address various uncertainties in distributed resources, building loads, and demand flexibility, accounting for weather variability, time-dependent building occupancy, and unpredictable demand patterns. To ensure safe operation and avoid potential violations arising from network constraints, a deep reinforcement learning controller is designed to refine control actions in real-time. A Monte Carlo tree search is used to enhance the learning agent’s exploration capability under dynamic conditions. Simulation results on the IEEE 123-node test feeder demonstrate enhanced performance, achieving a 13% average increase in economic benefits and significantly reducing voltage violations compared to baseline and conventional approaches.