A study on the optimal dispatch of storage and load in microgrids considering coordinated control of wind and solar power

Abstract

To address the increased output fluctuations, propagation of forecasting errors, and inconsistent responses between energy storage and flexible loads under high wind and photovoltaic penetration in microgrids, an optimal dispatch method integrating optical cloud image sensing, joint wind–solar–load forecasting, and safety-constrained rolling control is proposed. All-sky imaging data are used to extract cloud cover, cloud-motion direction, angular velocity, solar-occlusion probability, and estimated occlusion time. These features are temporally aligned with meteorological data, historical wind and photovoltaic power data, load data, and energy storage states. A Soft Actor-Critic algorithm based on maximum-entropy deep reinforcement learning is then combined with a safety action projection layer to generate feasible rolling dispatch decisions. Experimental results show that the proposed model reduces the root mean square errors of wind power, photovoltaic power, and load forecasting to 21.7, 16.9, and 18.6 kW, respectively. The daily operating cost is reduced to 5,128 yuan, while the wind and solar curtailment rate decreases to 2.4% and the peak-to-valley difference of the main-grid exchange power is compressed to 176 kW. No constraint violations occur during the dispatch process, and the computation time for each online decision is 0.043s. These results demonstrate that the proposed method can effectively coordinate renewable-energy forecasting, energy-storage operation, and flexible-load response, providing a computationally feasible solution for economical, reliable, and secure microgrid dispatch under high renewable-energy penetration.