Answers are grounded in this paper and cite their sources.
The growing integration of PV and energy storage systems (ESSs) poses operational challenges for power distribution grids (PDGs), mainly due to generation uncertainty and voltage-profile variations. Typical receding-horizon Model Predictive Control (MPC) fails to achieve efficient control due to forecast uncertainty. Meanwhile, it is difficult to directly improve the forecast due to limited feature availability, restricted data access, and insufficient domain knowledge. In this paper, we propose a Safe Deep Reinforcement Learning (SDRL) hybrid receding-horizon MPC to manage forecast uncertainty. Using an expert model based on historical data, the SDRL agent adjusts the receding-horizon MPC output to achieve better performance that is closer to the expert. The control objective is to minimize the impact of renewable energy on the external grid while ensuring critical grid operational safety, such as voltage levels. The proposed hybrid controller is tested under various forecast uncertainties, and the results show that it can improve receding-horizon MPC performance by 65% (reducing the PCC average daily fluctuation from 92 to 32) while ensuring voltage safety. Besides, the controller is tested over various forecast update intervals and MPC horizons to demonstrate its robustness. Furthermore, a scalability test shows that while the full SDRL suffers from scaling up, the proposed hybrid controller overcomes this issue and achieves better performance, minimizing the impact of the PDG by 42–62%.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: