An operational optimization strategy for home energy system: a deep reinforcement learning and mathematical programming approach

This article addresses the challenges of uncertainty in home energy system (HES) optimization, such as price fluctuations, temperature changes, and demand randomness, as well as the need for rapid decision-making. We propose a two-layer optimization framework that combines deep reinforcement learning (DRL) with mathematical optimization (MO). In the upper controller, DRL dynamically adapts to uncertainty, Optimizing the total household power by autonomously exploring, in the lower controller, mixed-integer programming (MIP) is used to allocate power to various household devices, ensuring both scheduling accuracy and speed. The lower controller optimization results are used as immediate rewards for the upper DRL controller, reducing the exploration and trial-and-error space of the DRL process and fundamentally improving the optimization speed. The proposed method performs well in uncertain environments and shows a significant advantage in convergence speed compared to existing DRL-based methods. In addition, since national standard documents regulate the self-description files of devices, our method is also applicable to industrial energy systems and demonstrates strong engineering applicability.