Long-Term and Short-Term Coordinated Scheduling for Wind-PV-Hydro-Storage Hybrid Energy System Based on Deep Reinforcement Learning
Huaiyuan Zhang, Kai Liao, Jianwei Yang, Zhe Yin, Zhengyou He
Abstract
For wind-photovoltaic-hydro-storage hybrid energy systems (WPHS-HES) grappling with the complexities of multiple scheduling cycles, traditional long-term strategies often impair short-term regulation capabilities, leading to extensive resource waste and critical power shortages. Thus, this paper introduces a novel framework that intricately nests short-term operational characteristics within long-term operating rules to synchronize multi-timescale scheduling for WPHS-HES. The cornerstone of our approach is the novel formulation of the long-term scheduling as a Markov Decision Process (MDP). It is integrated seamlessly with short-term generation schedules developed through an optimal model embedded at each MDP step. To achieve computational effectiveness and reliability, we propose a hybrid data-model-driven solution that harnesses the synergistic benefits of both data-driven and model-driven methodologies. By leveraging deep reinforcement learning our approach significantly streamlines long-term decision variables, while ensuring strict adherence to short-term operational constraints through mixed integer linear programming. Empirical simulations on an operational WPHS-HES validate the superior efficacy of our method over traditional scenario reduction and robust optimization techniques. The results are striking that it achieves a reduction in sustainable energy curtailment from 11.67% to 0.63% and slashes the load shedding rate from 3.3% to 0.69%, thereby setting a new benchmark for intelligent energy management in complex hybrid systems.