Bai Yunxiao, Sui Yu, Deng Xiaoyu, Wang Xiangbing
Guangdong Power Grid Co., Ltd., CSG, Guangzhou, 510080, China.
Sci Rep. 2025 Aug 12;15(1):29589. doi: 10.1038/s41598-025-12280-4.
The integration of photovoltaic (PV) generation and hydrogen storage in rural microgrids enables clean, long-duration energy supply, yet introduces operational challenges under high uncertainty. These include fluctuations in PV output, stochastic hydrogen demand, and volatile market prices. Traditional deterministic or static robust methods often fall short in handling such dynamic, multi-dimensional uncertainties. To address these gaps, this study develops a Quantum-Inspired Robust Optimization (QRO) framework that coordinates PV-H microgrid scheduling by dynamically adapting to real-time disturbances. The QRO approach leverages quantum-classical hybrid principles and integrates distributionally robust optimization with reinforcement learning. Uncertainty sets evolve adaptively based on operational feedback rather than remaining fixed, enhancing resilience to cyberattacks, extreme weather, and grid outages. Deep Q-learning and policy gradient methods are employed to continuously refine dispatch strategies, ensuring robust performance in non-stationary environments. A case study on a 5 MW PV-H microgrid with a 3 MW electrolyzer and 2 MW fuel cell demonstrates the practical effectiveness of the proposed framework. The model incorporates real historical solar profiles, stochastic demand, and price signals over a full-year horizon. Stress-testing under scenarios such as 48-hour grid failure, signal-based cyberattacks, and 40% PV output curtailment reveals substantial gains: operational costs are reduced by 9.3%, resilience scores improve by over 20% in adverse conditions, and convergence speed increases by 42% relative to classical optimization. These improvements reflect not only enhanced adaptability and computational efficiency, but also the practical feasibility of real-time learning in resilient energy scheduling. Importantly, the term "quantum-inspired" refers to classical algorithms that emulate quantum principles-such as probabilistic reasoning and solution diversity-without employing quantum hardware. By unifying quantum-inspired modeling, distributional robustness, and reinforcement learning, this framework offers a scalable and adaptive solution for next-generation hydrogen-based microgrid operations.
农村微电网中光伏发电与储氢的集成实现了清洁、长期的能源供应，但在高不确定性下带来了运行挑战。这些挑战包括光伏输出波动、随机的氢气需求以及波动的市场价格。传统的确定性或静态鲁棒方法在处理这种动态、多维度的不确定性方面往往不足。为了弥补这些差距，本研究开发了一种量子启发式鲁棒优化（QRO）框架，通过动态适应实时干扰来协调光伏 - 氢微电网调度。QRO方法利用量子 - 经典混合原理，并将分布鲁棒优化与强化学习相结合。不确定性集基于运行反馈自适应演变，而不是保持固定不变，从而增强了对网络攻击、极端天气和电网停电的恢复能力。采用深度Q学习和策略梯度方法不断优化调度策略，确保在非平稳环境中的鲁棒性能。对一个配备3兆瓦电解槽和2兆瓦燃料电池的5兆瓦光伏 - 氢微电网进行的案例研究证明了所提出框架的实际有效性。该模型纳入了全年的实际历史太阳能数据、随机需求和价格信号。在诸如48小时电网故障、基于信号的网络攻击以及40%光伏输出削减等场景下进行压力测试，结果显示出显著收益：运营成本降低了9.3%，在不利条件下恢复能力得分提高了20%以上，收敛速度相对于经典优化提高了42%。这些改进不仅反映了增强的适应性和计算效率，还体现了实时学习在弹性能源调度中的实际可行性。重要的是，“量子启发式”一词指的是模拟量子原理（如概率推理和解决方案多样性）的经典算法，而不使用量子硬件。通过统一量子启发式建模、分布鲁棒性和强化学习，该框架为下一代基于氢的微电网运营提供了一种可扩展且自适应的解决方案。
