AI 中文总结
研究针对氢能社区微电网运行的复杂问题，提出基于近端策略优化（PPO）的能量管理系统，经澳大利亚某社区数据评估，该系统在不同场景下展现出在电池 - 氢协同调度方面的潜力及对多种因素的敏感性。
AI 中文摘要
氢能社区微电网能提高可再生能源利用和本地弹性，但运行受间歇性发电、不确定居民需求、动态电价及电池和储氢耦合动态影响。本文为并网社区微电网开发了基于近端策略优化（PPO）的能量管理系统（EMS），该系统集成了光伏、风力发电、电池储能、电解槽、储氢、燃料电池和柴油备用。EMS被公式化为一个马尔可夫决策过程，有11维状态和三个连续控制动作。通过对澳大利亚罗克汉普顿一个1000户家庭社区的8760小时观测数据进行评估。正常运行场景下，学习到的策略产生了195,690.67澳元的年净运营收入，可再生能源占比91.2%，碳强度0.085千克二氧化碳/千瓦时，负荷满意度99.77%。当每小时停电概率从1%增加到5%时，该策略保留了169,892.21澳元的净运营收入，并满足了98.79%的需求，电池放电增加413%，柴油发电量增加429%。结果证明了PPO在电池 - 氢协同调度方面的潜力，同时也突出了对可再生能源分布变化、训练稳定性和评估边界选择的敏感性。
英文摘要
Hydrogen-enabled community microgrids require coordinated control of intermittent renewable generation and coupled battery and hydrogen storage. This paper presents a feasibility-aware proximal policy optimization (PPO) energy management system for a grid-connected microgrid comprising photovoltaic and wind generation, battery storage, an electrolyzer, a hydrogen tank, a fuel cell, and diesel backup. Raw continuous actions are projected onto the feasible operating set before evaluating the hourly power balance, ensuring operationally valid dispatch. The proof-of-concept study uses 8,760 hourly observations for a 1,000-household community in Rockhampton, Australia. The same annual chronology and one random seed were used for training and evaluation. Under a 1 percent independent hourly grid outage probability, the system achieved an annual net operating cash balance of AUD 195,690.67, load satisfaction of 99.77 percent, and a gross renewable share of 91.2 percent. After removing duplicated hydrogen electricity emissions, annual emissions were 1.342 kt CO2, equivalent to 0.328 kg CO2 per kWh of served demand and 0.087 kg CO2 per kWh of export-inclusive delivered energy. At a 5 percent outage probability, the cash balance decreased to AUD 169,892.21 and load satisfaction fell to 98.79 percent. Battery discharge and diesel generation increased more than fuel cell output. The results demonstrate feasible dispatch for the studied chronology, but broader validation requires unseen testing, multiple random seeds, benchmark controllers, export limits, and sustained outage scenarios.
