Reproducibility package for physics-constrained reinforcement learning and many-objective design of an Ordos green-hydrogen supply chain

This dataset provides the reproducibility package for the associated Journal of Energy Storage manuscript on physics-constrained reinforcement learning and many-objective design of an Ordos green-hydrogen supply chain. The archive contains source code, frozen configuration files, raw and processed inputs, trained PPO model artifacts, benchmark and optimization outputs, figure-source tables, tests, manuscript sources, and audit reports. Open-Meteo weather data are API-derived. Price, carbon, demand, and grid-interruption profiles are modeled scenarios rather than observed Ordos market, marginal-emission, industrial-demand, or outage records. The trained models are included to reproduce the reported evaluation and figures without repeating stochastic training. Re-training may vary slightly across hardware and software environments.本数据集为发表于《Journal of Energy Storage》的关联论文提供可复现性套件，该论文聚焦物理约束强化学习（physics-constrained reinforcement learning）与鄂尔多斯绿色氢能供应链的多目标设计。本归档文件包含源代码、固化配置文件、原始与预处理输入数据、训练完成的近端策略优化（Proximal Policy Optimization，PPO）模型工件、基准测试与优化输出结果、配图源数据表、测试脚本、论文原稿及审计报告。 Open-Meteo气象数据通过API接口获取生成。电价、碳排放、需求与电网中断曲线均为建模生成的场景数据，而非鄂尔多斯地区实测得到的市场、边际排放、工业需求或停电记录。本数据集包含训练完成的模型，可无需重复开展随机训练即复现论文中报告的评估结果与配图。不同软硬件环境下的重新训练结果可能存在小幅差异。