Reproducibility package for physics-constrained reinforcement learning and many-objective design of an Ordos green-hydrogen supply chain
Description
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.
Files
Steps to reproduce
After extracting the archive, install the Python dependencies listed in requirements-lock.txt and run the verification commands from the project root: python -m pytest -q python code/02_validate_inputs.py python code/09_audit_results.py The included trained PPO model artifacts can be used to reproduce the reported benchmark evaluations and figures without repeating stochastic training. Main scripts are numbered in the code directory from input construction through validation, training, benchmark evaluation, NSGA-III optimization, sensitivity/stress analysis, and figure generation.