Machine Learning Forecasting of Hydrogen Migration in Porous Media

This chapter presents a hybrid framework combining physics-based simulation and machine learning (ML) techniques to predict hydrogen migration in porous media for underground hydrogen storage applications. Recognizing the critical role of parameters such as permeability, injection rate, vertical position, and porosity, a comprehensive parametric study was conducted using MATLAB Reservoir Simulation Toolbox (MRST) to generate high-fidelity simulation data. Machine learning models, particularly tree-based ensemble methods like gradient boosting, were trained on this data to predict breakthrough times with high accuracy (R 2 = 0.98) and dramatically reduced computational costs. Feature importance analysis revealed permeability as the dominant factor influencing hydrogen migration, followed by injection rate and vertical position. The ML models not only achieved excellent predictive performance but also provided physically interpretable insights, validating the dominant influence of buoyancy and non-linear permeability effects. This integration of simulation and ML offers a powerful tool for rapid, accurate design and optimization of underground hydrogen storage systems, bridging computational efficiency with rigorous physical fidelity.