Machine Learning-Powered Forecasting of Hydrogen Solubility in Brines for Geological Storage: The Role of Pressure, Temperature, Salinity, and Salt Type

Underground hydrogen storage (UHS) in geological formations, such as brine-saturated reservoirs, presents a promising large-scale storage option. Therefore, an accurate prediction of hydrogen solubility under varying thermodynamic and geochemical conditions is crucial. This study utilizes a comprehensive machine learning (ML) framework to predict hydrogen solubility in brine systems, integrating experimental data across a wide range of pressures (1–1000 atm), temperatures (273–589 K), salinities (0–40%), and 10 different brine-constituting salts. A range of ML and deep learning models, including gradient boosting with decision trees (GB+DT), convolutional neural networks (CNNs), and multilayer perceptron (MLP), were rigorously evaluated for their suitability in the accurate prediction of H2 solubility in brine. The GB+DT model achieved the highest predictive accuracy (R2 = 0.9943; RMSE = 0.000211), outperforming other models and effectively capturing complex nonlinear interactions. SHAP analysis identified pressure as the primary driver of H2 solubility, with salinity, temperature, and salt type exerting secondary effects, which is consistent with Henry’s law and the salting-out behavior. Robustness tests on outlier handling showed that percentile filtering and isolation forest improved predictive stability relative to alternative approaches. Overall, the ML framework delivers scalable, accurate, and interpretable solubility estimates that complement equation-of-state methods and provide actionable guidance for the design and optimization of UHS systems.