Prediction of Hydrogen Solubility in Brine Using Machine Learning

Dilyara Sagandykova, Shams Kalam, Peyman Pourafshary, Masoud Riazi

2025

Closed access Metadata (CSL-JSON)
DOI 10.2118/230474-ms
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Abstract Deep saline aquifers are emerging as a viable option for underground hydrogen storage (UHS) to support low-carbon energy systems. A clear understanding of how hydrogen behaves when it comes into contact with brine is essential for such subsurface operations. In particular, reliably and rapidly predicting hydrogen solubility across different pressures, temperatures, and salinities is key to estimating potential dissolution losses. In this study, we present a novel ML predictive model for H2 solubility in brine in relation to pressure, salinity, and temperature. Some ML tools, including Decision Tree (DT), Gradient Boosting Regressor (GBR), and Extreme Gradient Boosting (XGBoost), were trained and tested with experimental data retrieved from the literature. Hyperparameter optimization was performed by applying k-fold cross-validation alongside an exhaustive grid search strategy. Model adequacy was established from an error metric including the coefficient of determination (R²), mean absolute percentage error (MAPE), root mean square error (RMSE), and the Akaike (AIC) and Bayesian (BIC) information criteria. The developed model shows that hydrogen solubility in brine is governed mainly by pressure, with temperature and salinity exerting secondary effects. Performance comparison among the tested algorithms revealed that GBR achieved the highest accuracy, reaching R² values exceeding 99 % for the training set and 95 % for the test set. Our model estimates hydrogen solubility under a wide range of subsurface conditions. Experimental solubility data used for training were derived from the literature and included pressures up to approximately 40 MPa, temperatures between 298 and 423 K, and salinity levels up to 5 mol/kg NaCl. The model was successfully validated against unseen data. A trend analysis confirmed the model's integrity, showing that hydrogen solubility increases significantly with pressure, while temperature exerts a minor influence. Salinity consistently reduces solubility via the salting-out effect, which is more pronounced at higher ionic strengths. These insights support the accurate prediction of hydrogen behavior in deep geological storage environments. By offering a fast, data-driven alternative to complex thermodynamic formulations and experimental work, this study contributes to the practical deployment of H2 solubility prediction. The novel digital model will be helpful in both industry and academia.