Integrating molecular dynamics and machine learning to predict hydrogen–brine interfacial tension for subsurface storage

Hydrogen energy has gained significant attention as part of the global transition to cleaner energy systems. Underground hydrogen storage (UHS) in depleted hydrocarbon reservoirs offers a promising option for large-scale, long-duration storage. A key parameter controlling UHS performance is the interfacial tension (IFT) between hydrogen and brine, which governs capillary trapping, injectivity, and storage security. In this study, molecular dynamics (MD) simulations were used to calculate IFT between hydrogen and CaCl 2 brines under pressures of 10–50 MPa, temperatures of 50–250 °C, and concentrations of 1–6 mol/kg. Results show that IFT decreases with temperature, increases with salinity, and is only weakly influenced by pressure, with simulated values ranging from 50.23 to 110.58 mN/m. To extend predictive capability beyond the simulated dataset, newly generated H 2 –CaCl 2 results were combined with previously published H 2 –NaCl data to train machine learning (ML) models. Among 24 algorithms tested, gradient boosting regression, multilayer perceptron, and XGBoost achieved the highest accuracy, with R 2 of test data up to 0.973 and average absolute relative error (AARE) below 2.6 %. This integrated MD–ML framework provides an efficient, data-driven approach for predicting IFT across diverse conditions, supporting the design and safe deployment of UHS systems. • MD simulations predicted IFT between H 2 and CaCl 2 brine at various conditions. • IFT decreases with temperature and increases with higher salt concentration. • ML models were trained using IFT datasets from this work and the literature. • ML models achieved R 2 up to 0.973 and AARE below 2.6 %.