Аннотация
Abstract Underground hydrogen storage (UHS) in deep saline aquifers is gaining recognition as an efficient long-duration energy solution in the context of the low carbon energy transformation process. The hydrogen and brine interaction, specifically in the context of the presence of methane gas, is crucial to reduce the dissolution losses taking place to some extent in the operational efficiency of the process. The presence of methane, also common in depleted gas fields, affects the solubility of the hydrogen gas due to the intersecting gas equilibrium phenomenon. The current work introduces another machine learning approach to predict the solubility of H2-CH₄ mixtures in brine conditions based on pressure, temperature, and salinity variables. A number of machine learning algorithms, including Decision Tree, Gradient Boosting Regressor, Extreme Gradient Boosting, are developed, evaluated, and validated with the help of the available experimental data compiled from the existing literature. The hyperparameters of these machine learning algorithms are optimized by utilizing the k-fold cross-validation search method with the help of the exhaustive search technique. The result validation is carried out on the basis of error estimation techniques, including the value of the Coefficient of Determination, Root Mean Square Error, AIC, and BIC. The proposed model shows how the solubility of hydrogen in the presence of methane in brine is dominantly affected by pressure, followed by the effects of temperature, and then salinity. The comparison study between the different models used indicates that the GBR method is the best-performing one, with R² values of 99.7% and 98.9% on the training and testing datasets, respectively. The proposed model predicts the hydrogen solubility in the subsurface based on the influences of pressure, temperature, salinity, and the components of the gas mixture. The experimental study on the hydrogen solubility data, which was used to train the proposed models, consists of 332 data points, with pressure values below 50 MPa, temperature values between 298.05 K and 423.15 K, with the gas mixtures composed of H2 and CH₄ with varying proportions from 10% to 100% H2. The proposed model was validated with successful outcomes on unseen data. The trend analysis of the proposed model is correct, indicating that with the increment of pressure, the solubility increases due to the compression of the gas, which readily dissolves in the brine, followed by the moderate effect of temperature, wherein the increase in temperature leads to the decrease in the hydrogen solubility in the brine due to the reduced density of the gas. The model also captures the effect of gas components, wherein the gas with lower proportions of H2 has higher solubility because CH₄ is more soluble in water compared to H2. By offering a fast, data-driven alternative to complex thermodynamic formulations and experimental work, this study contributes to the practical deployment of H2-CH₄ gas mixtures solubility prediction. The novel digital model will be helpful in both industry and academia.