Machine learning models for the prediction of hydrogen solubility in aqueous systems

Area of Science:

• Geosciences
• Chemical Engineering
• Data Science

Background:

• Hydrogen storage is crucial for reducing CO2 emissions, especially in the oil and gas sector.
• Hydrogen solubility in saline aquifers presents a significant challenge to storage efficiency and reservoir stability.
• Understanding hydrogen dissolution dynamics is vital for optimizing subsurface storage technologies.

Purpose of the Study:

• To predict hydrogen solubility in saline aquifers using machine learning algorithms.
• To identify key factors influencing hydrogen solubility, including pressure, temperature, and salinity.
• To evaluate the performance of various machine learning models for this prediction task.

Main Methods:

• Utilized machine learning algorithms: Bayesian inference, linear regression, random forest (RF), artificial neural networks (ANN), support vector machines (SVM), and least squares boosting (LSBoost).
• Trained models on experimental data and numerical simulations covering a wide range of thermodynamic conditions.
• Applied techniques to identify nonlinear relationships between solubility and influencing parameters.

Main Results:

• The random forest (RF) model demonstrated superior performance, achieving an R² of 0.9810 for test data and 0.9915 for training data.
• Root Mean Square Error (RMSE) values were low, with 0.048 for test data and 0.032 for training data.
• Machine learning models accurately predicted hydrogen solubility, which is highly sensitive to pressure, temperature, and salinity.

Conclusions:

• Machine learning offers a powerful approach to accurately predict hydrogen solubility in saline aquifers.
• Optimized hydrogen storage and reservoir management can be achieved through these predictive capabilities.
• This research supports the advancement of hydrogen as a clean energy carrier and industrial precursor.