Machine-learning based prediction of hydrogen/methane mixture solubility in brine.

Area of Science:

• Geochemistry
• Chemical Engineering
• Data Science

Background:

• Underground hydrogen storage (UHS) often involves pre-existing methane reservoirs, leading to hydrogen/methane mixtures.
• Accurate solubility data for these mixtures in brine is essential for safe and efficient UHS operations.
• Laboratory measurements are challenging due to the corrosive and flammable nature of the gases.

Purpose of the Study:

• To develop accurate data-driven intelligent models for predicting hydrogen/methane mixture solubility in brine.
• To utilize hybrid machine learning approaches optimized with metaheuristic algorithms.
• To investigate the influence of pressure, temperature, mixture composition, and brine salinity on solubility.

Main Methods:

• Development of hybrid models combining Adaptive Neuro-Fuzzy Inference System (ANFIS) and Least Squares Support Vector Machine (LSSVM).
• Optimization of LSSVM models using Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Coupled Simulated Annealing (CSA).
• Validation using laboratory data and evaluation of model performance through statistical metrics (AARE%, MSE, R-squared).

Main Results:

• The developed models demonstrated high accuracy in predicting hydrogen/methane mixture solubility.
• Sensitivity analysis revealed pressure and hydrogen mole fraction as key influencing factors.
• LSSVM-GA and LSSVM-CSA models exhibited superior performance with the lowest AARE% and MSE, and highest R-squared values.

Conclusions:

• Machine learning, particularly LSSVM-GA and LSSVM-CSA, provides a reliable and efficient method for predicting hydrogen solubility in brine for UHS.
• These models can support the development of intelligent, cost-effective, and secure underground hydrogen storage technologies.
• The findings contribute to de-risking hydrogen storage operations and advancing the hydrogen economy.