Machine Learning–Based Estimation of Hydrogen Solubility in Brine for Underground Storage in Saline Aquifers

Abstract

Saline aquifers are considered among the most attractive porous media systems for underground hydrogen storage (UHS) because of their wide availability and the considerable capacity of storage. The successful implementation of UHS in saline aquifers depends on many vital factors and parameters. Among these factors, the solubility of hydrogen (H₂) in brine remains a relevant consideration, particularly due to its influence on potential bio-geochemical reactions that may occur within underground formations. Given the significant expense and time demands associated with experimental methods for determining hydrogen solubility in brine, there is a growing need for a reliable and low-cost alternative capable of delivering accurate predictions. In this research, a suite of robust machine learning (ML) schemes, including multilayer perceptron (MLP), genetic programming (GP), and the group method of data handling (GMDH), is employed to construct predictive models for hydrogen solubility in brine, specifically under challenging high-pressure and high-temperature scenarios. The obtained results demonstrated the promising performance of the newly suggested ML-based paradigms. MLP optimized with Levenberg–Marquardt (MLP-LMA) yielded the best statistical metrics, including an R² of 0.9991 and an average absolute relative error (AARE) of 0.9417%. The findings of this study are important because they demonstrate that ML-based approaches embodied in intelligent paradigms are accurate and efficient and therefore have potential for use in reservoir simulators to assess dissolution processes associated with UHS in porous media.