Novel intelligent models for prediction of hydrogen diffusion coefficient in brine using experimental and molecular dynamics simulation data: Implications for underground hydrogen storage in geological formations
Piroozi G. Kouhi M.M. Shafiei A.
15 May 2025Elsevier Ltd
Journal of Energy Storage
2025#118
Saline aquifers are considered prime candidates for geological storage of H2 because of their relative abundance and notable storage capacity, worldwide. The typical formation water in other geological formations suitable for underground H2 storage is also brine. Diffusion coefficient of H2 in brine is vital for design and modeling of underground H2 storage operations. There is a lack of accurate and reliable models for accurate prediction of H2 diffusion coefficient in brine in the literature. Diffusion coefficient of H2 can be determined in laboratory. However, such experiments are usually time-consuming, costly, and suffer from safety issues because of highly flammable nature of H2 gas and high pressure-high temperature experiments involved. Hence, development of accurate, quick, and robust models for a wide range of pressure, temperature, and salinity is inevitable. To fill this gap, we developed and introduced novel intelligent models for the first time for accurate prediction of H2 diffusion coefficient in H2-brine/pure water system for temperatures of up to 973 K and pressures of up to 999 atm using four AI algorithms of Random Forest (RF), Decision Tree (DT), Adaboost, Catboost, along with linear regression (LR) were used. The dataset consists of 238 experimental and molecular dynamics simulation data points collected from the literature. The input features include pressure, temperature, and salinity. Shapley Additive explanations or SHAP analysis indicated that temperature and salinity have a positive and negative effect on the output parameter, respectively. The data modeling results showed that the Catboost model performed better than the other models with coefficient determination (R2), root mean square error (RMSE), and mean absolute error (MAE) of 0.9961, 2.987, and 1.46, respectively. The developed models can be ranked based on their accuracy and performance as the following: Catboost > DT > Adaboost > RF > LR. The developed novel models can serve as a useful tool for accurate prediction of H2 diffusion coefficient in brine/pure water in a wide range of operational conditions during H2 storage in geological formations.
Adaboost , Catboost , H2 diffusion coefficient , Machine learning , Saline aquifers , Underground H2 storage
Text of the article Перейти на текст статьи
Petroleum Engineering Program, School of Mining & Geosciences, Nazarbayev University, Astana, 010000, Kazakhstan
Petroleum Engineering Program
10 лет помогаем публиковать статьи Международный издатель
Книга Публикация научной статьи Волощук 2026 Book Publication of a scientific article 2026