Improving wettability estimation in carbonate formation using machine learning algorithms: Implications for underground hydrogen storage applications

• Published in: International journal of hydrogen energy Vol. 111; pp. 781 - 797
• Main Authors: Mwakipunda, Grant Charles, Ibrahim, AL-Wesabi, Kouassi, Allou Koffi Franck, Nafouanti, Mouigni Baraka, Yu, Long
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 20.03.2025
• Subjects: Carbonate formation; Contact angle; Underground hydrogen storage; Wettability; Wettability; Underground hydrogen storage; Contact angle; Carbonate formation
• ISSN: 0360-3199
• DOI: 10.1016/j.ijhydene.2025.02.342

The wettability of carbonate formations is a critical factor in evaluating underground hydrogen storage (UHS) applications, influencing fluid behavior and storage efficiency. This study investigates the estimation of contact angle, a key parameter in wettability assessment, using a novel hybrid model, Extreme Learning Machine - Mixed Effects Random Forest (ELM-MERF). The significance of this research lies in its ability to accurately predict contact angles, which are essential for optimizing UHS strategies in carbonate systems, where environmental factors such as pressure, temperature, and salinity play a significant role. We developed the ELM-MERF model to integrate the advantages of extreme learning machines with the robust predictive capabilities of random forests while accounting for mixed effects in the data. The dataset used in this study was sourced from the literature, including variables such as pressure (P), temperature (T), and salinity (S) as model inputs. We compared the performance of the ELM-MERF model to other established models, including EXtreme Gradient Boosting (XGBoost) and Random Forest (RF), using standard evaluation metrics. Key findings reveal that ELM-MERF significantly outperforms the other models. ELM-MERF achieved coefficient of determination (R2) of 0.9999 and 0.9996, root mean square error (RMSE) of 0.022 and 0.055, and mean absolute error (MAE) of 0.012 and 0.035 compared to XGBoost with R2 of 0.9952 and 0.9728, RMSE of 0.099 and 0.215, and MAE of 0.025 and 0.062, and RF with XGBoost with R2 of 0.9871 and 0.9508, RMSE of 0.177 and 0.522, and MAE of 0.054 and 0.099, during training and testing phase, respectively. Moreover, ELM-MERF offers computational efficiency, requiring only 30 s for processing, a substantial reduction in time compared to alternative models. This work is novel in its application of a hybrid machine learning approach to wettability estimation, offering a more accurate, efficient, and scalable solution for UHS applications in carbonate formations. These advancements extend beyond previous efforts by combining extreme learning and random forest techniques to account for mixed data effects, making it an invaluable tool for real-time analysis and large-scale UHS optimization. •ELM-MERF algorithm was developed for wettability estimation for UHS purpose.•ELM-MERF was compared with XGBoost, and RF.•ELM-MERF outperformed XGBoost, and RF.