PREDICTION OF DIFFUSION COEFFICIENT OF CARBON DIOXIDE USING ADVANCED MACHINE LEARNING MODEL IN BOTH BRINE AND HYDROCARBONS
Loading...
Date
2025-04-14
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Nazarbayev University School of Mining and Geosciences
Abstract
The diffusion coefficient (DC) of carbon dioxide (CO₂) in both brine and hydrocarbon play a critical role in geological carbon sequestration and CO₂-enhanced oil recovery (EOR), governing mass transfer efficiency and subsurface storage capacity. This study used three advanced machine learning (ML) algorithms — Random Forest (RF), Gradient Boost Regressor (GBR), and Extreme Gradient Boosting (XGBoost) — for the prediction of CO₂ diffusion coefficient using a dataset of 176 experimental and simulation data spanning pressures (0.10–30.00 MPa), temperatures (286.15–398.00 K), salinities (0.00–6.76 mol/L), and DC values (0.13–4.50 × 10⁻⁹ m²/s) utilized for the storage purpose in brine. The dataset was divided into 80% training and 20% testing sets to evaluate model generalizability. Performance metrics revealed RF as the most robust model, achieving an R² of 0.95, RMSE of 0.03, and MAE of 0.11 on test data, outperforming GBR (test R²: 0.925) and XGBoost (test R²: 0.91). Feature importance analysis identified temperature as the dominant predictor of diffusion coefficient, followed by salinity and pressure. A parallel investigation focusing on CO₂-EOR in hydrocarbon systems demonstrated RF adaptability, achieving R² values of 0.95 (training) and 0.92 (testing) using temperature (292.65–473.15 K), pressure (1.72–8 MPa), and API gravity (8.517–13.97) as input parameters. A dataset of 313 points was used for the hydrocarbon case. Contrastingly, tornado chart analysis in this context highlighted pressure as the most influential parameter, followed by temperature and API, suggesting context dependent variable significance. These findings establish ML frameworks as powerful tools for optimizing CO₂ injection strategies and storage security, with RF emerging as a versatile model for diverse subsurface conditions. The study underscores the potential of data-driven approaches to replace costly experimental methods while providing actionable insights for industrial carbon
management.
Description
Keywords
machine, earning, diffusion, CO2, hydrocarbon, brine, Type of access: Embargo
Citation
Khan, Q.(2025). Prediction of diffusion Coefficient of carbon dioxide using Advanced Machine Learning Model in both brine and hydrocarbons. Nazarbayev University School of Mining and Geosciences