A Data-Driven Model for Predicting Competitive Adsorption of CH4 and CO2 in Shallow Coal Seams During CO2-Enhanced Coalbed Methane Recovery

Abstract CO2-Enhanced Coalbed Methane (CO2-ECBM) has gained a lot of attention in recent times. Efficient methane recovery from Coalbed Methane (CBM) reservoirs depend on understanding the complex interactions of CH4 and CO2. During CO2-ECBM, injected CO2 adsorbs onto the coal surface and displaces CH4 which enhances the methane production. This work proposes a novel machine learning (ML) tool for predicting the amount of adsorbed CH4 and CO2 during CO2-ECBM process at various conditions. An experimental dataset was gathered from the literature with various operational conditions, such as pressure, injected gas composition, and coal properties. Dataset pre-processing and analytics were initially conducted to ensure optimal inputs for ML modeling. Several ML tools, including Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Decision Tree (DT), were employed to predict gas adsorption. The hyperparameters were optimized using K-fold cross-validation, integrated with the grid search approach. The performance of ML models was evaluated using the following error metrics: coefficient of determination (R2), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Akaike Information Criterion (AIC), and Bayesian Information Criterion (BIC). The total dataset was divided into an 80:20 ratio for model training and testing. Results show that XGBoost outperforms other models, showing R2 more than 90%. The proposed model can predict gas adsorption (0 – 905.36 scf/ton) as a function of pressure (0 – 1507.33 psi), temperature (39.9 °C – 44.3 °C), ash content (19.94 – 41.17 %), fuel ratio (2.65 – 3.88), vitrinite (11.4 – 51.28 vol. %), liptinite (0 – 52.2 vol. %), inertinite (1.36 – 76.8 vol. %), CO2 concentration (0 – 100 %) and gas type (1 = CH4 and 2 = CO2). The trained model was tested using a blind dataset to confirm its robustness. In addition, the proposed model was also compared to another set of unseen experimental datasets. The model reflects that CO2 has a stronger adsorption affinity toward coal than CH4, which may help enhance the methane recovery from CBM reservoirs. This approach enhances our understanding of complex gas-coal interactions in CO2-ECBM processes. This study presents a novel model for analyzing the competitive adsorption of CH4 and CO2 without the need for labor-intensive simulations or laboratory experiments. The proposed XGBoost model accurately predicts gas adsorption in CBM reservoirs, significantly reducing computation time. The findings highlight the potential use of ML to optimize CO2-ECBM strategies by improving methane recovery and reducing environmental impact by effective carbon sequestration.