Modeling of Carbon Dioxide Adsorption Capacity in Unconventional Reservoir using Machine Learning

Summary

Abstract

Carbon Capture, Utilization, and Storage (CCUS) has turned out to be a serious measure to the reduction of worldwide greenhouse gases.CO2-Enhanced Coalbed Methane (ECBM) recovery has dual advantage as it captures CO2 and increases methane yields. Conventional experiment techniques to measure adsorption isotherms are time-consuming and expensive. Present investigation gives a strong suggestion to forecast the capacity of CO2 adsorption using machine learning (ML) algorithm. This study includes the database of 1,101 coal samples representing different coalfields, considering the key physicochemical parameters including moisture content, ash yield, volatile matter, fixed carbon, vitrinite reflectance, and total organic carbon (TOC), pressure and temperature as input. The Random Forest (RF) model was developed using an 80:20 training–testing data split and optimized through 10-fold cross-validation combined with grid-based hyperparameter tuning. The optimal RF configuration consisted of 50 decision trees with a maximum tree depth of 30, where bootstrap aggregation was used so that each tree was trained on a randomly sampled subset of the training data. The small difference between training and testing R2 values (<0.03), together with low training errors (root mean Squared Error (RMSE) = 0.0268 mmol/g, mean absolute error (MAE) = 0.0143 mmol/g), indicates minimal overfitting and strong generalization capability. Finally predictive performance of the RF model evaluated on three unseen samples data. The model showed great predictive ability with Coefficient of Determination (R2) of more than 0.97 and RMSE less than 0.073 mmol/g with MAE (0.04). These findings confirm that the Random Forest algorithm is a reliable, computationally efficient, and cost-effective method for predicting CO2 adsorption capacity based on the physiochemical properties of coal. This forecasting model can significantly assist rock engineers and geologists in the preliminary evaluation and design of CO2 sequestration in coal seams.