Abstract
Greenhouse gases, particularly CO(2) and CH(4), are key contributors to climate change and global warming. Consequently, effective management and reduction of these emissions, especially in subsurface storage applications, are crucial. Adsorption presents a promising strategy for mitigating CO(2) and CH(4) emissions in the energy sector, particularly in the storage and utilization of fossil fuel resources, thereby minimizing the environmental impact of their extraction and consumption. In this study, the adsorption behavior of CO(2) and CH(4) in tight reservoirs is examined using experimental data and advanced machine learning (ML) techniques. The dataset incorporates key variables such as temperature, pressure, rock type, total organic carbon (TOC), moisture content, and the CO(2) fraction in the injected gas. Various ML models were employed to predict gas adsorption capacity, with CatBoost and Extra Trees demonstrating high predictive performance. The CatBoost model achieved superior results, with R² values of 0.9989 for CO₂ and 0.9965 for CH₄, along with low RMSE and MAE values, indicating strong stability and accuracy across all metrics. Sensitivity analysis identified pressure as the most influential factor, followed by TOC and CO(2) percentage, while temperature had a restrictive effect on adsorption. Secondary variables, such as rock type and moisture content, also contributed, though to a lesser extent. Graphical analyses further validated the high accuracy of the ML models, particularly CatBoost and Extra Trees. The findings underscore the effectiveness of ML approaches and optimized hyperparameter tuning in enhancing the prediction of gas adsorption capacity, thereby improving the design of gas injection and storage processes. This research provides valuable insights for optimizing gas composition and operational parameters in storage applications, serving as a foundation for future studies in gas sequestration and reservoir engineering.