Answers are grounded in this paper and cite their sources.
Carbon dioxide foam (CO 2 foam) injection is of great importance in the oil and gas industry, including enhanced oil recovery (EOR), drilling operations, etc. In addition, CO 2 foam injection is an efficient carbon dioxide (CO 2 ) storage technology due to its favorable industrial and environmental benefits. Therefore, knowledge of foam rheology behavior is essential to the optimal performance of these operations. Several laboratory tests must be conducted to investigate foams and determine the optimal foam composition. However, it is essential to minimize the expensive and time-consuming experiments required. Machine-learning (ML) algorithms for predicting the apparent viscosity of CO 2 foam provide a cost-effective and efficient substitute for lab experiments. The purpose of this study is to accurately forecast the apparent viscosity of CO 2 foam by applying machine learning (ML) techniques to experimental data to correlate influencing variables to foam rheology. The apparent viscosity of CO 2 foam is estimated using a set of robust ML methods, including two white-box approaches (genetic programming (GP) and group method of data handling (GMDH)) and four black-box approaches (categorical boosting (CatBoost), light gradient boosting machine (LightGBM), adaptive boosting (AdaBoost), and extreme gradient boosting (XGBoost)). A database including 105 experimental data points is considered to develop the algorithms. The target variable in this research is the apparent viscosity (cp), whereas the input variables include the permeability (D), temperature ( ° C), pressure (MPa), salinity (wt.%), concentration of surfactant (wt.%), Darcy velocity (ft/day), and foam quality (%). The CatBoost algorithm emerged as the most precise, with a determination coefficient (R 2 ) of 0.9985 and a total root mean square error (RMSE) of 104.48. Sensitivity analysis is performed to show the impact of input variables on the model output. In this regard, Darcy velocity is identified as the most effective factor on the output value. Finally, the leverage approach is used to confirm the created algorithms' precision and dependability.
No takes yet. Share an insight, caveat, or question.
Ghalenavi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: