Modeling minimum miscibility pressure of pure/impure CO2-crude oil systems using adaptive boosting support vector regression: Application to gas injection processes

Abstract

Carbon dioxide based enhanced oil recovery (CO2-EOR) techniques have been of great interest due to CO2 effectiveness as an oil solvent under supercritical condition and potential application to the sequestration purposes of CO2. Appropriate design of these techniques such as CO2 gas injection process needs accurate knowledge of the minimum miscibility pressure (MMP) allowing to the injected gas to be miscible with the oil. In this paper, three intelligent models named group method of data handling (GMDH), adaptive boosting support vector regression (AdaBoost SVR) and multi-layer perceptron (MLP) were implemented for MMP prediction in systems oil - CO2 streams. To this end, a widespread databank covering extended range of reservoirs and thermodynamic conditions was considered. The data bank covers dead and live oils under reservoir temperatures up to 388.73 K, while the gathered real MMP values were ranged between 6.54 MPa and 31.30 MPa. Graphical and statistical assessment criteria were performed to examine the reliability of the implemented paradigms and to compare them with the most well-known models. The results of this study indicated that the established models outperform the existing ones. In addition, among the proposed models, AdaBoost SVR has the highest accuracy with an Average Absolute Percent Relative Error (AAPRE) of 3.09% and Root Mean Square Error (RMSE) of 0.9 MPa. All of the proposed models can be applied in gas injection processes to prevent overestimation and underestimation of MMP, leading to appropriate performance of gas injection.