Modeling n-alkane solubility in supercritical CO2 via intelligent methods

Abstract

Abstract Injection of carbon dioxide is a familiar, cost-effective and influential technology of enhancing oil recovery whose application has been limited owing to the low n-alkane solubility in supercritical CO2. Thus, determining the amount of dissolved n-alkane in supercritical CO2 is of importance. Accordingly, in this study, least-squares support vector machine (LSSVM), tuned with two different optimizing algorithms, namely particle swarm optimization (PSO) and cross-validation-assisted Simplex algorithm (CV-Simplex), has been used for this simulation process. Based on the results, the predicted values for dissolved n-alkane mole fraction in supercritical CO2 by PSO–LSSVM model were quite in line with experimental data. Furthermore, the accuracy of these models was compared with Chrastil correlation. Absolute average relative error for PSO–LSSVM, CV-Simplex–LSSVM and Chrastil was calculated to be 3.88%, 13.49% and 18.22% for total dataset, respectively, which leaves PSO–LSSVM as the superior model with the highest accuracy. Finally, the statistical parameters of absolute average relative error, mean square error and determination coefficient equal to 3.88%, 0.0164 and 0.994 for total dataset, respectively, proved that PSO–LSSVM model is an efficient method that can predict n-alkane solubility in supercritical CO2 with high precision within 8.99–45.90 MPa pressure and 308.15–344.15 K temperature range.

Keywords

• CV-Simplex–LSSVM
• PSO–LSSVM
• n-alkane solubility
• Supercritical CO2