Genetic algorithm based support vector machine regression for prediction of SARA analysis in crude oil samples using ATR-FTIR spectroscopy

In the current research, an analytical method was proposed for rapid quantitative determination of saturates, aromatics, resins and asphaltenes (SARA) fractions of crude oil samples. Rapid assessments of SARA analysis of crude oil samples are of substantial value in the oil industry. The conventional SARA analysis procedures were determined with the standards established by the American Society for Testing and Materials (ASTM). However, the standard test methods are time consuming, environmental nonfriendly, expensive, and require large amounts of the crude oil samples to be analyzed. Thus, it be would useful to approve some supportive approaches for rapid evaluation of the crude oils. The attenuated total reflection Fourier-transform infrared spectroscopy ATR-FTIR coupled with chemometric methods could be used as analytical method for crude oil analysis. A hybrid of genetic algorithm (GA) and support vector machine regression (SVM-R) model was applied to predict SARA analysis of crude oil samples from different Iranian oil field using ATR-FTIR spectroscopy. The result of GA-SVM-R model were compared with genetic algorithm-partial least square regression (GA-PLS-R) model. Correlation coefficient (R-2) and root mean square error (RMSE) for calibration and prediction of samples were also calculated, in order to evaluate the calibration models for each component of SARA analysis in crude oil samples. The performance of GA-SVM-R is found to be reliably superior, so that it can be successfully applied as an alternative approach for the quantitative determination of the SARA analysis of crude oil samples. (C) 2020 Elsevier B.V. All rights reserved.

Automated summary

A rapid quantitative determination method for SARA fractions of crude oil samples was proposed, utilizing ATR-FTIR spectroscopy coupled with chemometric methods. The GA-SVM-R model successfully predicted the SARA analysis of crude oil samples from different Iranian oil fields, showing superior performance compared to other models. This alternative approach is reliable for quantitative determination of SARA analysis in crude oil samples.