Experimental Measurement and Accurate Prediction of Crude Oil Viscosity Utilizing Advanced Intelligent Approaches

In this study, experimental measurements and modeling investigations were performed to predict crude oil viscosity under a wide range of conditions. For this purpose, after measuring the viscosity of a considerable number of Iranian crude oils, three advanced intelligent models, including group method of data handling optimized by genetic algorithm, artificial neural network and Gaussian process regression were developed to estimate saturated and under-saturated oil viscosity by considering crude oil API, solution gas oil ratio, bubble point pressure, molecular weight and specific gravity of C12+ fraction, mole percent of C?11com-ponents, temperature and pressure as input parameters. To assess the ability of the proposed intelligent approaches, a wide variety of statistical and graphical error analyses were applied. The results demonstrated that the Gaussian process regression model with average absolute relative errors of 0.18 and 0.07% for saturated and under-saturated oil, respectively, had the best performance in viscosity prediction under different circumstances. Also, the findings of the Leverage technique, which was implemented for detection of suspected data, indicated the reliability of all measured data. Moreover, the results of sensitivity analysis showed that API, pressure and temperature had the greatest effect on oil viscosity in both saturated and under-saturated conditions.

Automated summary

In this study, experimental measurements and modeling investigations were conducted to predict crude oil viscosity under various conditions. Three advanced intelligent models were developed to estimate saturated and under-saturated oil viscosity using input parameters such as crude oil API, solution gas oil ratio, bubble point pressure, molecular weight, specific gravity of C12+ fraction, mole percent of C?11components, temperature, and pressure. The results showed that the Gaussian process regression model had the best performance in viscosity prediction, with average absolute relative errors of 0.18% and 0.07% for saturated and under-saturated oil, respectively. The findings of the Leverage technique and sensitivity analysis further supported the reliability and importance of the study.