Predicting Petroleum SARA Composition from Density, Sulfur Content, Flash Point, and Simulated Distillation Data Using Regression and Artificial Neural Network Techniques

Abstract

The saturate, aromatic, resin, and asphaltene content in petroleum (SARA composition) provides valuable information about the chemical nature of oils, oil compatibility, colloidal stability, fouling potential, and other important aspects in petroleum chemistry and processing. For that reason, SARA composition data are important for petroleum engineering research and practice. Unfortunately, the results of SARA composition measurements reported by diverse laboratories are frequently very dissimilar and the development of a method to assign SARA composition from oil bulk properties is a question that deserves attention. Petroleum fluids with great variability of SARA composition were employed in this study to model their SARA fraction contents from their density, flash point, sulfur content, and simulated distillation characteristics. Three data mining techniques: intercriteria analysis, regression, and artificial neural networks (ANNs) were applied. It was found that the ANN models predicted with higher accuracy the contents of resins and asphaltenes, whereas the non-linear regression model predicted most accurately the saturate fraction content but with an accuracy that was lower than that reported in the literature regarding uncertainty of measurement. The aromatic content was poorly predicted by all investigated techniques, although the prediction of aromatic content was within the uncertainty of measurement. The performed study suggests that as well as the investigated properties, additional characteristics need to be explored to account for complex petroleum chemistry in order to improve the accuracy of SARA composition prognosis.