Prediction of °API Values of Crude Oils by Use of Saturates/Aromatics/Resins/Asphaltenes Analysis: Computational-Intelligence-Based Models

Summary The °API value is an important physicochemical characteristic of crude oils often used in determining their properties and quality. There exist models—predominantly linear ones—for predicting the °API magnitude from the molecular composition of a crude oil. This approach is tedious and time-consuming because it requires quantitative determination of numerous crude-oil components. Usually, the hydrocarbons present in a crude oil are grouped according to their molecular average structures into saturates, aromatics, resins, and asphaltenes (SARA) fractions. An °API-value prediction model dependent on these four fractions is relatively easier to develop, although this approach has been rarely used. A rigorous scrutiny suggests that some of the dependencies between the individual SARA fractions and the corresponding °API value could be nonlinear. Accordingly, in this study, SARA-fraction-based nonlinear models have been developed for the prediction of values using three computational-intelligence (CI) formalisms: genetic programming (GP), artificial-neural networks (ANNs), and support-vector regression (SVR). The SARA analyses and °API values of 403 crude-oil samples covering wide ranges have been used in developing these models. A comparison of the CI-based models with an existing linear model indicates that all the former class of models possess a significantly better °API-value prediction and generalization performance than those exhibited by the linear model. Also, the SVR-based model has been found to be the most accurate °API-value predictor. Because of their better prediction accuracy, CI-based models can be gainfully used to predict °API values of crude oils.