Smart models for predicting under-saturated crude oil viscosity: a comparative study

Abstract

ABSTRACT In this study, radial basis function (RBF) and multilayer perceptron (MLP) neural networks were proposed for accurate prediction of under-saturated oil viscosity. To this end, more than 600 viscosity data were collected from various geological locations worldwide which cover oil API gravity from 6.5 (extra heavy crude oils) to 53 (very light crude oils), reservoir temperature from 300.15 to 445.15 K, and reservoir pressure from 1.68 to 105.52 MPa. Statistical and graphical comparison of the proposed models with other existing models indicates that the prediction accuracy and applicability extent of the suggested models are much better compared to the previously published ones by providing average absolute relative errors of 3.09% and 3.88% for MLP and RBF models, respectively.