Development of Artificial Intelligence Models for Prediction of Crude Oil Viscosity

Abstract

Crude oil viscosity is a significant parameter for the fluid flow in both porous media and pipe lines. Therefore, it has to be determined using highly accurate methods. Oil viscosity is usually predicted with the correlations obtained from the laboratory measured data. However, some of the presented correlations have very complicated assumptions which make them very difficult to apply in most of the case studies reported. On the other hand, simplified correlations companies the accuracy. The present work in this paper studies predictive capabilities of Artificial Intelligence (AI) to estimate the oil viscosity. Artificial Neural Network (ANN) models are proposed to predict the undersaturated, saturated and dead oil viscosity in Yemeni fields. A data set consisting 545 of laboratory measurements on oil samples was gathered from different oil fields in Yemen. 70% of the data points were used to train the proposed ANN models while the remaining data set was tested the model performance. The performance of the ANN methods was compared with some of the conventional correlations such as (Beal's correlation, Khan's correlation, Kartoatmodjo and Schmidt correlation, Vasquez-Begg's correlation, Chew and Connaly correlation, Beggs and Robinson correlation, Elsharqawy correlation and Glaso's correlation). The result of this study shows the superiority of the Artificial Neural Network (ANN) models over the current models for predicting oil viscosity from PVT data. The comparative results displayed that the proposed ANN models performed better with higher accuracy than those obtained with published correlations.