Modeling approach for predicting PVT data

Abstract

Neural networks are tested successfully in so many fields as pattern recognition or intelligent classifier, prediction, and correlation development. Recently, artificial neural network has gained popularity in petroleum applications. In this paper, two directly connected neural networks are designed, using Matlab5.3, for PVT parameters determinations. The first neural network predicts the bubble point pressure values using the four following variables: reservoir temperature, APIo, relative gas density, and solution gas oil ratio. These estimated values directly used with the same four input variables in the second network to determine oil Formation Volume Factor (FVF). A comparison study between the designed Neural Network Model and other published correlations displayed excellent performance, smallest average absolute relative error, and highest correlation coefficient for the designed Networks among all other correlations.