The Application of Artificial Neural Networks in Determination of Bubble Point Pressure for Iranian Crude Oils

Abstract

Reservoir fluid characterization is an important issue in reservoir and production engineering calculations. Accurate determination of bubble point pressure is of major importance since it affects phase behavior of crude, which is indeed influential in further upstream and downstream computations. Several correlations have been proposed in the recent years to predict fluid properties using linear or non-linear regression and graphical techniques. In this study, artificial neural network is applied to predict bubble point pressure from reservoir temperature, solution gas oil ratio, oil API gravity, and gas specific gravity. A predictor model is developed based on 157 PVT data sets from southwest Iranian oil fields. Investigations of different network architectures show that a network with two hidden layers of six and three neurons has the best efficiency. Predictions of the developed neural network model are compared to empirical correlations. Results show that new model gives highest correlation coefficient and lowest average absolute relative error in estimation of bubble point pressure.