Development of a Universal Artificial Neural Network Model for Pressure Loss Estimation in Pipeline Systems; A comparative Study

This study aims to develop a universal artificial neural network model for estimating pressure drop at pipelines under multiphase flow conditions. Three phase flow data have been collected from different geographical locations; especially from Middle-Eastern fields in order to construct, test, and validate the model. The data covered a wide range of variables such as oil rate (up to 25000 STB/D), water cut (up to 60%), angles of inclination (from -80 to 210), pipe length up to 26.0 km and pressure drop (from 10 to 250 psi). the model has been generated using the Back-propagation technique with Bayesian Regularization training algorithm for predicting pressure drop in pipelines under various angles of inclination. A total number of data points consists of 335 sets has been used for generating, validating, and testing the ANN model. A model performance has been evaluated against the best empirical correlations and mechanistic models (Xiao et al, Gomez et al, and Beggs & Brill). A series of statistical and graphical analysis were conducted to show the superiority of the generated model. A thorough literature review is also conducted to check the applicability of the existed correlations and mechanistic models and their drawbacks compared with the new proposed ANN model. The new developed model outperforms all the investigated models with correlation coefficient reaches 99.91%.