PREDICTION OF HORIZONTAL OIL-WATER FLOW PRESSURE GRADIENT USING ARTIFICIAL INTELLIGENCE TECHNIQUES

Abstract

In oil-water flow, the two-fluid and the homogeneous models are commonly used to predict the pressure gradient. However, these models fail in many cases to predict the pressure gradient, especially in dual continuous flow. In this work, an artificial neural network (ANN) model with five inputs—oil and water superficial velocities, pipe diameter, pipe roughness, and oil viscosity—was developed to predict the pressure gradient of horizontal oil-water flow based on a databank of around 765 measurements collected from the open literature. Statistical analysis showed that the ANN model has an average error of 0.30%, average absolute error of 2.9%, and standard deviation of 7.6%. A comparison with the two-fluid model, the homogeneous model, and the Al-Wahaibi (2012) correlation showed that the ANN model better predicts the pressure gradient data over a wide range of superficial oil (U so = 0.05–3.0 m/s) and water velocities (U sw = 0.05–2.7 m/s), oil viscosity values (1–35 cp), pipe diameters (14–82.8 mm), and different pipe materials.