Accurate Prediction of Pressure Drop in Two-Phase Vertical Flow Systems using Artificial Intelligence

Abstract

Abstract One of the significant parameters affecting flow rate in oil production wells is the pressure drop between the well bottom-hole and tubing head. The pressure drop calculation in two-phase flow systems is very complicated due to the variations in gas and liquid flow rates across the two-phase flow stream. As the pressure of crude decreases while climbing a well tubular, more gas comes out of solution. This gradual increase in gas volumes leads to the reduction of liquid slip velocity and creating new flow patterns that are not only different in shape, but also complicated in pressure drop calculations. To overcome this difficulty in calculating pressure drop in two-phase flow systems, scientists came up with two main approaches: flow correlations and mechanistic models. These two approaches are applicable within certain conditions and their accuracy in pressure drop prediction degrades outside their design boundary ranges. The raising popularity of Artificial Intelligence (AI) techniques during the past two decades proved that AI can be an alternative solution to many of the complicated problems where physics and classic statistics fail to provide satisfactory solutions. These techniques applied in different upstream fields have provided fast, robust and reliable numerical models in a variety of areas, e.g., geological modeling, reservoir engineering, petrophysics and well testing. This paper describes the utilization of Fuzzy Logic, which is one of the famous AI techniques, in predicting flowing bottom-hole pressure in oil producer wells. Real well testing data from the Middle East were used in constructing the Fuzzy Logic model. After training the model using 596 well testing data samples, it was successfully able to predict the flowing bottom-hole pressure at 199 well testing samples with an average absolute error of 4.9%. A comparison analysis was conducted to evaluate multiple flow correlation in predicting flowing bottom-hole pressure and compare their results with the developed Adaptive Neuro-Fuzzy Inference System (ANFIS) model.