Investigating the Use of Machine Learning Models for the Prediction of Pressure Gradient and Flow Regimes in Multiphase Flow in Horizontal Pipes

During multiphase flow, there is a variation of the physical distribution of the phases within the conduit leading to different flow regimes and consequently variation in the pressure gradient along with the flow regime, hence flow parameter is of vital importance in the prediction of flow regime and pressure gradient in multiphase flow. Analytical solutions and empirical correlations have been developed to predict the flow regime and the pressure gradient respectively. However, in this study, we seek to use supervised machine learning to make predictions taking parameters such as relative phase volume, bulk fluid flow rates, individual phase flow rates, conduit diameters, inclination, phase densities and temperature as input to the model. The data representing these parameters can be regularly updated to reflect the flow conditions in the well. The flow is composed of water, oil and air at different temperatures. The machine learning models used are Logistic Regression, Decision Trees and Principal Component Analysis. The first two is supervised and are tuned for accuracy dependent on pressure gauge readings while the third seeks to determine the parameters of greatest influence on the predicted output, the flow regime and pressure gradient. The model is constrained to learning and making predictions for fluid production through the tubing only. The trained model shows promise for application in the industry as it allows for automation of systems used to control flow and affords a more comprehensive approach to mitigating flow problems in pipeline systems and flow systems in oilfields.