Study on the prediction method of oil-water two-phase flow pattern and oil holdup

Oil-water flow often occurs during pipeline gathering and oil field production. Complex flow patterns and oil holdup changes cause technical difficulties in pipeline maintenance. Accurate and rapid prediction of the flow patterns distribution and oil holdup is significant. A basic database of oil and water flow patterns and oil holdup was constructed by collecting experimental data. The pipeline inclination angle ranged from 0 to 90 degrees, the apparent velocity of the oil phase ranged from 0.01 to 4.71 m/s, and the apparent velocity of the water phase ranged from 0.01 to 7.58 m/s. The characteristic parameters of the oil-water flow pattern and oil hold-up were selected, and an estimated model of liquid hold-up and oil-water flow pattern distribution was constructed by applying the deep learning method. The results show that the prediction accuracy of the established deep learning model for the oil-water flow pattern and oil holdup in the training set exceeded 95%. Predicting flow transition boundaries in both the horizontal and vertical directions is superior to traditional flow boundary maps that are based on experimental partitioning. The new method has a prediction accuracy of over 85% for oil holdup. Applying this model to the calculation of pressure drop in oil-water two-phase flow, the accuracy of pressure drop calculation reaches 95%, which can be extended to the application of oil-water pipeline transportation.