Water holdup prediction of oil-water two-phase flow in horizontal well using a 12-probe conductance array

This paper presents a water holdup prediction method for oil-water two-phase flows in horizontal well using a circular 12-probe conductance array. By carrying out experiments in a 16m-long simulator of horizontal well with an inner diameter of 125mm, the analog voltage data under 220 different flow conditions was acquired and then divided to act as training set and testing set respectively. A machine learning method, namely gradient boosting regressor (GBR), was employed to establish the relationship between the raw data and the references of water holdup in the training set. Comparisons between GBR and popular SVR (support vector regressor) were made to demonstrate the validity and superiority of GBR in water holdup prediction. The prediction accuracy can reach 99.51%. Finally, the prediction results by a bicircular 24-probe conductance array were taken as reference to show the prediction accuracy by the circular 12-probe array which is simpler in structure and more reliable when working in underground wells. It can be concluded that the prediction accuracy by the circular one was no worse than the bicircular one.