Prediction of split-phase flow of low-velocity oil-water two-phase flow based on PLS-SVR algorithm

Abstract

In the middle and late stages of water flooding development in oil fields, oil wells will have the characteristics of low production fluid and high water cut content, making traditional production profile combination instruments unable to meet field needs. In order to accurately obtain the flow rate of each phase of the oil-water two-phase flow under the condition of low velocity, this paper proposes a new production profile logging method that combines an ultrasonic Doppler logging tool and a combined annulus logging tool. The measurement characteristics are verified by simulation experiments. Through frequency spectrum transformation of the original ultrasonic logging data, the power spectrum curve of the oil-water two-phase flow is obtained, the oil frequency, oil amplitude and other parameters are obtained by Gaussian fitting, and the distance analysis of the obtained experimental data is performed to obtain the correlation between the variables. Finally, the PLS-SVR algorithm is used to predict the oil and water flow. The mean absolute percentage error of oil flow is 7.96%, and the mean absolute percentage error of water flow is 9.4%. The prediction results show that the proposed low-velocity oil-water two-phase split-phase flow prediction model can meet actual needs, and has higher accuracy than support vector regression (SVR) and back propagation (BP) neural network prediction models, which can provide a new theoretical support for logging interpretation of production profiles.