A research on a GA-BP neural network based model for predicting patterns of oil-water two-phase flow in horizontal wells

Abstract

At present, unconventional oil reservoirs are usually exploited by means of horizontal wells. However, due to the influence of multiple factors such as well deviation, oil flowrate and water flowrate, the distribution (flow pattern) of oil and water in a wellbore is complex. In addition, an accuracy of flow pattern identification is also vital for accurately interpreting the oil-water two-phase production profile of a horizontal well. Therefore, the research on flow pattern prediction of wellbores is of great application significance. Firstly, at normal pressure and temperature, based on the multiphase flow simulation experiment device, a transparent glass wellbore with a diameter similar to the actual downhole well was used to carry out oil-water two-phase simulation experiments with different well deviations, different flowrates and different water cuts, while visual observation and whole-process HD video recording of oil-water flow states were performed. Secondly, with reference to the typical horizontal well oil-water two-phase theoretical flow pattern, the flow patterns under different experimental conditions were identified, and the distribution diagram of flow patterns at four well deviation angles was drawn. Then, the GA-BP neural network was used for learning and prediction of experimental flow pattern data, and a model for predicting four experimental flow patterns was established to realize the flow pattern prediction under different well deviations and different flowrates in horizontal well. Finally, the above prediction model was verified based on 24 data points of two sample wells. After comparison with logging data, it was found that the accuracy of identifying flow patterns through the prediction model was high, and the overall coincidence rate reached up to 87.25%, indicating that the prediction model met the requirements for interpretation of actual logging data.