Research on Correction of Temperature and Mineralization and Prediction of Water Holdup Based on Machine Learning

Accurate detection of water holdup in oil-water two-phase flow is crucial for optimizing production and improving crude oil recovery. The transmission lines method is currently one of the few effective methods to measure the water holdup of oil-water two-phase flow. However, variations in temperature and mineralization will alter the dielectric constant and conductivity of the oil-water mixture respectively, posing challenges for precise water holdup measurement. The complex nonlinear relationship between these factors limits the prediction range and accuracy of widely used models, such as the BP neural network and Support Vector Machine (SVM). In order to overcome these issues, this paper establishes a multi-sensor oil-water two-phase flow indoor experiment system and studies the complex relationship between the phase shift of sensor signal and influencing factors. On this basis, this paper proposes a combined water holdup prediction model (BO-XGBoost) of Bayesian optimization (BO) algorithm and extreme gradient boosting (XGBoost). The results demonstrate that the XGBoost model outperforms traditional BP neural network and SVM in predicting water holdup across the full range of 0%-100%. The average absolute error of the BO-XGBoost model is only 1.50%. The above research achieves a full-range, high-precision water holdup prediction, providing a new solution for oilfield development and possessing practical engineering significance.