A Novel Multiphase Flow Water Cut Modeling Framework Based on Multilevel Multiscale Convolutional Neural Network

Water cut measurement is crucial in oil–water multiphase flows, particularly in late-stage oilfield extraction, where high water production presents significant operational challenges. This article proposes a novel multilevel multiscale convolutional neural network (MLMS-CNN) to achieve water cut estimation. The model is designed to extract and analyze complex flow characteristics through three key modules. The multilevel feature learning (MLFL) module fuses spatial and amplitude–phase features from sensor data, while the multiscale feature fusion module captures flow structures across multiple scales. Additionally, the fully convolutional measurement (FCM) module ensures precise water cut prediction. Experimental results demonstrate that the model achieves a mean square error of 0.013%, highlighting its potential for enhancing real-time industrial multiphase flow monitoring and optimization.