Enhancing Accuracy in Gas–Water Two-Phase Flow Sensor Systems Through Deep-Learning- Based Computational Framework

Multiphase flow is a critical component in contemporary industrial operations, yet the accurate quantification of multiphase parameters presents a substantial obstacle. This research enhances gas-water two-phase flow measurement accuracy via a deep learning framework, leveraging a multisensor array in a laboratory-simulated dual-layer pipeline. Employing electrical resistance tomography (ERT), electromagnetic flow meters (EMFs), and temperature and pressure sensors, it captures real-time data for a deep learning model integrating a classical drift flux model (DFM) for a nonintrusive, comprehensive measurement system. Two models, 1-D convolutional bidirectional long short-term memory neural network (1D CNN-BiLSTM) and multiphase flow estimation neural network (MFENet)—featuring positional encoding, multiattention mechanisms, and a sliding window—were developed. Testing across 185 different flow conditions demonstrated superior precision of MFENet in flow predictions with the average relative errors of 2.45% for gas volumetric flow rate and 1.38% for water volumetric flow rate, outperforming 1D CNN-BiLSTM. This emphasizes the capability of deep learning to improve the accuracy of multiphase flow measurement techniques.