Multiscale Cross-Channel Transformer for Parameter Measurement in Oil–Water Two-Phase Flow

Abstract

The accurate measurement of water cut and total flow rate in oil–water two-phase flow is crucial for effective oilfield management, particularly in high water-cut environments. The inherent complexity of oil–water flows, characterized by diverse flow patterns and significant nonlinear behaviors, poses substantial challenges for traditional measurement techniques. To address these challenges, we propose a multiscale cross-channel transformer (MSCC-Transformer) model designed to analyze high-frequency signals collected by a double-helix microwave sensor (DHMS). The MSCC-Transformer employs cross-channel multiscale embedding (CCME) and multiscale multihead self-attention (MMHSA) mechanisms to capture intricate interchannel dependencies and extract both fine-grained and long-term features. Moreover, a global average pooling (GAP) layer is used to integrate multiscale information, enhancing feature representation and improving measurement accuracy. The experiment determined 1.3 GHz as the optimal operating frequency for the DHMS. Additionally, dynamic experiments show that the MSCC-Transformer significantly outperforms the existing time-series models in measuring water cut and total flow rate, demonstrating its robustness and accuracy.