SSIFNet: Oil–Water Two-Phase Flow Pattern Identification Based on Spatial Scale Internal Attention Feature Fusion Network

Abstract

In response to the constraints of current methods in swiftly and accurately identifying complex flow patterns under actual conditions, this article proposes an innovative identification method for oil-water two-phase flow patterns, based on a spatial scale internal attention feature fusion network (SSIFNet). Specially, a spatial scale attention (SSA) module is designed to equip the model with the scale-aware ability to capture the flow characteristics at varying scales for complex flow patterns. Moreover, an internal attention strategy (IAS) is developed to realize the local context and global dependency modeling, so as to realize the accurate identification of the flow pattern. The proposed method leverages a hybrid architecture to combine the strengths of convolutional neural networks (CNNs) in local feature extraction with the ability of transformers to model global dependencies, thereby enhancing the overall identification performance of the oil-water flow patterns. The experimental results demonstrate that the proposed method achieves an accuracy of 89.21% while the number of parameters to 15.93 M, outperforming traditional standard-network and lightweight-network methods in terms of identification accuracy and the number of parameters. The proposed method exhibits high identification accuracy and stability across a range of oil-water flow patterns, particularly when confronted with complex and dynamic logging scenarios.