A Flow Rate Estimation Method for Gas–Liquid Two-Phase Flow Based on Transformer Neural Network

Abstract

Accurately estimating the flow rate of gas–liquid two-phase flow is crucial for reducing costs, improving efficiency, and optimizing management in multiphase industries. However, due to the inherent complexity of multiphase flow, this task remains highly challenging. In this article, we propose a flow rate estimation method for gas–liquid two-phase flow based on Transformer neural network (TNN). The proposed method utilizes Transformer encoder layers to extract flow rate features and effectively fuses these features at various levels through the stacking of layer outputs. Compared with previous methods based on recurrent neural network (RNN), this method can process data more effectively in parallel by extracting flow rate features through the self-attention mechanism. The experimental results show that the proposed method achieves more accurate flow rate estimation across various performance indicators, compared with other deep learning-based models. In the robustness experiments against noise, the proposed method shows good robustness, demonstrating its significant potential in accurately estimating flow rates in gas–liquid two-phase flow scenarios.