Gas-liquid two-phase flow rate measurement with differential pressure and density ratio synergistic dual neural network

Abstract

Machine learning has been widely applied in the field of fluid measurement. Establishing the mapping relationship between feature data, quality, and total mass flow rate is crucial for accurate measurement of gas-liquid two-phase flow. This study proposes two MLP models for gas-liquid two-phase flow measurement. A throttling experiment was conducted using a nozzle with a throat diameter of 12 mm, a total of 122 sets of experimental data were collected, with superficial gas velocity and superficial liquid velocity coverage ranges of 1.73–20.72 m/s and 0.0173–0.242 m/s, respectively, covering four flow patterns: stratified flow, wave flow, slug flow, and annular flow. By combining throttling mechanisms for feature selection, the input features of the quality prediction model are determined to be square root differential pressure ratio and square root gas-liquid density ratio, while the input features of the total mass flow rate prediction model are square root difference of double differential pressure and square root gas-liquid density ratio. And validate the effectiveness of features using Spearman and Pearson correlation analysis methods. The study results indicate that the relative error of the test samples for quality prediction model is within ±7 %, and the mean absolute percentage error (MAPE) is 2.78 %. The relative error of the test samples for total mass flow rate prediction model is within ±6 %, and the MAPE is 2.05 %. Both models were validated through 5-fold cross validation to ensure no overfitting occurred. This work avoids the problem of error accumulation through a dual model parallel prediction architecture, and achieves high-precision flow rate prediction with a small number of features and a small sample dataset, providing a data-driven new solution with practical engineering value for gas-liquid two-phase flow rate measurement.