An integrated approach of throttling physical model and machine learning for gas-liquid two-phase flow rates measurement

Abstract

In the field of flow measurement, machine learning methods have been widely applied, and the training data used is often related to physical principles. In this study, integrating the hydraulic and thermal characteristics resulting from two-phase flow throttling, the data-physics model-based flow measurement (DPMFM) without separation for gas-liquid two-phase flow by combining throttling physical model with machine learning technique is proposed. Throttling experiments are conducted with a 12 mm nozzle. The throttling physical model is obtained through theoretical derivation and experimental data fitting. It serves as a physical constraint in the prediction model of mas quality and the basis for the calculation of flow rates. Representative features are selected as network inputs based on correlation calculation results. The performance of the model is assessed with untreated datasets and compared with the model without a physical constraint. Measurements of gas and liquid flow rates are in a good agreement with the experiments with Mean Absolute Percentage Error (MAPE) of 3.82 % and 3.62 %. The precision is favorable, with relative uncertainties for gas and liquid flow rate at 0.39 % and 0.57 %, respectively.