Study on prediction model of liquid holdup based on back propagation neural network optimized by tuna swarm algorithm
Rongge Xiao,Guoqing Liu,Dongrui Yi,Bo Liu,Qi Zhuang
DOI: https://doi.org/10.1080/15567036.2023.2229269
2023-06-28
Abstract:It is unavoidable that there will be liquid accumulation in the low-lying areas of the pipelines during the operation of wet gas pipelines. The existence of liquid accumulation can generate a variety of safety issues and, in extreme circumstances, accidents. The accurate calculation of liquid holdup in gas-liquid two-phase flow is of great significance for the study of flow pattern identification, pressure drop calculation, pigging cycle determination, hydrate prediction, wax deposition prediction, pipeline corrosion evaluation and prediction, and transportation efficiency calculation of gas pipelines. Therefore, it is crucial to predict the liquid holdup of wet gas pipelines. 2141 independent experimental data samples were collected and screened out from literatures. Based on the gray theory, gray relation analysis was carried out on the influencing factors of liquid holdup, and the factors with greater influence were selected as the influencing variables; the liquid holdup prediction model based on tuna swarm algorithm optimized BP neural network was established, with pipe diameter, inclination angle, apparent gas velocity, apparent liquid velocity, average temperature, average pressure, and liquid viscosity as input parameters, and liquid holdup as output parameter. Liquid holdup was predicted for upward inclined, downward inclined, and horizontal pipelines respectively. The results show that the prediction model of liquid holdup established in this paper has high accuracy, with the MAPE value of 5.3223%, RMSE value of 0.0213, and R 2 value of 0.9924 for upward inclined pipelines; the MAPE value of 10.1859%, RMSE value of 0.0174, and R 2 value of 0.9922 for downward inclined pipelines; the MAPE value of 4.8037%, RMSE value of 0.0113, and R 2 value of 0.9974 for horizontal pipelines. The predicted results are generally stable and have a wider scope of application, providing a new idea and approach for predicting the liquid holdup of wet gas pipelines.
energy & fuels,engineering, chemical