Steady-State Pressure Drop for Two-Phase Flow in Pipelines: An Integrated Genetic Algorithm-Artificial Neural Network Approach

This paper develops an artificial neural network (ANN) model for steady-state two-phase flow pressure drop estimation in pipelines. Mechanistic models are traditionally considered in pipeline flow modeling. However, their reliance on fundamental laws of physics can negatively impact their accuracy when dealing with large experimental data sets and various pipeline inclinations. Hence, ANN models prove to be highly accurate compared to mechanistic models. Dimensional analysis is used to derive a broad reservoir of dimensionless groups and form candidate inputs to the ANN model. Identifying the groups leading to the best correlation of the output variable requires a laborious and nonsystematic trial-and-error procedure. To circumvent this problem, genetic algorithms (GA) were considered to identify the best ANN input combination, thereby allowing a good prediction of steady-state two-phase flow pressure drop in pipelines with all inclinations. The sensitivity of the model accuracy to some GA parameters such as the population size and the parent selection scheme was investigated. The proof of concept of the proposed approach was illustrated using the Stanford multiphase flow database. Based on the obtained results, the proposed model was shown to outperform existing mechanistic models when cross-examined using the same database. In addition, the proposed model allowed good prediction accuracy for all pipe inclinations and all flow patterns.