Development of ANN structural optimization framework for data-driven prediction of local two-phase flow parameters

摘要

In this paper, a workflow of ANN-based model development for data-driven prediction of the local two-phase flow parameters has been proposed. To effectively construct an ANN-based model, a framework of ANN structural optimization was also developed based on Genetic Algorithm method and classical Trial-and-Error method. Two case studies with 638 data points of local void fraction (in subcooled boiling flow) and 3125 data points of local interfacial area concentration (in adiabatic air-water flow) of two-phase flow were investigated. The results clearly proved the effectiveness of optimization framework as well as the predictive capability of developed ANN-based models. Also, comparison results with CFD (computational fluid dynamics) showed that better agreements with experimental data can be obtained with well-trained ANN structure.