A pre-training model based on CFD for open-channel velocity field prediction with small sample data

Abstract Accurately obtaining the distribution of open-channel velocity field in hydraulic engineering is extremely important, which helpful for better calculation flow and analysis water characteristics. In recent years, machine learning has been used prediction. However, effective training data-driven models heavily depends on diversity quantity data. this paper, a CFD-based pre-training neural network model (CFD–PNN) proposed accurate prediction, allowing adaption to task with small sample Also, cross-sectional prediction method combining computational fluid dynamics (CFD) established. By comparing CFD–PNN six other algorithm CFD model, results show that, case data, can predict more reasonable higher accuracy than models. The average error trapezoidal cross-section about 3.62%. Compared models, improved by 0.3–2.8%.