Prediction of local scour depth downstream of sluice gates using data-driven models

Abstract

Numerous investigations have already been conducted to characterize the effects of influencing parameters on scouring process in order to derive an accurate predictive equation of local scour depth downstream of a sluice gate. However, due to the complexity of the scour phenomena, available empirical equations, originally derived on the basis of regressive methods, do not always offer accurate scour depth prediction. Previous studies have clarified that artificial intelligence techniques may be alternatively considered to solve a complex phenomenon like local scouring. In present study, Gene-Expression Programming, Model Tree (MT), and Evolutionary Polynomial Regression approaches, which are among the newest artificial intelligence approaches, were evaluated for prediction of local scour depth downstream of sluice gates with an apron. The input variables affecting the scour depth are sediment size and its gradation, apron length, sluice gate opening, and the flow conditions downstream of the sluice gate. Six non-dimensional variables were selected to determine a functional relationship between the input and output parameters. Results of training and testing stages indicated that MT approach yielded the most precise predictions in comparison with the other proposed models and conventional equations.