Comparative study of soft computing models for prediction of scour below two symmetric crossing jets

Abstract

The scour estimation downstream of the dam's hydraulic structures is a serious issue and has long been considered an important topic by hydraulic engineers. The literature shows that the computational methods are good alternatives for the prediction of the scour in various hydraulic structures in which the conventional methods demonstrate shortcomings. In the present paper, the different machine learning techniques have been used for the first time to predict the scour below the two symmetric crossing jets. Four different artificial neural networks (ANNs), including feed forward back propagation (FFBP), cascade forward back propagation (CFBP), radial basis function (RBF), generalized regression neural network (GRNN) along with the adaptive neuro-fuzzy inference system (ANFIS), and support vector regression (SVR) were employed and evaluated. Two important scour hole characteristics, namely the maximum scour hole depth, and its location relative to the scour hole origin, are predicted at different crossing angles. The soft computing models are also compared with the traditional empirical methods. The results indicate that the applied computing techniques perform satisfactorily and improve the results remarkably. They can estimate the scour more precisely than the regression models and can be considered robust alternative tools. A detailed sensitivity analysis is also performed on the FFBP that shows the crossing angle parameter has a powerful effect on the predictions. The SVR improved the results to 47% and 31.71% for the scour hole depth and its location in terms of the coefficient of correlation.