A new hybrid decision tree method based on two artificial neural networks for predicting sediment transport in clean pipes

Abstract

A new hybrid decision tree (DT) technique based on two artificial neural networks (ANN), namely multilayer perceptron (MLP) and radial basis function (RBF), is proposed to predict sediment transport in clean pipes (i.e. without deposition). The parameters affecting densimetric Froude number (Fr) prediction were extracted from the literature in order to build the model proposed in this study. The effect of each parameter is first examined using MLP and RBF and a sensitivity analysis. According to the sensitivity analysis, the optimal model indicates that using the volumetric sediment concentration (CV), median diameter of particle size distribution to pipe diameter (d/D) and ratio of median diameter of particle size distribution to hydraulic radius (d/R) parameters yield the best Fr prediction results. Subsequently, the hybrid DT-MLP and DT-RBF model results are compared with MLP and RBF. According to the results, MLP with all models predicted Fr more accurately than RBF, and DT-MLP exhibited the best performance (R2 = 0.975, MARE = 0.063, RMSE = 0.328, SI = 00.081, BIAS = −0.01). Moreover, the comparison between DT-MLP and existing regression-based equations indicates that the models presented in the current study are superior. Keywords: Artificial Neural Network (ANN), Decision Tree (DT), Hybrid model, Multilayer Perceptron (MLP), Radial Basis Function (RBF), Sediment transport