Estimation of River Bedform Dimension Using Artific ial Neural Network (ANN) and Support Vector Machine (SV M)

Abstract

Movement of sediment in the river causes many chang es in the river bed. These changes are called bedform. River bedform has significant a nd direct effects on bed roughness, flow resistance, and water surface profile. Thus, h aving adequate knowledge of the bedform is of special importance in river engineeri ng. Several methods have been developed by researchers for estimation of bed form dimensions. In this investigation, bedform has been estimated using Artificial Neural Network (ANN) and Support Vector Machine (SVM) methods. The results obtained from th ese two methods were compared with empirical formulas of Van Rijn. The accuracy o f the model was evaluated using (RMSE), (MSRE), (CE), (R 2 ) and (RB) statistical parameters. Higher values of statistical parameters indicated that the SVM model with RBF ke rnel function predicted the bedform more accurately than the other method. The values calculated for R 2 , RMSE, MSRE, CE and RB parameters were 0.79, 0.024, 0.066, 0.786, -0.081, respectively. Comparison of the results of the SVM model with RBF kernel with other models indicated that SVM had a higher capability for esti mating and simulating height of the bedform than Artificial Neural Networks. Keywords : Bed roughness, RBF kernel function, River enginee ring.