A combination of computational fluid dynamics, artificial neural network, and support vectors machines models to predict ow variables in curved channel

This study presents the combination of Computational Fluid Dynamics (CFD) and soft computing techniques to provide a viewpoint for two-phase ow modelling and accuracy evaluation of soft computing methods in the three-dimensional ow variables prediction in curved channels. Artificial Neural Network (ANN) and Support Vectors Machines (SVM) models with CFD are designed to estimate velocity and ow depth variables in 60 sharp bend. Experimental results for 6 different ow discharges of 5, 7.8, 13.6, 19.1, 25.3, and 30.8 l/s to are used to train and test ANN and SVM models. The results of numerical models are compared with experimental values and the accuracy of models is conrmed. Evaluation of the results shows that all the three models of ANN, SVM, and CFD perform well in ow velocity prediction with correlation coeffcients (R) of 0.952, 0.806, and 0.680 and ow depths (R) of 0.999, 0.696, and 0.614, respectively. ANN model, with Mean Absolute Relative Errors (MAREs) of 0.055 and 0.004, is the best model in prediction of both velocity and ow depth variables. Then, SVM and CFD models with MAREs of 0.069 and 0.089 in velocity prediction and CFD and SVM models with MAREs of 0.007 and 0.011 in ow depth prediction are the best models, respectively.