Improving Modelling of Discharge Coefficient of Triangular Labyrinth Lateral Weirs Using SVM, GMDH and MARS Techniques

Abstract

AbstractIn this study the discharge coefficient (Cd) of labyrinth lateral weirs was modelled and predicted using artificial neural network (ANN), adaptive neuro fuzzy inference system (ANFIS), support vector machine (SVM), group method of data handling (GMDH), and multivariate adaptive regression splines (MARS) techniques. To this end, a related data set was collected from the literature. Results indicated that the proposed ANN model includes two hidden layers with eight and five neurons in the first and second hidden layers, respectively. Testing transfer functions demonstrated that the radial basis function (RBF) has the best outcome. Evaluating the performance of the ANN model shows that this model with a coefficient of determination (R2 = 0.96) and root mean square error (RMSE =0.07) in the testing stage has suitable functionality. Reviewing the ANFIS model showed that this model with one hidden layer containing seven membership functions within the Gaussian function could achieve R2 = 0.94 and RMSE =0.136 for predicting the discharge coefficient. During preparation of SVM, it was found that this model with RBF as the best kernel function obtained R2 = 0.96 and RMSE =0.07. The structure of the MARS and GMDH models pointed out that the Froude number, grandiose index and ratio of upstream flow depth to the height of weir are the most effective parameters for predicting the discharge coefficient. Minimum R2 and RMSE achieved were equal to 0.93 and 0.08, respectively, in the testing stage for the MARS and GMDH models. Copyright © 2017 John Wiley & Sons, Ltd.