Predicting the longitudinal dispersion coefficient by radial basis function neural network

Abstract

Study on the pollution transmission in rivers is the main part of environmental engineering studies. In this paper, the performance of empirical formulas, the radial basis function (RBF) and multilayer perceptron (MLP) neural network for predicting the longitudinal dispersion coefficient (DL) in rivers were assessed. To this purpose, 150 data set related to the most influence parameters on the DL was collected. The results of MLP model indicated that the performance of model with error indices such as correlation coefficient (R2 = 0.90) and root mean square error (RMSE = 0.09) had suitable performance for predicting DL. The result of RBF model with correlation R2 (0.61) and RMSE (0.82) also has suitable ability to predict the DL during the MLP model development, found that the MLP model included one hidden layer which ten neurons with log-sigmoid equation as transfer function and pure line equation as transfer function in the output hidden layer. The RBF model contained one hidden layer with 15 neurons.