PREDICTING THE LONGITUDINAL DISPERSION COEFFICIENT IN NATURAL STREAMS USING DEVELOPED ARTIFICIAL NEURAL NETWORK MODEL

The main objective of the present work is to predict the longitudinal dispersion coefficient in natural streams using a neural network (NN) model which was developed based on Quasi-Newton training functions. For this reason, we used the hydraulic and geometric data easily obtained in natural streams. A total number of 100 data sets was used which were split into three subsets: training, validation, and testing sets.The most cited literature in the field was first reviewed in an attempt to identify possible deficiencies and inadequacies in previous studies. In a second stage, a new approach less commonly used by researchers, i.e. the NN model based on QuasiNewton training functions, was employed for predicting the longitudinal dispersion coefficient in natural streams. Finally, the effect of Quasi-Newton training function on the performance of the NN model was investigated and the best architecture was selected for the model developed. The results obtained in this study showed that the proposed model enjoys a satisfactory level of accuracy. The two statistics of the model, i.e. determination coefficient and mean absolute error in testing step, were found to be equal to 0.85 and 53, respectively.