Ecology, Environment and Conservation Paper

A COMPARATIVE STUDY OF THE ROLE OF OPTIMAL TRAINING FUNCTION OF ARTIFICIAL NEURAL NETWORK IN SUSPENDED SEDIMENT LOAD APPROXIMATION

Mohammad Ali Izadbakhsh

Abstract

In recent years many studies have been carried out using artificial neural network in order to predict the suspended sediment load concentration. However, the selection of appropriate training functions has not received enough attention at artificial neural network training stage. The present study compares three different and common training functions for artificial neural network training in order to approximation the suspended sediment load. Moreover, considering the importance of peak and total suspended sediment load in suspended sediment load approximation, the present study gives adequate emphasis to the mentioned training functions performance; thatis the point neglected by previous researchers. The results of the study indicatet hat all three training functions have acceptable Accuracy in approximating the suspended sediment load. The findings also show that LM and SCG functions performed slightly more accurate. But concerning the peak and total suspended sediment load in the predicted data, it was revealed that “Levenberg Marquardt” training function represents a better capability in suspended sediment load approximating, especially for peak and total suspended sediment load.