عنوان مقاله [English]
The precise modeling and prediction of suspended sediment in the river is a key element of water resources management and environmental policies. In this study, the efficiency of artificial neural network models and adaptive neuro-fuzzy network models were evaluated in prediction of daily suspended sediment load in Gero station located in Gero watershed. For this purpose, 782 samples (2001-2014) suspended sediment in milligrams per liter and flow rate measured corresponding to sediment in terms of cubic meters per second were used in three different input patterns. To implement artificial neural network and adaptive neuro-fuzzy network models, the data were divided into two groups of training and testing, in which 80% of the data were for training and 20% for the test. In the artificial neural network, there are two sigmoid functions and a hyperbolic tangent function in the middle layer and a linear function in the output layer, and to perform adaptive neuro-fuzzy network model, a network segmentation method with three membership functions (triangular, Gaussian, and generalized bells) was used with the optimal membership number, which was determined by trial and error. The results obtained from prediction of suspended sediment showed that the best prediction with correlation coefficient (0.96), coefficient of efficiency (0.95) and mean square error (4789.12 mg/l) related to the input pattern 2 with the input variables of current flow rate of current day (Qt), and the daily delayed daily flow rate to 1 day before the origin of prediction time (Q t-1) and the daily suspended sediment delays up to 1 day before the origin of the prediction time (St-1). The results of adaptive neuro-fuzzy network and artificial neural network models showed that the adaptive neuro-fuzzy network model had better performance than artificial neural network in predicting daily suspended sediment in all three patterns.