Application of Artificial Neural Networks in Order to Predict Mahabad River Discharge

Abstract

Estimating of river discharge is one of the more important parameters in the water resources management. In recent years, due to increasing population, increased water consumption in industrial, agricultural and health sections, thus water shortge becomes a global problem. Accurate estimation of the river discharge is one of the most important parameters in surface water resources management, especially in order to determine appropriate values in flood, drought, drinking, agricultural and industral topics. The case study in this research is Mahabad River that is located in west Azarbaijan province in west north of Iran. In this study, we used 70%, 15% and 15% data in order to train, validate and test, respectively. In this study, data of Kawtar and Baitas stations were used in order to determine Mahabad River discharge. In each ststion, several different networks were prepared using NeuroSolutions V.6.0 software. The neural models included Multilayer Perceptron (MLP), Generalized Feed Forward, Jordan/Elman, Radial Basis Functions (RBF) and Principle Component Analysis (PCA), and different transfer functions included Tanh, Sigmoid, Linear Tanh, Linear Sigmoid and the number of hidden layers of.The different number of nodesin layers with different learning algorithms (Momentum, Levenberg Marquardt, Quickprop, DeltaBarDelta, Conjugate Gradient) and different networks were compared. The results showed the artificial neural networks. They predicted the river discharge with 10.67 and 0.94 ( m 3 /s) 2 and the high value of correlation coefficient with 0.88 and 0.75 for Kawtar and Baitas stations respectivly.