مرکز منطقه ای اطلاع رساني علوم و فناوري - Modeling of Discharge in Rivers by Artificial Neural Network

Abstract :

In order to provide the sustainable use of the water resources and prevent flood damages it is indispensable to measure regularly the river flow. Regular flow measurements take much time and cost. For that reason, in practice it is generally estimated approximately by the rating curves. But a rating curve must be obtained for each station and the prediction shows mistakes because rating curves change by time. Some of the methods of artificial intelligence, artificial neural networks and fuzzy logic models make up an alternative for the rating curves. In this study, 5 different stations, for 22 different stream flow conditions have been selected at side branch of the Kızılırmak River. Dividing the cross sections in slices, the geometry of the river is identified and discharge is determined with velocity-area method. Using the determined bottom slope, bottom roughness coefficient, water level flowing through slice and river cross section width value as input data; discharge from each slice was predicted, with multilayer perceptron (MLP), radial basis neural networks (RBNN) and adaptive-network based fuzzy inference systems (ANFIS) models. Errors between the measured values and model predictions were determined, the model performances were compared with each other, also the effectiveness of input parameters in determining the discharge was examined. ANN and fuzzy logic models showed to be fairly successful in the determination of discharge, and the performance of models observed were close to each other. The best performance was obtained from ANFIS model. The water level did not produce adequate results individually, but along with other input data is regarded as the most effective input parameter in the discharge prediction.