دانلود مقاله Flood hydrograph modeling and forecasting using artificial neural networks: A numerical example
سال انتشار: ۱۳۸۸
محل انتشار: اولین کنفرانس بین المللی مدیریت منابع آب
تعداد صفحات: ۷
A. Kamkar-Rouhani – Faculty of Mining, Petroleum and Geophysics, Shahrood University of Technology, Shahrood, Iran.
In this research work, the flood hydrograph at downstream of a river is predicted from the recorded upstream hydrograph using artificial neural network (ANN) method. For this, a model based on the ANN method is developed and trained using the first three data sets at the upstream and downstream sections. The fourth data sets are excluded from the training phase as they should be used to check the validation and predictive capability of the model. Using the software package WinNN, a number of models with different ANN parameters are obtained, which establish the relationships between the flood hydrographs at the upstream and downstream sections. In this research work, a suitable 3-layer feed-forward neural network model is chosen and trained using the first three data sets as the input. The optimal network architecture in the 3-layer model is examined by changing numbers of nodes in the input and hidden layer. As an alternative and for comparison with the 3-layer model, the case of 4-layer feed-forward neural network model is also briefly investigated. By running the program for the above-described pattern files using 3- and 4-layer neural networks, in which different number of nodes were specified for the input layer and the hidden layers, the results were obtained. Also a normalization procedure was carried out on the input and output data. Since the results of some of the cases were not promising, we only presented here the cases which produced relatively good results, i.e. contained less RMS error and higher percentage of patterns with error less or equal to the target error. The training will stop when all patterns have an error that is less or equal to the target error in a reasonable time of run. The best trained neural network is then applied to predict the flood hydrograph at the downstream cross-section based upon the fourth data set available for the upstream section. Finally, the predicted flood hydrograph (the outputs from ANN) is compared with the fourth data set for the downstream cross-section.