عنوان مقاله [English]
نویسندگان [English]چکیده [English]
One of the techniques that has been used in different areas of sciences, and probably would be able to simulate complicated process of rainfall-run off is artificial neural network (ANN).
The purpose of this research is to evaluate the applicability of ANN in rainfall- run off simulation and the comparison of the results to those produced by HEC-HMS model in Azam river basin of Herat in Yazd province. The data used in this study are daily rainfall and daily discharge as well as instantaneous peak discharge measured in a time period of 24 years (1982-2006). At first, preparation of rainfall data and the related run off hydrographs of some events was the basic of work, then artificial neural network with error back propagation algorithm and the sigmoid transfer function was trained using the prepared data. The criterion to select and set the network parameters was the production of the least amount (RMSE) value between outputs. HMS model using SCS suggestive procedure (using CN) was also calibrated and used. To evaluate the efficiency of the models, observations and models outputs related to whole flow data, run off volume, the time to peak, and the peak discharge were compared. Findings of this research show that the correlation coefficient (r) between measured and estimated flow data is 0.978 and 0.823 respectively for ANN and HEC-RAS models, showing higher accuracy of ANN outputs. About the run off volume and peak discharge, r obtained respectively 0.986 and 0.981 for ANN where for the HEC-HMS model these are respectively 0.979 and 0.972.Comparing the estimated time to peak to the observed values shows more efficiency of ANN model over the HEC-HMS model as the r for outputs of these to models are respectively 0.833 and 0.491. In overall, comparing the application of ANN and HEC-HMS models indicates that for all evaluated parameters the efficiency of ANN is higher than HEC-HMS. However, using t-test it was observed that statistically there is no meaningful difference between the observed and estimated values at the levels of 99 and 95 percents for all parameters.