|
Evaluation of intelligent models in predicting the discharge of Aladyzga and Arbabkandy stations
|
Fariborz Ahmadzadeh Kaleybar    , Ahad Molavi * , Bahman Mehrvarz Qoje Begloo |
| Assistant Professor, Department of Water Science and Engineering, Ta.C., Islamic Azad University, Tabriz, Iran, Email: ahad.molavi@iau.ac.ir |
|
|
Abstract: (246 Views) |
| The present study was conducted to evaluate the performance of artificial neural networks, support vector machine models, and their hybrid mode with the wavelet model in predicting the discharge of the Aladyzga and Arbabkandy hydrometric stations located in the Qara Su watershed. Considering the correlation index, the discharge of two months ago and the discharge of one month ago were considered as the input of the runoff model at the Arbabkandy and Aladyzga hydrometric stations, respectively. The optimal state in the artificial neural network and hybrid wavelet-artificial neural network models was achieved in two and five neurons at the Aladyzga station, respectively, and in 12 and one neuron at the Arbabkandy station, respectively. The results indicated that the agreement between the observed runoff and predicted runoff values was high when using the wavelet-artificial neural network combination and the wavelet-support vector machine combination compared to the cases of using the conventional artificial neural network and support vector machine. Thus, at Arbabkandy station, hybridizing the single model of the artificial neural network with the wavelet model increased the R parameter from 0.44 to 0.91 and also reduced the RE and RMSE parameters from 41% and 2.03 m3 s-1 to 23 % and 1.33 m3 s-1, respectively. The NSE and GMER indices in the wavelet-artificial neural network and wavelet-support vector machine models had better acceptance in both stations than in the other models, so that in Arbabkandy station, the values of these indices in the wavelet-artificial neural network model were 0.78 and 0.94, respectively. After the hybrid wavelet-artificial neural network model, which had the best fit and consistency with the observational data, the hybrid wavelet-support vector machine model had good accuracy and efficiency compared to other models used in both stations |
Article number: 3 |
| Keywords: Artificial neural network, Support vector machine, Wavelet transform, Time delay |
|
|
Full-Text [PDF 2244 kb]
(80 Downloads)
|
|
Type of Study: Research |
Subject:
Special
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
