Application Of Intelligent Autocorrelated Models For Runoff Simulation: A Case Study Of The Samiyan And Doostbighlou Rivers, Iran
Auteurs : Molavi Ahad .
Résumé
Auto-correlated simulation is one of the effective methods for predicting river flows in basins that do not have information and statistics on various climatic factors. In this study, the flow of Doostbighlou and Samiyan Rivers located in Ardabil province, Iran, was predicted with intelligent meta-exploration models of artificial neural networks, support vector machines, and their combination with wavelet transform, considering acceptable runoff delay times. The evaluation results showed that considering the wavelet transform in the individual methods of artificial neural networks and support vector machines can be an effective help in improving the performance of the above models. The t-test analysis indicated that at a 5% significance level, the null hypothesis (H0: µ1=µ2) was accepted only for the hybrid Wavelet-ANN and Wavelet-SVM models, while the alternative hypothesis (H1: µ1≠µ2) was accepted for the standalone ANN and SVM models. Thus, at Samiyan station, hybridizing the individual artificial neural network model with the wavelet model has increased and decreased the R and RMSE parameters from 0.48 and 1.96 m3s-1 to 0.82 and 1.02 m3s-1, respectively. Also, at Doostbighlou station, integrating artificial neural network and support vector machine models with wavelet transform analysis significantly improved their correlation coefficients, increasing them from 0.39 and 0.45 to 0.77 and 0.72, respectively.
Mots clés
Hybrid wavelet ; time-delayed ; hydrological variables ; stream flow
