Application of Wavelet Denoising and Artificial Intelligence Models for Stream Flow Forecasting

In this study, the ability of threshold based wavelet denoising Least Square Support Vector Machine (LSSVM) and Artificial Neural Network (ANN) models were evaluated for forecasting daily Multi-Station (MS) streamflow of the Snoqualmie watershed. For this aim, at first step, outflow of the watershed was forecasted via ad hoc LSSVM and ANN models just by one station individually. Therefore, MS-LSSVM and MS-ANN were employed to use entire information of all sub-basins synchronously. Finally, the streamflow of sub-basins were denoised via wavelet based thresholding method, then the purified signals were imposed into the LSSVM and ANN models in a MS framework. The results showed the superiority of ANN to the LSSVM, MS model to the individual sub-basin model, using denoised data with regard to the noisy data, e.g., DCLSSVM=0.82, DCANN=0.85, DCMS-ANN=0.91, DCdenoised-MS-ANN=0.94.

Keywords:

Stream flow , Denoising , Artificial Neural Network , Least Square Support Vector Machine , Multi-Station , Snoqualmie watershed