Improvement of Long-Term Run-Off Forecasts Approach Using a Multi-model Based on Wavelet Analysis

This paper promotes a new multi-model method of long-term run-off forecasting based on wavelet analysis to produce an combined estimated output to be used an alternative to that obtained from a single individual long-term run-off forecasts model. Three methods of long-term run-off forecasting are considered, namely Nearest neighbor bootstrapping regressive model (NNBR), Support vector machine model (SVM), Wavelet analysis (WA). Firstly, the original annual runoff series is decomposed into different component sequences with WA, and then the NNBR and SVM models are used to get the forecast results of different component sequences, at last they are reconstructed with WA again. The observed discharges of a 42-year period getting from Yichang station, Yangtze River basin, are used to test the performance of the proposed method. The results confirm that better discharge estimates can be obtained by the multi-model approach based on wavelet analysis compared with the single NNBR and SVM models.