Selecting of Combinations Forecasting Techniques of Annual Stream-Flow and Application *

# suxiaoling17@126.com Received: Apr. 24 th , 2012; revised: May 7 th , 2012; accepted: May 19 th , 2012 Abstract: In order to improve runoff forecast accuracy, combination forecasting method is selected by error sequence stability judgment and applying a technique with a bias correction component. Eight combinations technology were applied to combine the four single-value forecasts. The relative deviation and relative root mean square error index were used to compare the accuracy of the various single-value forecast and com- bined forecasts. Select the Dongda River as an example. The major findings include that: 1) SVM model and ARIMA model performs best among the four individual prediction models; 2) The accuracy of combining the corrected single-value forecasts is higher than combining the non-corrected single-value forecasts; 3) WA performs better than SA combination method; 4) The Regression and ANN combining methods can remove the effects of bias in the constituent forecasts and yield unbiased combining forecasts.