Runoff Characterization Of Three Gorges Basin And The Study Of Mid-long Term Runoff Forecasting

Runoff system is highly nonlinear characteristics under the synthetic action ofvarious factors，and its changing trend of flow takes on a great deal of randomness,uncertainty, chaos, fuzziness and grey. The experts and scholars at home and abroad havedone a lot of research work on the long-term variant characteristics of runoff. Byanalyzing the hydrology, climate, characteristics of the basins and underlying surfaceecological environment, they expect to find new methods of mid-long runoff forecastingwhich can be used to direct scientific management of hydraulic engineering and rationalexploitation and optimal allocation of water resources.Taking the annual runoff and monthly runoff of Yichang station, Cuntan station andWulong station at Three Gorges basin as an example, using stepwise regression analysis,wavelet analysis technique and mathematical statistics method to reveal the law of cyclicevolution and the multiple time scale features of the runoff at Three Gorge basin, showingthe trend of high and low water. Two mid-long term runoff forecasting models areestablished on this basis.Considering the traditional hydrological forecasting model being overly dependent onthe quality and quantity of the sample, a coupling prediction model which combines thesupport vector machine and wavelet analysis method is proposed in this paper. Particleswarm optimization is applied to select the parameters of the model as to optimize globalsearching capacity and improve the precision of the model. This method can guaranteethat the obtained extremal solution just is the global extremal solution while finding thedecision function from limited data set. By comparing the prediction effect of differentmodels with different time scale input, we can find that the model performance is bestwhen we use the monthly runoff of first seven months as input.The noise of hydrological series can increase the saturation correlation dimension ofthe hydrological dynamic system which can make runoff characteristics more complicatedto affect the prediction precision, we use the Daubechies wavelet in the multi-scaleanalysis to decompose the monthly runoff series and use the Stein unbiased risk thresholdmethod to eliminate noise in the monthly runoff time series in this paper, of which thepahse space restructure of chaos theory is used to calculate the best delay time andsaturated embedding dimension of the runoff series of Three Gorge basin. At last, takingthe times series computed by the phase space restructure as the input of BP neural network to get the chaos characteristics of the runoff series and monthly runoff prediction of ThreeGorge basin.