A hybrid Forecasting Model of Discharges based on Support Vector Machine

Abstract Forecasting is one of the important research topics in the analysis of the hydrological time series. In order to improve the prediction accuracy for complex flood process, this paper presents a hybrid prediction method, which is based on combining multiple support vector machine (SVM) models. According to different discharge levels, multiple sub-models are established respectively, from which the final result is integrated. For each sub-model, the input is optimally determined by elaborate correlation analysis. Experimental results on the discharge prediction of Wangjiaba station on Huaihe River of China show that the hybrid model can significantly improve the prediction accuracy, compared to the single model without partitioning of the discharge.