SVM-based prediction of runoff in headwater region of the Yellow River

The hydrological system is a system of high nonlinearity. Runoff is the result from the comprehensive action of climate conditions and drainage area underlying surface. The support vector machine (SVM) is a new machine learning method based on the statistical learning theory and it can solve the high nonlinearity, regression, etc in the sample space and also can be used as the hydrological system identification tool. By means of phase space reconstruction, it establishes the SVM model input/output samples; with small sample runoff series, it sets up SVM predicting models. The prediction results show that SVM model has strong generalization ability and very satisfactory prediction results. It effectively solves such problems as small samples, over-learning, high dimension, local minimum, etc. The prediction of the future runoff evolution trend with this model will provide the basis for water regulation and water resources reasonable configuration.