Study on improving peak flood forecast accuracy with SVM model

In this paper, support vector machine (SVM) theory is introduced for flood forecast model. In response to the imbalance of training samples, a modified SVM algorithm with peak recognition theory(SVMPR) is proposed. In this algorithm, weight to the peak samples is properly increased to objective function of structural risk minimization. Consequently, the accuracy of flood peak forecast is greatly improved. Flood forecast model was established for Yangkou station in Saxikou basin by SVM and SVMPR algorithm. The contrastive analysis showed the validity of SVMPR algorithm.