SVM回归法在汛期旱涝预测中的应用研究  

Study of application of regression method based on support vector machine on Zhejiang drought and flood's forecast at flood season

作　　者：滕卫平 俞善贤 胡波 胡丰成 

机构地区：[1]浙江省气象科学研究所,浙江杭州310017 [2]浙江省气象台,浙江杭州310017 [3]厦门市气象局,福建厦门361012

出　　处：《浙江大学学报（理学版）》2008年第3期343-347,354,共6页Journal of Zhejiang University（Science Edition）

基　　金：浙江省科技厅重点项目资助(气候模式统计降尺度集成技术的应用研究)(2005C23077)

摘　　要：研究基于统计学习理论的支持向量机(SVM)回归在汛期旱涝预测中的应用.根据浙江省38个测站的降水量资料,用正态化Z指数对汛期旱涝等级进行划分,得到了能够反映全省旱涝状况的指标.以此指标作为预测量,通过相关分析从前期大气环流场、海温场中选取高相关预测因子,应用逐步回归和SVM回归技术分别建立浙江省汛期旱涝短期气候预测模型,并进行了对比分析.结果表明:SVM回归模型集中了众多预测因子的预测信息,有效地利用了支持向量机方法的非线性映射能力,无论在历史样本拟合的精度上还是模型实际预测的能力上都比传统的逐步回归方法有一定提高,具有较好的应用效果.The application of SVM regression method on Zhejiang drought and flood＇s forecast at flood season was studied. By use of normalized Z-index calculated from 38 observatory stations＇ rainfalls, the grades of intensity of drought and flood were classified, and then corresponding index is obtained which reflects the whole state of drought and flood in Zhejiang province. Take the index as predictand, and atmospheric circulation and SST in the past, which exhibit high correlation with predictand, as predictor, the prediction model of Zhejiang drought and flood based on support vector machine and successive regression were built respectively. The comparisons of forecast results of two methods show that the SVM forecast model could make use of plentiful predictor＇s information, and show better performance of non-linear projection than successive regression method, which is confirmed by both training and testing samples.

关 键 词：支持向量机回归 汛期旱涝 预测模型 

分 类 号：P468[天文地球—大气科学及气象学]