摘要
以2003～2006年ECMWF数值预报产品、Japan和T213降水预报场资料为基础，应用支持向量机（SVM）方法和CMSVM应用软件平台，通过对训练样本进行交叉验证和模型核参数的逐渐逼近，分区建立了武汉区域16，个区SVM24h暴雨预报模型，并在武汉区域5～7月进行实时业务应用。通过预报结果检验，5～7月武汉区域暴雨预报TS评分为33．59％，其中湖北省平均成绩为34．69％、湖南省平均成绩为34．15％、河南省平均成绩为31．71％。检验结果较好，表明SVM方法在区域性暴雨预报中具有一定的预报能力和参考价值。
Based on the ECMWF, Japan, and T213 precipitation data from 2003 to 2006, using the SVM method and CMSVM application software, the cross-validation with random samples and the gradual approaching of model kernel parameters are performed, and then the 24-hour SVM heavy rainfall forecasting models are established for 16 sub-regions in the Wuhan region, which have been put to use in the real-time operation from May to July in 2007. The results show that the TS points for Hubei, Hunan and Henan are 33.59%, and 34.69%, 34.15% and 31.71%, respectively, which proved the forecasting capability and reference value of the SVM method.
出处
《气象科技》
北大核心
2009年第2期145-148,共4页
Meteorological Science and Technology
基金
中国气象局2007年多轨道业务建设项目"长江中游暴雨洪水定量预报和水循环影响评估系统"资助
关键词
支持向量机(SVM)
暴雨预报
业务应用
Support Vector Machine （SVM）, heavy rainfall forecasting, operational application
