摘要
将气候预测中常用的74项环流特征量进行归一化处理后,与郑州市冬季气温进行相关普查,利用SVM(SupportVectorM ach ine)两类分类方法,同时考虑气温的年代际变化,建立郑州冬季温度距平趋势预测推理模型,并对因子个数多少和年代际变化对预测模型的影响进行了试验。试验结果表明:用25个和15个因子分别建模时,产生最优模型时样本平均Ts评分均为56%,但后者预报准确率为75%,较前者提高了10%。用20世纪50年代和60年代做试验集,效果较好,产生最优模型时的样本Ts评分和预报准确率较高;用90年代做试验集,效果较差。
Normalizing the 74 kinds of characteristic quantity on circulation which are broadly used in weather forecast, the author proceeds a widespread relation check between these characteristic quantity and the winter temperature of Zhengzhou. Considering the decadal temperature variability, he builds a reasoning model on Zhengzhou＇s wintertime temperature anomalous trend prediction by making use of the two - classification method of SVM. Besides, he makes a test about how much the number of the factors and the decadal variability influence upon the forecast model.
出处
《河南气象》
2006年第1期15-16,共2页
Meteorology Journal of Henan
关键词
支持向量机
训练集
实验集
检验集
Support Vector Machine （SVM）
Training Data
Testing Data
Checking Data
