k-近邻法及铜川降雹预报试验  

k-Nearest Neighbor Method and Its Application to Tongchuan Hail Prediction

作　　者：燕东渭[1,2] 杨艳[1] 孙田文[1] 赵国令[1] 李明[1] 

机构地区：[1]陕西省气象局,陕西西安710015 [2]西安电子科技大学,陕西西安710077

出　　处：《高原气象》2009年第1期209-213,共5页Plateau Meteorology

基　　金：国家科技基础条件平台建设项目“气象科学数据共享中心”(2005DKA31700)资助

摘　　要：分析了传统概率统计方法在天气预报中的局限性,简要介绍了基于实例的学习方法中最典型的k近邻法的基本原理。根据冰雹产生的物理机制,选取当日08时的7个高空因子,以连续8年的248个完整的数据样本为基础,将k近邻法应用于陕西铜川5月份冰雹预报,得到了较好的准确率(TS评分最高达到了0.444)。进而选取了2005年和2006年两年的数据,运用优选出的模型做了进一步预报检验,TS评分优于实际业务工作中预报人员用的传统方法,以及向量机(SVM)和人工神经网络(ANN)。最后分析了实验结果,针对天气预报科学的特点,对k近邻法在其中的应用做了进一步展望。Limitations of traditional statistic methods used in weather predicting are analyzed. And the basic theory of k-Nearest Neighbor （kNN） is briefly introduced. Seven factors from high altitude are picked up according to the physic mechanism of hail＇s generation. Based on the 248 intact examples of 8 consecutive years, kNN is used to predict hail in Tongchuan in May. Experiment shows kNNts better performance（the best TS reaches to 0. 444）. The data in 2005 and 2006 are picked up to do further experiment using the preferred model. Its TS is superior to that of SVM, ANN and the traditional methods which are used by many forecasters. At last, the experiments＇ results are analyzed, and the applications of kNN in weather broadcasting are further analyzed.

关 键 词：铜川 K-近邻 冰雹 

分 类 号：P456.9[天文地球—大气科学及气象学]