副热带高压强度变化的模糊聚类诊断预测  

Fuzzy Clustering Diagnosing and Forecasting on Subtropical High Intensity Index

作　　者：洪梅[1] 张韧[1] 吴国雄[1] 何金海[2] 余丹丹[3] 

机构地区：[1]中国科学院大气物理研究所LASG,北京100029 [2]南京信息工程大学KLME暨大气科学博士后流动站,南京210044 [3]解放军理工大学气象学院海洋与空间环境系,南京211101

出　　处：《应用气象学报》2006年第4期459-466,共8页Journal of Applied Meteorological Science

基　　金：国家自然科学基金项目(40375019);江苏省气象灾害重点实验室开放课题(KLME0507);中国博士后科学基金项目(2004036012)共同资助

摘　　要：采用模糊C均值聚类(FCM)、遗传算法(GA)和模糊减法聚类(FSC)交叉融合、优势互补思想进行副热带高压影响因子的综合聚类分析和副热带高压指数的诊断预测。在统计分析的基础上,通过选择若干与副热带高压指数关系密切的影响因子构成了高维特征空间,进行了综合聚类分析,实现了副热带高压指数的聚类判别和诊断预测。该文提出的综合聚类方法既可克服FCM/GA算法全局/局部寻优的不足,又可客观确定聚类数目。试验结果表明,该方法具有良好的分类效果,判别结果与实况基本相符。In order to analyze and diagnose the correlative characters and phase relation between subtropical high and the east-Asia summer monsoon circulation and thermal factors as well as southern hemispheric pressure field, an idea of synthetically combining FCM, GA with FSC to make clustering analyze and diagnosis on subtropical high and its influence factors is presented. Based on observational and statistical facts, some initial subtropical high influence factors in east-Asia monsoon system are chosen to form a high-dimensional feature spaces. Firstly, the FSC technique is used to objectively estimate the clustering number of samples in the feature space, then the GA technique is used to make global optimization search and the FCM technique is used to make further local optimization searching and adjusting on the GA global optimization results. Finally, the synthetical optimized clustering results of subtropical high indexes and its influence factors in the feature mapping space are taken as the classification criterion for subtropical high intensity index classification. By the clustering analyze and diagnosis of using above synthetical techniques, some important portentous characteristic facts of subtropical high changing are revealed and discovered, and a corresponding clustering diagnosis and prediction are carried out. On the basis of above researches, some examples of subtropical high index and its monsoon influence factors are analyzed, diagnosed and predicted by using the synthetical clustering technique, the experiments results show that the classification correctness and forecast precision of 5 grades subtropical high indexes are superior to 60 %, especially, the correctness and precision in the case of the weakest and the strongest subtropical high indexes are better than 80 %, and similar better subtropical high classification results are also gotten by different examples clustering experiments, the validity and reliability of the synthetical clustering technique are well supported and approved by above

关 键 词：副热带高压面积指数 季风系统 模糊聚类 遗传算法 

分 类 号：P434[天文地球—大气科学及气象学]