数据挖掘技术在天气预报中的应用研究
详细信息    本馆镜像全文|  推荐本文 |  |   获取CNKI官网全文
摘要
本文是数据挖掘技术在天气预报中有针对性的应用。本文收集、整理并分析了大量降雹、沙尘以及降水天气资料。首先整理分析了近年从巴盟、包头及呼市各防雹办收集的降雹及雹情资料,且对相应的天气状况进行了初步分析;其次总结分析了全区80年代以来的沙尘暴资料,按能见度的程度及分布状况对沙尘暴个例进行了分类;另外初步分析了中西部地区各站点1995年后4—8月的日降水量,按不同量级和分布进行了统计,对其分门别类建立了样本库。在此基础上根据不同的数据挖掘技术的要求,对相应采用的数值预报产品进行客观处理。针对强对流天气(冰雹)将历史样本天气分为西北气流型、槽区型、西风气流型和西南气流型等4个类型,并建立其HLAFS资料400hPa四种特征场,按照模板匹配的原理,用相似分析方法在大、小两个关键区中对实时HLAFS预报产品进行计算、比较,再建立预报方程,得出结论;对沙尘暴天气将其按区域分为全区、西部、中部、中西部以及中东部5类,再按强弱分别分为强和一般2类,建立ECMWF的3个场(500hPa高度场、850hPa温度场和海平面气压场)的历史资料库,在用传真资料消空之后,用相似离度方法计算实时ECMWF资料,做出不同时次的预报;对降水天气按出现时间和区域分为适合和不适合飞机增雨作业2类,选择T106资料中恰当的气象物理量因子,用BP神经网络算法建立人工增雨降水预报模型,实际应用中将实时T213相关预报资料代入预报模型即可。这些工作表明,在完善气象信息数据的基础上,数据挖掘技术在气象预报,尤其在数值预报产品的释用中将有广阔的前景,在提高灾害性天气预报准确率的进程中,必将发挥愈益明显的作用。
In this paper , the data mining techniques have been applied to the weather forecast with a clear aim .The large quantity weather data about hail shooting , duststrom and precipitation have been collected , put in order and analyzed . First not only carefully analyzed the hail event data gathered from the Hail Suppress Offices of Bayan Naoer League , Baotou and Huhhot in recent years , but also preliminarily analyzed the relevant weather patterns .Second comprehensively analyzed the duststrom data since 1980' s,classified them according to the visibility and distribution .Besides these ,the precipitation data after 1995 within April and August in middle-west of Inner Mongolia were analyzed preliminarily and added up in accordance with different quantity degree and distribution . Futhermore set up the different sample databases .On this foundation ,the numerical forecast products have been objective handled according to different require of every data mining technique .For strong convective weather (hail) , i
    t classified all historical sample events into 4 weather patterns (like northwest , trough area ,west wind and southwest current), established 4 characteristic fields of 400hPa height of HLAFS , then according to the principle of the pattern match calculate and compare the real-time HLAFS forecast products
    -2-
    
    
    using the similar method inside the big and small key areas , establish forecast equation , finally gain conclusion .To duststrom weather, the historical samples were sorted into duststrom and severe duststrom types in 5 regions (they are whole area, west, central, middle-west and middle-east of Inner Mongolia ),moreover set up different sample databases about ECMWF fields (including 500hPa height, 850hPatemperature and sea-level pressure).In order to making duststrom forecast at different time level , we first filtered the real-time data by FAX data ,and then used the method of similar range degree to compare the historical data to the actual data of ECMWF .To precipitation weather ,they were divided into 2 types that suit or unsuit airplane artificial precipitation stimulation in line with their emergence time and district .The appropriate weather physical factors come from T106 were chosen to establish the artificial precipitation stimulation prediction model . In the actual application , we can get predi
    ction result as long as use the real-time forecast data of T213 in the prediction model .These work indicate on the foundation of the data of perfect weather information , be sure that the data mining techniques will have wide prospect in the weather forecast .By all means it will bring into play more function of obvious benefit in the progress of increasing the prediction accurate rate of disaster weather.
引文
[1] Hanjiawei, Kamber, M. Data Mining: Concepts and Techniques (中译本书名:数据挖掘概念与技术/译者:范明等)北京:机械工业出版社,2001
    [2] J.P. Marques de sa. Pattern Recognition: Concepts, Methods and Applications (中译本书名:模式识别—原理、方法及应用/译者:吴逸飞)北京:清华大学出版社,2002
    [3] 孔玉寿 章东华.现代天气预报技术.北京:气象出版社,2002
    [4] 王耀生.人工智能、模式识别在气象领域应用的现状与展望.气象,1994,20(6):9-14.
    [5] 刘景涛,罗孝逞.内蒙古自治区天气预报手册(下册).北京:气象出版社,1987:122-125.
    [6] 吴鸿宾等.内蒙古自治区主要气象灾害分析..北京:气象出版社,1990:100-114.
    [7] 李一平,薄玉华等.内蒙古中西部地区降雹天气预报方法.气象,1999,25(7):37-40.
    [8] 中央气象局.地面气象观测规范.北京:气象出版社,1979:21-27.
    [9] 王式功.中国北方地区沙尘暴变化趋势初探.自然灾害学报,1996,5(2):86-94.
    [10] 徐启运,胡敬松.我国西北地区沙尘暴天气时空分布特征.应用气象学报,1996,7(4):479-482.
    
    
    [11] 杨东贞,房秀梅,李兴生.我国北方沙尘暴变化趋势的分析.应用气象学报,1998,9(3):352-358.
    [12] 周自江.近45年中国扬沙和沙尘暴天气.第四纪研究,2001,21(1):9-17.
    [13] 叶笃正,丑纪范,刘纪远等.关于我国华北沙尘天气的成因于治理对策.地理学报,2000,55(5):513-521.
    [14] 周自江,王锡稳.西北地区东部群发性强沙尘暴序列的建立与分析,地理学报,2002,57(4):437-442.
    [15] 方宗义,朱福康,江吉喜等.中国沙尘暴研究.北京:气象出版社,1997:1-158.
    [16] 刘景涛,杨耀芳,李运锦等.中国西北地区1993年5月5日黑风暴的机理探讨.应用气象学报,1996,7(3):371-376。
    [17] 周自江,王锡稳,牛若芸.近47年中国沙尘暴气候特征研究.应用气象学报,2002,13(2):193-200.
    [18] 方宗义,张运刚,郑新江等.用气象卫星遥感监测沙尘暴的方法和初步结果.第四纪研究,2001,21(1):48-55.
    [19] 杨行峻,郑君里.人工神经网络.北京:高等教育出版社,1992
    [20] Hagan,M.T.etc.Neural Network Design(中译本书名:神经网络设计/译者:戴葵等)/北京:机械工业出版社,2002
    [21] 张冬峰,董存仙,任璞等.数值预报产品省级降水预报B—P神经元网络释用系统.山西气象,1999,47(2):1-4.
    [22] 金龙,陈宁,林振山.基于人工神经网络的集成预报方法研究和比较.气象学报,1999,57(2):198-207.
    [23] 王繁强,徐文金,陈杰伦等.B-P算法在青海降雨分区分级预报中的应用.高原气象,1997,16(1):105-111.
    [24] 师春香,吴蓉璋,项续康.多阈值和神经网络卫星云图云系自动分割试验.应用气象学报,2001,12(1):70-78.
    [25] 郝家学,张经珍,孙秀忠等.用神经网络法试报黄河三角洲汛期暴雨.河南气象,2000,(3):19-20.
    [26] 李朝兴,冶林茂,乔春贵等.BP神经元网络算法改进试验.河南气象,1999,(1):41.
    [27] 楼顺天,施阳.基于 MATLAB的系统分析与设计—神经网络.西安:西安电子科技大学出
    
    版社,2000.
    [28] 罗建军,杨琦.精讲多练MATLAB.西安:西安交通大学出版社,2002.
    [29] 王沫然.MATLAB 5.X与科学计算.北京:清华大学出版社,2000.
    [30] 施阳,李俊等.MATLAB语言工具箱—TOOLBOX实用指南.西安:西安西北工业大学出版社,1998.