区域降水数值预报产品人工神经网络释用预报研究  

Application of Numerical Forecast Products to Regional Rainfall Forecasting by Artificial Neural Network

作　　者：林健玲[1] 金龙 彭海燕[3] 

机构地区：[1]南京信息工程大学,南京210044 [2]广西自治区气象局气象减灾研究所,南宁530022 [3]江苏省气象局,南京530022

出　　处：《气象科技》2006年第1期12-17,共6页Meteorological Science and Technology

基　　金：广西自然科学基金项目(0339025);国家自然科学基金项目(40075021)共同资助

摘　　要：利用T213、日本细网格降水预报等数值预报产品,采用人工神经网络方法进行预报释用.通过聚类分析方法对广西自治区测站进行分类,简化预报对象,对数量众多的T213数值预报产品采用自然正交分解（EOF）方法,浓缩大量因子的有效信息,并结合日本降水预报因子建立广西5～6月区域降水量级的逐日人工神经网络预报模型.运用与实际业务预报相同的方法进行逐日预报试验.结果表明,用这种数值预报产品释用方法建立广西3个预报区域的B-P人工神经网络预报模型对中雨以上降水量级预报的TS评分分别为0.55、0.5和0.26,比目前业务预报中参考使用的T213和日本数值预报产品降水预报具有更好的预报效果.The numerical forecast products （T213, Japanese fine grid rainfall forecast, etc. ） are used in forecasting by means of the artificial neural network technology. Classification is conducted on all stations over Guangxi by use of Cluster Analysis. A large amount of valuable information about numerous factors is concentrated in terms of the Empirical Orthogonal Function （EOF）. Three artificial neural network day-by-day regional rainfall forecast models for May and June in Guangxi Province are established in combination with the rainfall forecast of japan. A day-by-day forecast test is carried out using the operational method. The TS evaluation of the three models is 0.55, 0.5, 0.26, respectively, for moderate rainfall prediction. These results suggest that the three artificial neural network regional rainfall forecast models perform better than the operationally used numerical forecast products-T213 and Japanese model.

关 键 词：人工神经网络 EOF 数值预报产品释用 降水量级预报 

分 类 号：P456.7[天文地球—大气科学及气象学] TS941.732[轻工技术与工程—服装设计与工程]