克立格法在时间域上做外延预报的可行性  

Feasibility of Goestatistical Method on Time Series Forecast

作　　者：常文渊[1] 戴新刚[1] 封国林[2] 

机构地区：[1]中国科学院大气物理研究所 [2]扬州大学理学院物理系,江苏扬州225009

出　　处：《计算物理》2006年第3期335-342,共8页Chinese Journal of Computational Physics

基　　金：国家重点基础研究发展规划项目(G1999043400);国家自然科学基金(No.49875024)资助项目

摘　　要：利用Lorenz系统混沌解序列和全国160个站的气象资料研究了线性地质统计学在时间域上做外延预报的可行性并与自回归过程AR模型进行比较.结果表明:时间域上普通克立格法(Ordinary Kriging,简称OK)的外延计算虽与矩估计法建模的AR模型法相似,但由于前者使用了无偏性条件使其在平稳或非平稳序列上都具有更好的外延能力;泛克立格法(Universal Kriging,简称UK)由于引入内插中漂移概念反而限制其外延计算,使在平稳序列上的外延解也呈现出类似漂移多项式的变化.同时还提出克立格法有关外延计算的3个试验方案.对实际气象资料的计算表明试验方案存在对OK法外延精度改善的可能性,但这还有赖于外延序列的性质.To study the feasibility of goestatistical method on time series forecast, we use a section of Lorenz system for extrapolation and try to give a theoretic explanation. Numerical investigation points out that though Ordinary Kriging＇ s（OK） solution resembles that of AR model. OK is better for forecast whether the series is stable or not. Universal Kriging＇s （UK） results present a drift which deviates from the real value obviously. Three experimental schemes were designed. In a comparison calculation of the ideal datum and meteorological datum, it indicates that the schemes is possible to improve the extrapolation accuracy. It lies on the characteristics of the series and needs further study.

关 键 词：克立格 AR模型 LORENZ系统 时间序列 

分 类 号：P456.8[天文地球—大气科学及气象学]