均值生成函数的周期性延拓在回归分析中存在的问题及其改进方案  

THE PROBLEM OF PERIODIC EXTENSION OF MEANGENERATING FUNCTION IN REGRESSION ANALYSIS AND ITS RESOLUTION SCHEME

作　　者：杨昕[1] 张仁健[1] 

机构地区：[1]LAPC中国科学院大气物理研究所

出　　处：《气象学报》1998年第4期493-499,共7页Acta Meteorologica Sinica

摘　　要：针对均值生成函数的周期性延拓在回归分析中存在的回归前提不同，预报因子是预报量的非独立表现等缺点，给出了改进方案。实例分析计算表明：新方案可以有效地消除原方案中存在的非独立虚假相关现象，从而使得筛选出周期性预报因子更加客观。基于本方案所建立的数学预报模型，具有历史拟合率与多步长预报精度基本一致的特点，是一种具有使用价值的长期预报手段，也有一定的隐含周期分辨能力。ince the periodic extension of mean generating function in regression analysis has the defect of unequal premise, and the predictors are not independent to the predictants. An improving scheme is suggested. Analysis and experiments of samples result in that the new scheme can effectively eliminate the false correlation appearance which exist in old scheme, so the selected periodic predictors are more objective. Basing this new mathematical model, fitting rate and forecast accuracy is basically identical. It is a very useful method for longterm prediction and analysis. What is more, this model has some resolving power of hidden periodicity.

关 键 词：均值生成函数 回归分析 数学模型 天气预报 

分 类 号：P456[天文地球—大气科学及气象学]