用于高维时序数据预测的非同步尺度主成分分析  

Asynchronous Scaled Principal Component Analysis for High-dimensional Time Series Data Prediction

作　　者：张怡平 金文玲 董晨昱[2] 牛家敏 武月 ZHANG Yiping;JIN Wenlin;DONG Chenyu;NIU Jiamin;WU Yue(Department of Housing and Urban-Rural of Shanxi Province,Taiyuan 030013,China;School of Mathematical Sciences,Shanxi University,Taiyuan 030006,China)

机构地区：[1]山西省住房和城乡建设厅,山西太原030013 [2]山西大学数学科学学院,山西太原030006

出　　处：《山西大学学报（自然科学版）》2023年第2期321-325,共5页Journal of Shanxi University(Natural Science Edition)

基　　金：国家自然科学基金(62076156)。

摘　　要：针对时间序列数据预测过程中可能面对高维或超高维的预测变量,同时考虑变量的时序特征及预测的非同步性,提出用于时序数据预测的非同步尺度主成分分析方法。首先构建单个预测变量和被预测变量的非同步线性回归,通过可决系数选取单变量的最佳滞后阶数,并将回归系数赋权与相应的预测变量得到赋权预测变量,并通过主成分分析对赋权预测变量降维,即非同步尺度主成分分析。将该方法用于消费者物价指数增长率的预测,结果表明经非同步尺度主成分分析降维的预测精度高于传统降维预测的方法。Considering the possible high dimensional or ultra-high dimensional prediction variables in the process of time series data prediction,as well as non-synchronization between the time series and predictions,this paper proposes an asynchronous scaled principal component analysis method for time-series data prediction.Firstly,the asynchronous linear regression of a single prediction variable and the predicted variable is constructed.The optimal lag order of a single prediction variable is selected through the decisive coefficient,and the regression coefficient is weighted to the corresponding prediction variable to obtain the weighted prediction variable.Finally,the dimension is reduced through asynchronous principal component analysis.This method is used to predict the growth rate of consumer price index(CPI) and the results show that the prediction accuracy of asynchronous scaled principal component analysis is higher than that of traditional methods.

关 键 词：尺度主成分分析 宏观经济预测 非同步性 时间序列 

分 类 号：O212[理学—概率论与数理统计]