时间序列平稳性分析的自动机制研究

Research on Automatic Mechanism for Time Series Stationarity Analysis

平稳性分析是建立时间序列自回归滑动平均模型的一个预处理过程,已有的分析方法研究颇多,但尚缺乏一个自动判定平稳性的机制。文章在分析自相关函数的基础上,引入非线性转换理论进行改进,并采用聚类方法建立时间序列平稳性分析的自动机制,为大批量时间序列的平稳性自动判定提供了一条新途径。文中采用了一组模拟数据和一组金融数据来进行实证分析,实验表明,平稳性分析的自动机制能得到较好的结果。

Stationarity analysis has been widely studied,which is a pre-process of time series ARMA modeling,but there is no automatic mechanism for stationarity analysis.In this paper we better autocorrelation function with the non-linear translation theory.The automatic mechanism for stationarity analysis is also proposed,using time series clustering.We utilize two dataset to experiment in this paper,one financial data and one synthetic.Empirical evidence has strongly suggested that our method is shown to yield useful and get robust result.