作　　者： (管河山）; (王谦）;

机构地区： 南华大学经济管理学院,湖南衡阳421001

出　　处： 《统计与决策》 2017年第17期28-32,共5页

摘　　要： Box-Pierce Q检验采用近似卡方分布分析时间序列的平稳性特征,其检验统计量的参数选取将影响到检验结果。文章多个Q值提取平稳性特征,在此基础上建立新的平稳性判定准则,该准则是自相关函数序列收敛的充分条件;采用欧氏函数作为平稳性特征的相似性度量,借助k-means聚类建立平稳性分类方法;该方法在平稳性分析过程中充分考虑了样本之间的关联性,避免了传统Box-Pierce Q检验对统计分布和临界表的过度依赖。实验结果表明,新方法能有效地处理海量时间序列数据,且准确率高于Q检验和ADF检验。 Approximate chi-square distribution is used to analyze the stationarity characteristics of time series in Box-Pierce Q test, whose parameter selection of statistical magnitude affects the test results. This paper extracts a kind of stationary characteristic based on multiple Q values and establish a new judging criterion, which is a sufficient condition for autocorrelation function sequence convergence. Furthermore, the paper uses Euclidean function as a similarity measure for the stationary characteristics, and falls back on K-means clustering to establish stationarity classification method, which fully considers the correlation among samples in the process of stationarity analysis in order for traditional Box-Pierce Q test to avoid excessive dependence on the statistical distribution and the critical table. The experimental results show that the proposed method is able to effectively handle huge amounts of time series data, and that its accuracy is better than that of Q test and ADF test.

关 键 词： 时间序列 平稳性 检验 聚类