单分位数方法对时间序列尾指数变点检测及应用  

Multivariate Time Series Tail Exponential Change Point Detection and Application

作　　者：周江娥 胡尧 商明菊 ZHOU Jiang′e;HU Yao;SHANG Mingju(School of Mathematics and Statistics,Guizhou University,Guiyang 550025,China;Guizhou Provincial Key Laboratory of Public Big Data,Guiyang 550025,China)

机构地区：[1]贵州大学数学与统计学院,贵州贵阳550025 [2]贵州省公共大数据重点实验室,贵州贵阳550025

出　　处：《贵州大学学报（自然科学版）》2019年第2期22-27,共6页Journal of Guizhou University:Natural Sciences

基　　金：国家自然科学基金项目资助(11661018;11361015);全国统计科学研究项目资助(2014LZ46);贵州省自然科学基金项目资助(黔科合J字[2014]2058号);贵州省科技计划项目资助(黔科合平台人才[2017]5788号)

摘　　要：多元时间序列中的尾指数变点检测在理论和实际应用中都有着广泛应用。本文利用单分位数方法(Single Quantile Method)构造检验统计量检测和估计出多元时间序列数据尾指数变点,证明其极限分布。在模拟研究中,分别产生三个经典的厚尾分布类型随机数进行模拟研究,结果表明,单分位数方法对多元时间序列尾指数的变点检测是有效的,尤其对分布变化造成的尾指数变化的情形更加敏感与准确。最后将该方法应用于深圳市香蜜湖路市委党校南行路段车流量数据,结果显示该方法能准确检测出交通流变点,根据存在的变点分析出交通流的变化规律。Tail Exponential detection in multivariate time series is widely used in both theory and practice. In this paper,the Single quantile method is used to construct test statistics to detect and estimate the tail exponential change points of multivariate time series data,and to prove its limit distribution. In the simulation study,three classical random numbers of the thick-tailed distribution type were generated for the simulation study,and the results show that the single quantile method was effective in detecting the change points of the tail index of the multivariate time series,especially more sensitive and accurate to the variation of the tail index caused by the distribution change. Finally,this method was applied to the traffic flow data of the southbound section of the Xiangmi Lake Road Party School in Shenzhen. The results show that the method can detect the traffic flow change point accurately and analyze the change rule of traffic flow according to the existing change point.

关 键 词：单分位数方法 变点 多元时间序列 厚尾分布 尾指数 

分 类 号：O212[理学—概率论与数理统计]