基于EEMD与WT组合的变点检测研究  

Research on Change-point Detection Based on Combination of EEMD and WT

作　　者：郭建平 赵立龙 Guo Jianping;Zhao Lilong(School of Management Engineering,Nanjing University of Information Science&Technology,Nanjing 210044,China;School of Physics and Optoelectronic Engineering,Nanjing University of Information Science&Technology,Nanjing 210044,China)

机构地区：[1]南京信息工程大学管理工程学院,南京210044 [2]南京信息工程大学物理与光电工程学院,南京210044

出　　处：《统计与决策》2025年第8期59-63,共5页Statistics & Decision

基　　金：国家社会科学基金一般项目(22BTJ061)。

摘　　要：准确检测出数据的变点对于统计分析而言具有重要价值。文章构建了一种对观测数据本身没有要求的自适应变点检测方法,通过对原始序列施以聚合经验模态分解,基于方差贡献率分析,确定系统演变的主要影响因子;利用小波变换优良的局域化能力,确定主要影响因子的变点位置;根据系统变化时刻和系统主要影响因子的变化时刻在时间上的相依关系,定位原始序列的变点。结果表明,与经典回归分析和直接使用小波变换检测变点相比,所提组合变点检测法能够大幅提升变点检出率。Accurately detecting change point in data is of great value for statistical analysis.This paper constructs an adaptive change-point detection method that does not require the observation data itself.By applying ensemble empirical mode decomposition to the original sequence and analyzing the variance contribution ratio,the main influencing factors of system evolution are determined.The excellent localization characteristics of Wavelet Transform(WT)are utilized to determine the location of the main influencing factors’change point.According to the temporal dependency relationship between the system change time and that of main influencing factors of the system,the change point of the original sequence is located.The results show that compared with classical regression analysis and direct use of WT to detect change point,the combined change-point detection method proposed in the paper can significantly improve the detection rate of change point.

关 键 词：变点 聚合经验模态分解 主要影响因子 小波变换 

分 类 号：C812[社会学—统计学]