期刊论文

中文

统计与决策

1002-6487

2022

38

10

16-20

10.13546/j.cnki.tjyjc.2022.10.003

教育部人文社会科学研究青年基金项目(21YJC910005) 安徽省优秀青年人才支持计划重点项目(gxyqZD2020100)

本校为第一机构

由于小波理论的高频面板数据聚类方法需要选择合适的母小波以及设置可行的分解层数，其聚类结果较易受到人为影响，缺乏自适应性。文章提出了一种基于经验模态分解特征提取的高频面板数据自适应聚类方法，该方法基于经验模态分解法分解高频时间序列信号，提取“绝对”特征、“趋势”特征、“高频波动”特征和“低频波动”特征，并运用熵值法进行客观赋权，进而构造基于加权余弦距离的层次聚类算法。实证检验结果表明该方法更适用于非平稳、非线性的高频面板数据聚类问题，正确率明显提高。

The clustering method for high frequency panel data based on wavelet theory needs to select the appropriate mother wavelet and set the feasible decomposition level,and thus the clustering results are easily affected by human beings and lack self-adaptability.This paper proposes an adaptive clustering method for high frequency panel data based on empirical mode decomposition(EMD)feature extraction.This method decomposes high frequency time series signals based on EMD,extracts“absolute”features,“trend”features,“high frequency fluctuation”fe...