基于小波变换多维时间序列最新变化点检测
投稿时间:2020-10-12  修订日期:2020-10-29  点此下载全文
引用本文:
摘要点击次数: 89
全文下载次数: 0
作者单位E-mail
陈雪文 河海大学理学院 1548062593@qq.com 
基金项目:国家自然科学基金项目“基于小波框架的散乱数据重构及其在计算生物中的应用”
中文摘要:现代数据科学中存在大量的多维时间序列数据,检测多维时间序列中的最新变化点对于短期预测很重要。一种改进的方法被提出以检测此类多维时间序列数据中最新变化点。通过使用小波变换,将多维时间序列中的变化点检测问题转化为相对较容易的多维面板数据中的变化点检测问题。该方法旨在跨时间序列合并信息,以便优先推断多个序列中同一时间点的最新变化。通过对每个时间序列的输出进行后处理,获取最新变化点的时间集及该时间最近变化点的序列集。最后使用R软件以模拟数据和BP500数据实验,证明了该方法的有效性。
中文关键词:多维时间序列  小波变换  惩罚成本  最新变化点
 
The latest change point detection of multidimensional time series based on wavelet transform
Abstract:There are a lot of multi-dimensional time series data in modern data science, and detecting the latest change points in multi-dimensional time series is very important for short-term forecasting. An improved method is proposed to detect the latest change points in such multi-dimensional time series data. By using wavelet transform, the problem of detecting change points in the second-order structure of multi-dimensional time series is transformed into a relatively easy problem of detecting change points in multi-dimensional panel data. This method aims to merge information across time series in order to prioritize the latest changes at the same time point in multiple series. By post-processing the output of each time series, the time set of the latest change point and the sequence set of the latest change point in time are obtained. Finally, the use of R software to simulate data and BP500 data experiments proved the effectiveness of the method.
keywords:multi-dimensional time series  wavelet transform  penalty cost  latest change point
查看全文   查看/发表评论   下载pdf阅读器