基于变换空间的形态子序列快速提取方法  

Fast Extraction Method for Morphological Subsequence Based on Transform Space

作　　者：张癸水 陈黎飞[1] 胡丽莹 ZHANG Guishui;CHEN Lifei;HU Liying(College of Computer and Cyber Security,Digital Fujian Internet-of-Things Laboratory of Environmental Monitoring,Fujian Normal University,Fuzhou 350117,China)

机构地区：[1]福建师范大学计算机与网络空间安全学院,福建师范大学数字福建环境监测物联网实验室,福建福州350117

出　　处：《福建师范大学学报(自然科学版)》2025年第2期55-64,124,共11页Journal of Fujian Normal University:Natural Science Edition

基　　金：国家自然科学基金项目(61672157)。

摘　　要：shapelets是时间序列数据中最具辨识度的子序列片段,能够准确表达时间序列的局部形态特征。现有的形态子序列提取方法通过设置关键点等方式,有效降低了形态子序列的数量,却忽略了噪声的影响且缺乏对信息冗余的优化。针对这种问题,提出一种改进方法,通过重叠式滑动窗口对时间序列进行局部平滑,以降低噪声的影响,并基于变换空间进行信息冗余优化。提出的方法可以在低维空间中提取时间序列的形态子序列,并进行数据表征。结果表明新方法在保持分类精度的同时,提高了分类效率。Shapelets are the most recognizable subsequence segments in time series data,capable of accurately expressing local morphological features of time series.Existing morphological subsequence extraction methods can effectively reduce the number of morphological subsequences by setting key points,but often ignore the influence of noise and lack optimization for information redundancy.To address this issue,this paper proposes an improved method:the time series is smoothed locally using an overlapping sliding window to reduce the influence of noise,and information redundancy is optimized based on the transformation space.The proposed method can extract morphological subsequences from time series in a low-dimensional space and represent the data.The experimental results show that the new method improves classification efficiency while maintaining classification accuracy.

关 键 词：shapelets 噪声 重叠式滑动窗口 优化 低维空间 

分 类 号：O211.61[理学—概率论与数理统计]