电子学报 ›› 2019, Vol. 47 ›› Issue (5): 983-991.DOI: 10.3969/j.issn.0372-2112.2019.05.002

• 学术论文 • 上一篇    下一篇

基于自适应增量学习的时间序列模糊聚类算法

王玲1,2, 徐培培1,2   

  1. 1. 北京科技大学自动化学院, 北京 100083;
    2. 北京科技大学自动化学院 工业过程知识自动化教育部重点实验室, 北京 100083
  • 收稿日期:2018-07-19 修回日期:2018-11-11 出版日期:2019-05-25 发布日期:2019-05-25
  • 作者简介:王玲 女,1974年生于北京.北京科技大学自动化学院副教授.研究方向为数据挖掘、机器学习.E-mail:lingwang@ustb.edu.cn;徐培培 女,1994年生于安徽淮北.硕士研究生,研究方向为数据挖掘、机器学习.
  • 基金资助:
    国家自然科学基金(No.61572073);北京科技大学中央高校基本科研业务费专项资金资助(No.FRF-BD-17-002A);北京市重点学科共建项目(No.XK100080537)

Adaptive Incremental Learning Based Fuzzy Clustering of Time Series

WANG Ling1,2, XU Pei-pei1,2   

  1. 1. School of Automation and Electrical Engineering, University of Science and Technology Beijing, Beijing 100083, China;
    2. Key Laboratory of Knowledge Automation for Industrial Processes of Ministry of Education, School of Automation and Electrical Engineering, University of Science and Technology Beijing, Beijing 100083, China
  • Received:2018-07-19 Revised:2018-11-11 Online:2019-05-25 Published:2019-05-25

摘要: 针对现存可用于时间序列的增量式模糊聚类算法往往需要设置多个控制参数的问题,本文提出了一种基于自适应增量学习的时间序列模糊聚类算法.该算法首先继承上一次聚类得到的簇结构信息以初始化当前聚类进程,然后在无需设置参数的情况下自适应地搜索当前数据块中的离群样本,并自动从离群样本创建新簇,最后检查空簇识别标识确定是否需要移除部分簇以保证后续聚类过程的效率.实验结果表明所提算法对等长和不等长时间序列均具有良好的聚类准确性及运行效率.

关键词: 时间序列, 模糊聚类, 自适应增量学习, 离群样本

Abstract: Existing incremental fuzzy clustering algorithms which can be used for time series often require setting multiple control parameters.To solve this problem,a fuzzy clustering algorithm of time series based on adaptive incremental learning is proposed.First,the cluster structure information obtained by the previous clustering process is inherited to initialize the current clustering process.Then,the outliers in current data block are adaptively searched without parameters,and new clusters are automatically created from the outliers.Finally,an empty cluster flag is checked to determine if some clusters need to be removed to ensure the favorable efficiency of subsequent clustering.The experimental results show that the proposed algorithm has good clustering accuracy and efficiency for both equal-length and unequal-length time series.

Key words: time series, fuzzy clustering, adaptive incremental learning, outliers

中图分类号: