基于时间序列形态的模糊聚类算法  

Fuzzy clustering algorithm based on time series morphology

作　　者：陈梅[1] 柳博雅 王钰 姚怡莹 尤远毓秀 王松[1] CHEN Mei;LIU Bo-ya;WANG Yu;YAO Yi-ying;YOU Yuan-yu-xiu;WANG Song(School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730000,China)

机构地区：[1]兰州交通大学电子与信息工程学院,兰州730000

出　　处：《控制与决策》2025年第4期1116-1126,共11页Control and Decision

基　　金：国家自然科学基金项目(62266029);甘肃省高等学校产业支撑计划项目(2022CYZC-36)。

摘　　要：针对现有时间序列聚类分析较少考虑到各簇时间序列的相似形态对聚类结果的影响,提出一种基于时间序列形态的模糊聚类算法.该算法使用线性时间复杂度的Jeffreys复合距离度量时间序列之间的距离,利用迭代过程中的隶属度为各簇选择能够映射簇内时间序列相似形态的核心特征,并在下一次迭代中对距离进行特征加权.当隶属度不再显著变化时,算法停止迭代,根据隶属度最大原则对时间序列进行簇划分.在14个公开时间序列数据集上与10种对比算法的实验结果表明,所提算法具有精确的聚类结果和较好的鲁棒性,综合性能优于对比算法.Aiming at the existing time series clustering analysis which seldom considers the influence of time series morphological commonality in clusters on the clustering results,this paper proposes a fuzzy clustering algorithm based on time series morphology.In this algorithm,Jeffreys complex distance with linear time complexity is used to measure the distance between time series,the membership degree in the iterative process is used to select the core features for each cluster that can map the similar shape of time series in the cluster,and the distance is weighted by features in the next iteration.When the membership no longer changes significantly,the algorithm stops iteration,and finally divides the time series into clusters according to the principle of maximum membership degree.The experimental results with 10 comparison algorithms on 14 time series public datasets show that the proposed algorithm has accurate clustering results and better robustness,and the overall performance is better than the comparison algorithms.

关 键 词：时间序列 模糊聚类 相似性度量 时间序列形态 特征加权 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]