基于LLE和高斯混合模型的时间序列聚类  

Time Series Clustering Based on LLE and Gaussian Mixture Model

作　　者：杨秋颖 翁小清[1] YANG Qiu-ying;WENG Xiao-qing(School of Information Technology,Hebei University of Economics&Business,Shijiazhuang 050061,China)

机构地区：[1]河北经贸大学信息技术学院,河北石家庄050061

出　　处：《计算机技术与发展》2022年第8期33-41,共9页Computer Technology and Development

基　　金：河北经贸大学科研基金(2019QR21)。

摘　　要：聚类分析是常见的数据挖掘方法,时间序列数据挖掘可以将海量时序信息转化成有组织的知识。由于时间序列具有高维度、非线性等特点,大多数聚类算法无法直接应用在原始时间序列数据上并取得令人满意的效果。研究如何在维数约简的同时尽可能多地保留数据的内蕴特征,识别代表知识的真正有趣的模式,具有重要意义。现有大多数时间序列聚类算法没有考虑数据集的局部结构,而数据集的局部结构对聚类性能有较大影响。提出一种基于局部线性嵌入(Locally Linear Embedding,LLE)和高斯混合模型(Gaussian Mixture Model,GMM)的时间序列聚类算法。首先从保留数据集局部结构的角度,使用LLE将每个高维时间序列样本表示为其k近邻的线性组合,并在低维空间进行重构,在保持数据集局部几何结构的同时实现维数约简;然后使用GMM从概率分布的角度进行聚类分析。与已有方法相比,该方法在单变量时间序列聚类上具有更优的效果。Cluster analysis is a common data mining method.Time series data mining can transform massive time series information into organized knowledge.In view of the high dimensionality,nonlinearity and other characteristics of time series,most clustering algorithms cannot be directly applied to the original time series data and achieve satisfactory results.It is important to study how to retain as many inherent features of the data as possible while reducing the dimensionality,and to identify interesting patterns that represent knowledge.Most of the existing nonlinear dimensionality reduction methods reduce the dimension from the perspective of preserving the global features and ignore the local linear features of the data set.A time series clustering algorithm based on LLE and GMM is proposed.Firstly,from the perspective of preserving local features,LLE is used to represent each sample of high-dimensional time series as a linear combination of its k-nearest neighbors and reconstruct it in the low-dimensional space,and dimension reduction is achieved while preserving the local geometric structure of data.Then,GMM is used to perform cluster analysis from the perspective of probability distribution.Compared with the existing methods,the proposed algorithm can obtain better clustering effect in univariate time series.

关 键 词：局部线性嵌入 高斯混合模型 流形学习 时间序列聚类 深度学习 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]