一种基于Markov链模型的动态聚类方法

A Clustering Algorithm Based on Markov Chain Models

对单变量时间序列的聚类 ,是一类有着广泛应用背景的特殊的聚类问题 由于该问题的特殊性 ,现有的聚类方法无法直接使用 ,故提出了一种新的基于Markov链模型的动态聚类方法 该方法首先对每一个时间序列建立一个描述其动态特征的Markov链模型 ,从而把对时间序列的聚类问题转化为对Markov链模型的聚类问题 然后通过定义各个Markov链之间的“距离” ,采用动态聚类算法完成对这些Markov链模型的聚类 使用该方法 ,分别对一批真实数据和仿真数据进行了聚类试验 。

Clustering a set of univariate time series based on their dynamics is a popular problem that can be widely found in many research areas Because the common clustering algorithm can't directly be used to resolve this problem, a new approach to grouping time series is proposed This approach firstly models each time series as a Markov chain that captures the dynamic character of the time series, and then gets the clustering result to these time series by clustering the Markov chains Using this method, two different data sets are clustered, in which, one is real data, and the other is artificial data By qualitative and quantitative analysis, it is proved that both of the clustering results are good