基于独立成分分析的时间序列谱聚类方法

Spectral clustering method based on independent component analysis for time series

为了对时间序列数据进行聚类分析,提出了一种基于独立成分分析的时间序列多路归一化割谱聚类方法,并给出了利用独立成分分析对时间序列数据进行特征提取和降维的理论解释.该方法首先利用独立成分分析对时间序列数据进行特征提取,然后利用多路归一化割谱聚类方法完成对时间序列特征数据的聚类分析,从而得到了一种新的基于特征的时间序列聚类方法.为了验证该方法的可行性和有效性,将其应用于仿真时间序列数据和实际的股票时间序列数据聚类分析中,取得了较好的数值结果.

In order to cluster time series data,this paper presents spectral clustering method based on independent component analysis for time series and gives some theoretical interpretations for feature extraction and dimension reduction of time series data by using independent component analysis.The method includes two steps.In the first step,it conducts feature extraction and dimension reduction of time series data by applying independent component analysis.In the second step,it clusters time series data using multiway normalized cut spectral clustering algorithm.Consequently,a new feature-based time series clustering method is derived.With the purpose of validating feasibility and effectiveness of the presented method,the method is used to analyze the simulation time series data and the real world stock time series data and much better numerical results are derived.