基于张量多线性PCA的多变量时间序列模式匹配

提出基于张量多线性PCA的多变量时间序列模式匹配方法,通过张量多线性PCA对多变量时间序列进行低维重构并获得其模式表示,然后利用Frobenius范数设计模式间的相似性度量.在四组公开的多变量时间序列数据集上进行实验,结果表明该方法的匹配准确率较高,时间开销较少,且适用于各种规模的数据集.

Pattern matching based on tensor multilinear pca for multivariate time series

The accuracy of existing pattern matching methods for multivariate time series is low on the time series datasets of shorter length. This paper presents a pattern matching method based on tensor multilinear principal component analysis for multivariate time series, the pattern matching method obtains the low dimensional reconstruction of multivariate time series by tensor multilinear principal component analysis and gets the pattern presentation of the multivariate time series. Our method uses the Frobenius norm as the measure of similarity between patterns .The experiment results show that the proposed method achieves higher matching accuracy and spends less time on four open multivariate time series datasets, and is suitable for different size datasets.