基于条件互信息的多维时间序列图模型

Graphical models for multivariate time series based on conditional mutual information

在多维时间序列的图模型中引入信息论方法,提出了多维时间序列中各分量之间直接线性联系存在性的互信息检验.定义了线性条件互信息图,图中的结点表示多维时间序列的分量,结点间的边表示各分量之间存在的直接线性相依关系.提出了分量之间条件线性联系存在性的信息论检验方法.图中边的存在性用基于线性条件互信息的统计量检验,统计量的显著性用置换检验决定.应用到实例中的结果表明本文的方法能迅速准确的捕捉各分量之间的直接线性联系.

The information theory is introduced for graphical models of multivariate time series. A method for testing direct linearity between two components is proposed. A class of graphical models, called linear conditional mutual information graph, is defined. The vertex set denotes the components of the series and the edges denote the direct linear dependence structure of the components. The presence of the edges is tested by a statistics based on linear conditional mutual information. The permutation procedure is used to determine the significance of the test statistics. Finally, the method is applied to a real series, and the results show that the method can efficiently capture the direct linear dependence between the components.