非线性结构向量自回归模型因果关系的图模型辨识方法

摘要

确定变量间的因果关系是时间序列分析的重要内容。传统的图模型因果推断算法有着明显的局限性，要求模型是线性的且噪声项服从Gauss分布。本文利用图模型方法辨识非线性结构向量自回归模型变量间的因果关系，给出了一种基于互信息和条件互信息的非线性结构向量自回归因果图模型结构的非参数辨识方法。数值模拟结果验证了方法的有效性。%It is important to detect and clarify the cause-effect relationships among variables in time series analysis. Traditional graphical models causality inference methods have a salient limitation that the model must be linear and with Gaussian noise. In this paper, we apply the graphical models to infer the causal relationships a-mong variables of nonlinear structural vector autoregressive models. We propose a nonparametric method which employs both the mutual information and condi-tional mutual information to identify the causal structure of nonlinear structural vector autoregressive causal graph model. Numerical simulations demonstrate the effectiveness of the method.