A Transfer Entropy Method to Quantify Causality in Stochastic Nonlinear Systems

Abstract

In modern chemical processes, identification of the process variable connectivity and their topology is vital for maintaining the operational safety. As a general information theoretic method, transfer entropy can analyze the causality between two variables based on estimation of conditional probability density functions. Transfer entropy estimation is typically a data driven method, however, the associated high computational complexity and poor accuracy are not acceptable in real applications. Using a nonlinear stochastic state-space model in conjunction with particle filters, a novel transfer entropy estimation method is proposed. The proposed approach requires less data, is fast and accurate.