Comparison of Methods for Granger Causality Network Estimation

Abstract

Causal relationships among variables serve as a useful tool for augmenting process understanding and to help diagnose the source of abnormalities should they occur in chemical and related process units. Several approaches to decipher the network of relationships from process data have been developed over the years. In this work, we describe an approach for causality determination using Canonical Variate Analysis and compare its results with two other recently proposed approaches that employ Vector Autoregressive (VAR) and Vector Autoregressive Moving Average (VARMA) models. Three simple case studies are presented to compare the efficacy of the approaches.