Estimation of Transfer Entropy for Capturing Connectivity and Causality in Industrial Processes

Abstract

This paper discusses the use of two information theory measures: transfer entropy and direct transfer entropy, as an approach for detection of causality and connectivity between continuous variables in industrial processes (continuous systems). These measures are asymmetric and identify and quantify linear or non-linear directional relationships between two variables. To estimate these measures, we used estimators based on distances between neighbors. The results obtained from simulations demonstrate the applicability of the measurements and their estimations in order to identify the connectivity map of two systems: autoregressive model (which can be compared with the analytical results) and four water tanks (an industrial systems). Keywords— Transfer entropy, causality, continuous processes, industrial processes.