Data-based causality detection from a system identification perspective

The problem of detecting causality, from routine operating data, is reviewed from a system identification perspective. It is shown that even simple examples from the literature under Granger causality analysis do not have adequate model fit. As an alternative, this study uses the system identification platform to capture causality from process data. For example, the model inadequacy test is considered an important reason to reject a given causal relationship. The rich framework of system identification techniques and the choice of models to deal with exogenous variables and nonlinearities are shown to be an extremely suitable foundation to detect casual relationships. The utility of the proposed approach is illustrated by several benchmark examples including the analysis of routine operating data in an industrial case study.