Deep causal mining for plant-wide oscillations with multilevel Granger causality analysis

Plant-wide disturbance such as oscillations are common in large-scale complex controlled processes whose effects propagate to many units and may deteriorate overall control performance. It is important to capture the major causal relationship within the plant and diagnose the root cause along with complete propagation paths. This paper presents a novel multilevel Granger causality framework for root cause diagnosis. The high level is regarded as a group-wise analysis, which is clustered by dynamic time warping-based K-means method and investigated using group Granger causality. The low level is individual causal reasoning within each group where a partial least squares modified Granger causal test is developed to overcome multicollinearity issue. The proposed causality analysis framework is validated through a benchmark industrial case study to show its effectiveness and superiority.