Selection of Physically Interpretable Data Driven Model Structures to Analyze Industrial Processes

This work presents a black-box input selection approach to reveal causal dependencies between process variables of complex industrial systems. This allows data based modeling with physically interpretable model structure. For this purpose a method is used which combines statistical and analytical approaches to find causal relations between measured data, detection of control loops and the interaction of conditional system behavior respectively. The quality of such models remains in comparison to a common statistical approach unchanged high. The benefit of this input identification approach is an improved insight in complex processes for modeling purposes and their applications.