Model Selection for Model based Controllers

Model based controllers (e.g. Inferential, Internal Model, Self-Tuning Regulators, Dynamic Matrix, etc.) are designed and implemented using a process model. The controller is generally designed to either explicitly or implicitly invert the process model. When the model is identically equal to the process, controller design is straightforward. If the model structure is known exactly, intuition suggests that the best approach is to try to identify all the unknown parameters. For many real systems the estimated parameters will be in error to some degree. Under these circumstances, simulation and experimental work suggests that simpler and/or lower order models can result in control performance comparable to models of the "correct" form (e.g. the z transform of the continuous process) when the controller is properly designed.