How to use simplified dynamics in model predictive control of superfractionators

In this paper, issues associated with the application of model predictive control algorithms to the product quality control of superfractionator columns are addressed. Full-order and reduced-order (pseudo-ramp) models are compared in different predictive control algorithms, and the effects of different disturbance models on the closed-loop performance are emphasized. The results show that the reduced-order model, which is easier to identify, can give excellent results when it is used along with an effective disturbance model and can be implemented with a relatively small sampling time. The results also show that the output disturbance model typically used by industrial algorithms, is the source of poor closed-loop performance, and although the input disturbance model represents a better choice, the rotation factor disturbance model used with the reduced model can represent a simple and relatively effective practical alternative.