Robustness of MPC and Disturbance Models for Multivariable Ill-conditioned Processes

The problem of robust predictive control for multivariable illconditioned systems is addressed in this work. A technique for designing robust disturbance models is presented, which is based on an off-line min-max optimization problem. It is shown that the most common disturbance model – the output disturbance model – is not robust to modeling errors when the process model is ill-conditioned. On the other hand the input disturbance model shows robustness to uncertainties, and the optimal disturbance model obtained with the technique proposed is close to the input disturbance model. Application to a well-known ill-conditioned distillation column is presented.