Constrained receding horizon predictive control of a binary distillation column

A state space predictive control algorithm, with state constraints at the end-point, is evaluated on a real plant showing nonlinear behaviour. Predictive control techniques based on linear state space model description can find difficulties when applied to a real system with nonlinear behaviour, for example the controlled system may present a steady state offset or bias in the step response. We illustrate, in a real application, a multivariable extension of the constrained receding horizon predictive control (CRHPC) with an error correction on the set point in such a way to avoid the above problem. Specifically, we estimate with a Kalman filter the error between the model prediction and the real response of the plant. The benchmark in this paper is a laboratory distillation column.