Nonlinear MPC and inferential sensing for PVC production

A nonlinear model-based predictive controller with new features including range control, multirate sampling and two degrees of freedom control for PVC batch control/optimization is introduced. The lumped uncertainty design of the semi-empirical batch model based on data-mining technologies is described. The controller is complemented by an inferential sensor to reduce the uncertainty about the reactivity of the batch. The knowledge of reactivity is used for batch time optimization under constraints given by shared cooling resources of the plant.