Closed loop identification of MPC models for MIMO processes using genetic algorithms and dithering one variable at a time: Application to an industrial distillation tower

Model Predictive Controllers (MPC) typically use step response models. The identification of these models is usually carried out under Open-Loop conditions, where large quantities of data are collected and/or large process perturbations are used. This makes the identification simpler, but is costly in terms of personnel requirements and degraded process performance during the test. This paper reports a methodology for identifying Multiple-Input Multiple-Output (MIMO) step response models while the process is operating under multivariable control. The identification of MIMO step response models can be achieved by adding an external test signal to one variable at a time. The method has been successfully applied to a distillation tower in a petroleum refinery. During the test the tower was controlled by an existing MPC with the constraint handling active.