Experimental comparisons between generalised predictive control algorithms using CARMA and CARIMA models

Generalised predictive control algorithms were initially designed using the CARIMA model to eliminate offset errors. They have been successfully used in many occasions. However, the experimental results from an application of these algorithms to an air-handling plant show that more stable control performance can be achieved when the CARMA model is used to generate the j-step-ahead predictions of the process output instead of the more commonly used CARIMA model. The offset errors caused by mismodelling can be eliminated by the use of a process gain adaptive scheme. Experimental results also show that the gain-adaptive scheme developed can give significant improvements on the global control performance in the presence of a nonlinear process gain when this is strongly linked to the operating conditions of the plant