Abstract
Dynamic models play a central role in MPC technology. The assumption during MPC design is that a reliable model of the plant under consideration is available. Normally identification methods deliver a nominal model of the plant with unknown dynamics. The performance achieved by this controller when applied to the plant is highly dependent on the accuracy of this model. However, in real life situations, models that are identified from open loop data are generally contaminated with errors and are inaccurate for many reasons such as equipment degradation (e.g. catalyst change, heat exchanger fouling etc.), low quality measurement data etc. Problem also arises when these models start exhibiting degradation in their performance after some span of time. A number of reasons can be attributed to this phenomenon like change in process operating conditions, drift in process conditions, environmental conditions, instability and inherent feedback mechanisms of the plant. Examples of such processes are refineries, where a change in the crude oil flow
