Cultivation of model predictive control systems

In the past, most research in process control has been in the design and application of control systems. Recently, System Cultivation has been studied as a framework for the continual monitoring and improvement of process units; however, techniques for advanced control systems have been given little attention. One class of such control systems is Model Predictive Control (MFC) whose performance is dependent upon model accuracy. A main problem in determining when to update MFC models is unmeasured disturbances, which certainly affect most real applications.

This work applies the philosophy of System Cultivation to the problem of updating the models of MFC and of Dynamic Matrix Control (CMC), in particular. In doing so, a general technique is developed to determine when updating should occur in the face of unmeasured disturbances. In addition, an existing identification technique for DMC and MAC is extended to pure integrating systems. Since this work relies heavily on simulation, a multiunit dynamic simulation framework is also developed.