Vortrag im Kolloquium Technische Kybernetik

An Introductory Overview to Output Feedback Nonlinear Model Predictive Control

Dipl.-Ing. Rolf Findeisen

    Time: Tuesday · 12. 2. 2002 · 16:00h
    Place: Room V 9. 31 · Pfaffenwaldring 9 · Campus Stuttgart-Vaihingen

Abstract

In the past decade, nonlinear model predictive control (NMPC) has witnessed steadily increasing attention from control theoretists and control practitioners. The practical interest is driven by the fact that today's processes need to be operated under tighter performance specifications. Often these demands can only be met when process nonlinearities and constraints are explicitly considered in the controller.
By now, several NMPC schemes exist that guarantee stability if the full state information is available. Often in practice, however, not all states needed for the application of NMPC are available by measurement. Thus, they must be recovered by a suitable state observer from output measurements. While, in principle, the combination of a stable state estimator with a stabilizing NMPC scheme does lead to asymptotic stability, nothing in general can be said about the resulting region of attraction. Often only local stability can be guaranteed. This problem is of substantial practical relevance, since even small observer errors can lead to significant performance decrease or possible instability of the closed loop.
This talk examines the output feedback NMPC problem. After a short review of NMPC and an introduction to the output feedback problem we turn to the question of how to achieve more than local stability in the output-feedback case. After outlining several possibilities to achieve this goal we focus on the use of high gain observers in combination with nonlinear predictive controllers. It is shown that under certain conditions the combination of a high-gain observer with an NMPC controller does lead to stability of the closed loop with a non-local region of attraction. Furthermore, the performance of the state feedback predictive controller can be recovered up to an arbitrary degree by increasing the observer gain and decreasing the NMPC controller sampling time.
The resulting controller is applied to an inverted pendulum-car system to explicate some of the key properties. The talk concludes with possible generalizations to the proposed output feedback NMPC controller.

Biographical Sketch

Rolf Findeisen graduated with an engineering degree in Technical Cybernetics from the University of Stuttgart in 1998 and an M.S. degree in chemical engineering from the University of Wisconsin, Madison in 1997. From mid 1997 to the end of 1999 he was a research assistant at the Automatic Control Laboratory at the Department of Electrical Engineering at ETH Zürich. At the end of 1999 he joint the newly founded Institute for Systems Theory in Engineering at the University of Stuttgart. His main research interests are in the area of nonlinear model predictive control, output feedback control and optimization based control.