Structured Lipschitz Models: Definition, Examples, and Behavior

Dr. Ronald K. Pearson

   Zeit: Dienstag, 10. 07. 2001, 16:00
   Ort: Hörsaal V 9.31 Pfaffenwaldring 9, Universitätsbereich Stuttgart-Vaihingen

Abstract:

Many authors have noted the difficulty of developing the models required for nonlinear model predictive control (NMPC) and other model-based computer control strategies. Part of this difficulty lies in the extreme range of behavioral characteristics that are possible in the enormously broad class of nonlinear dynamic models. In particular, it is difficult to select nonlinear model structures that exhibit desirable qualitative behavior or exclude undesirable behavior because connections between nonlinear model structure and nonlinear behavior are not well understood. This difficulty motivates detailed examinations of specific nonlinear model structure classes.

This talk introduces the class of structured Lipschitz models, a relatively rich class of nonlinear discrete-time dynamic models that includes as special cases all nonlinear FIR models, both recurrent and nonrecurrent dynamic artificial neural networks, the Lur'e model class, and a variety of other less well-known model structures like the EXPAR and TARMAX classes. Further, the structure of this model class lends itself nicely to stability analysis and gives some interesting insights into the differences between BIBO stability and stronger stability notions (e.g., exponential stability) in these model classes.

Biographical Sketch:

Dr. Pearson received his PhD in electrical engineering from M.I.T. in 1982, working in the area of optimal fixed-structure compensators for distributed parameter systems. Upon completing his degree, he joined the DuPont Company in Wilmington, Delaware in the U.S.A. where he remained until 1997, working in a variety of areas including the development of on-line process measurement sensors, analysis of process operating data, and various practical aspects of dynamic model development (in particular, data pretreatment procedures and nonlinear model structure selection). In 1997, Dr. Pearson joined the Institut fuer Automatik at ETH Zuerich, where he taught courses on exploratory data analysis and nonlinear dynamic model development and completed two books: "Discrete-Time Dynamic Models," published by Oxford University Press in 1999 and "Identification and Control Using Volterra Models," co-authored with F.J. Doyle, III and B.A. Ogunnaike, to be published this summer by Springer-Verlag.