Optimization and Systems Theory Seminar
Friday, March 19, 2004, 11.00-12.00, Room 3721, Lindstedtsv. 25

Moritz Diehl
Interdisciplinary Center for Scientific Computing
University of Heidelberg
D-69120 Heidelberg

Real-time optimization of large scale systems

Nonlinear Model Predictive Control (NMPC) is a feedback control technique that uses a nonlinear dynamic process model for prediction and optimization. Feedback is obtained by using the observed system state as initial value of an optimal control problem on a prediction horizon, solving the problem online, and implementing the first part of the optimized control trajectory at the real process. The optimization is repeated after a short time, sufficiently fast to react to disturbances or to the effects of modelling errors.

A major challenge for any nontrivial NMPC application is the real-time optimization of large scale process models of differential algebraic (DAE) or partial differential equation (PDE) type. We present an efficient embedding technique to initialize subsequent problems, implemented in an online algorithm for NMPC, that has already been applied to experimentally control a pilot plant distillation column described by DAE. We show how the technique can be exteded to very large scale DAE systems, arising from the discretization of instationary PDEs by the method of lines, and apply this technique to a periodically operated chromatographic separation process.

Calendar of seminars
Last update: February 6, 2004 by Anders Forsgren.