Real-Time Strategies for Optimization in Nonlinear Model Predictive Control
Moritz Diehl (University of Heidelberg)
Nonlinear model predictive control(NMPC) techniques require the
reliable solution of subsequent dynamic optimization problems in real time.
In this talk, efficient direct multiple shooting
strategies are presented that exploit the specific structure of the
subsequent NMPC optimization problems. These techniques comprise the
off-line precalculation of derivative information as well as the
re-use of solution information from one optimization problem to the
next.
A 'dovetail connection' of the iterative optimization
algorithm with the real process dynamics
is described that may reduce response times considerably, though at the
expense of optimality.
Numerical simulations of the nonlinear model predictive control of a
continuous stirred tank reactor are presented, as well as preliminary
results for a distillation column described by differential algebraic
equations.
Back to
Control Seminars Page.