Efficient numerical solution methods for nonlinear model predictive
control and the effect of numerical approximations on stability
Dr Rolf Findeisen (University of Stuttgart)
The purpose of this talk is twofold. In the first part we briefly review
the main ideas and principles of nonlinear model predictive control.
As shown, one of the main obstacles for the application of nonlinear
model predictive control is that for practical implementation an often
large-scale optimal control problem must be solved in real-time. Thus,
we review in the second part efficient on-line solution methods for
nonlinear model predictive control. The efficiency of one specific
solution method is underlined considering the real-time control of a
high-purity distillation column. Furthermore, we discuss the influence
of numerical approximations and errors on stability and performance. For
this purpose we consider one specific real-time implementation of
nonlinear model predictive control, which performs only one SQP
iteration per sampling-time. We underline that under certain conditions
the resulting closed-loop, consisting of the system to be controlled and
the numerical iteration scheme, is stable.