How to make Nonlinear Model Predictive Control real-time feasible for fast processes

Event details
Date | 21.11.2014 |
Hour | 10:15 |
Speaker | Georg Bock |
Location | |
Category | Conferences - Seminars |
NMPC - Nonlinear Model Predictive Control - is probably one of today’s most promising model-based approaches to deal with nonlinear processes under uncertainties. Its simple but challenging idea is to use detailed nonlinear dynamical models to predict the performance of the process to be controlled over a sufficiently long time horizon. In the omnipresence of per-turbations and model-plant mismatch, an optimal control law must then be repeatedly com-puted online, which ensures that the economic and/or tracking performance of the process is optimized subject to all kinds of important restrictions.
However, even the fastest "all-at-once" optimization solvers available today are much too slow to compute an optimal control quick enough to be used in very fast and/or very complex processes with correspondingly complex models, as they are frequently met, e.g., in produc-tion processes or in vehicle dynamics.
The lecture will review some basic concepts of NMPC including theoretical results showing that global stability can be achieved by economic performance criteria. It will then outline some of the ideas behind the new so called “multi-level real-time iterations” and “online ac-tive set” optimization algorithms that have produced a speed-up by 4 – 5 orders of magnitude over earlier approaches to NMPC. Extensions to mixed-integer optimal control will be sketched.
The performance of the new methods will be demonstrated - with the support of some videos - by engineering applications such as distillation processes, wind power generating kites, and energy conservation in heavy duty trucks.
However, even the fastest "all-at-once" optimization solvers available today are much too slow to compute an optimal control quick enough to be used in very fast and/or very complex processes with correspondingly complex models, as they are frequently met, e.g., in produc-tion processes or in vehicle dynamics.
The lecture will review some basic concepts of NMPC including theoretical results showing that global stability can be achieved by economic performance criteria. It will then outline some of the ideas behind the new so called “multi-level real-time iterations” and “online ac-tive set” optimization algorithms that have produced a speed-up by 4 – 5 orders of magnitude over earlier approaches to NMPC. Extensions to mixed-integer optimal control will be sketched.
The performance of the new methods will be demonstrated - with the support of some videos - by engineering applications such as distillation processes, wind power generating kites, and energy conservation in heavy duty trucks.
Practical information
- Expert
- Free