Abstract:

Nonlinear model predictive control (NMPC) is a technique for feedback control of nonlinear systems that is based on optimization of the predicted future system behaviour, using a first-principles differential equation model. A practical application of NMPC has to overcome the difficulty to solve the occuring nonlinear optimal control problems reliably and in real-time.

In this talk, a newly developed algorithm to achieve this aim is presented. The approach is based on the direct multiple shooting method for dynamic optimization and is characterized by a dovetailing of the iterative solution procedure with the system dynamics in a very fast "real-time iteration" scheme.

The performance of the algorithm is demonstrated for the NMPC of a pilot plant distillation column situated in Stuttgart, using an optimization model with over 200 states.

In a numerical experiment, the algorithm is also tested for the NMPC of a looping kite, which shows its good performance and robustness even for unstable periodic systems.