Model Predictive Control (MPC) has established itself as the advanced control method of choice in industrial practice -- especially in the process industries -- due to its ability to handle systems with multiple inputs and outputs and economically motivated cost functions. Moreover, MPC is increasingly being used for control of fast systems described by nonlinear models, despite the high demands in terms of computing power that such applications entail.

In many applications where process constraints must be respected despite model uncertainties, control system designers are currently forced to rely on heuristics. For example, constraints may be artificially tightened. These heuristics result in suboptimal control performance and cannot guarantee that the intended control goals are reached. More advanced methods that have been developed in the field of robust MPC lead to optimization problems that are extremely difficult to solve online. Therefore, these methods have only rarely been used in real applications.

The aim of the two research projects in this project cluster is to develop new methods for robust MPC and to test them in realistic simulation studies and at a laboratory plant, with the aim that they will also be transferable to industrial practice. The main goals are to find less conservative and more efficient robust problem formulations, as well as to reduce the computation times for the online solution of the resulting optimization problems by tailored numerical methods.

The cluster is funded by DFG and comprised of the following two projects:

• P1: Numerical methods for ellipsoid based and tree sparse robust MPC formulations.
  M. Diehl, University of Freiburg
  DFG grant 424107692
• P2: New methods for robust MPC with high-dimensional uncertainty.
  S. Lucia, TU Dortmund
  DFG grant 423857295