To path indicator Subpages of “Nachrichtendetail“ To navigation To quick access To footer with other services
To content
The Process Automation Systems Logo consists of a green box with rounded corners through which a set of trajectories are crossing as white lines. Next to the box is the name of the laboratory.
Fakultät BCI
Studierende sitzt am Arbeitsplatz und schaut auf den Laptop. © Aliona Kardash​/​TU Dortmund

New Publication in the Journal of Process Control: Advancing Real-Time Robust MPC for Mixed-Integer Nonlinear Systems

-
in
  • News
  • Events
  • Research
© Dalga Merve Özkan
Joshua Adamek and Lukas Lüken, under the supervision of Prof. Lucia, have published a new article in the Journal of Process Control. The work presents a novel learning-based approach that enables robust nonlinear model predictive control with integer decisions to be solved in real time, even under significant uncertainty.

The joint work of Joshua Adamek and Lukas Lüken under the supervision of Prof. Lucia, titled “Enabling robust mixed-integer nonlinear model predictive control via self-supervised learning and combinatorial integral approximation”, has been published in the March 2026 issue of the Journal of Process Control (Volume 159).

Model predictive control (MPC) is a powerful optimization-based control strategy that is widely used in industrial applications. However, solving MPC problems in real time becomes particularly challenging when nonlinear dynamics, discrete decisions, and uncertain operating conditions must all be considered simultaneously. In this work, the authors address this challenge for the demanding class of mixed-integer nonlinear systems.

The proposed methodology combines three key ideas: combinatorial integral approximation, scenario-based decomposition for uncertainty handling, and self-supervised learning for accelerated optimization. By decomposing the robust MPC problem into many smaller subproblems that can be solved in parallel and accelerated on GPUs using learned iterative solvers, the approach achieves significantly faster computation times while maintaining high solution quality.

Simulation studies on an uncertain nonlinear reactor demonstrate that the proposed framework enables real-time applicability for robust mixed-integer nonlinear MPC and achieves an order-of-magnitude speedup compared to conventional numerical optimization methods.

The publication further highlights the integration of combinatorial integral approximation into robust multi-stage MPC and the use of self-supervised learning for GPU-based acceleration.

You can read the full article here.