Nathalie Janes

University of Freiburg

Tuesday, November 24, 2026, 11:00 - 11:59

SR 01-012

This master thesis investigates decomposition methods for solving Mixed-Integer Nonlinear Programs (MINLPs) that arise from discretizing Optimal Control Problems (OCPs) where both continuous and integer controls influence the dynamics. In closed-loop Model Predictive Control (MPC), those MINLPs must be solved repeatedly over short time intervals. Since solving the MINLPs directly with a standard MINLP solver is unsuitable in the context of MPC, because computation times can get too high, we need another approach.

We study a decomposition approach, where we first solve a relaxed NLP, then reconstruct integer controls and lastly solve a polishing NLP. We present five different methods of reconstructing the integer controls from the solution of the relaxed NLP and review their theoretical foundation.

These decomposition methods are then applied in a closed-loop MPC framework to test examples, where we analyse performance and computation time.