Michael Reichel (ari)
Jan Heiland(TU Ilmenau, Fachgebiet Optimization-based Control)
Numerical methods for nonlinear controller design with high-dimensional models address the complexity of the nonlinearities and the computational challenges of solving huge systems of equations. A general approach, that does not exist yet, would need to resort to elaborated numerical backends, to system-theoretic and data-driven model reduction, to heuristics, and possibly a bit of everything.
One promising approach is the embedding of the nonlinear system in so-called quasi linear parameter-varying (qLPV) systems which opens a direct path to decreasing the model complexity in terms of the nonlinearity while retaining high-dimensional linear structures for which suitable solvers exist.
Oftentimes, the needed embedding is induced by the model structure. For the subsequent reduction, one can call on classical model order reduction or recent developments that use techniques from machine learning. The resulting, structurally reduced, models can then be controlled with standard design techniques even more if combined with state-of-the-art linear algebra backends.
In this talk, I will give an overview over my recent research activities in this challenging and multi-faceted field.
In a second, more informal part of my presentation, I want to highlight my plans and motivations for research, teaching, and collaboration at our institute of mathematics and beyond.
Dienstag, 3. Juni 2025, 15 Uhr, Curie-Hörsaal