System identification of port-Hamiltonian systems

Abstract: System identification is essential in modeling, analysis, and control of dynamical systems, particularly when first-principles models are incomplete or unavailable. In this talk, we begin with a brief introduction to system identification, outlining its main objectives, challenges. We then focus on structured modeling frameworks, with particular emphasis on port-Hamiltonian systems, which have attracted significant attention due to their strong ties to physics, energy-based interpretation, and interesting properties for control and stability analysis. We study system identification under explicit structural and physical constraints, using the port-Hamiltonian formalism as a unifying framework, starting with the identification of linear port-Hamiltonian systems, and highlighting how structure-preserving approaches can be leveraged to recover physically consistent models from data. We then move to nonlinear port-Hamiltonian systems and discuss recent methods that enable their learning from data, including generalizations to higher-order and more complex systems through neural scaling laws. The talk concludes with a discussion of current research directions, including recently proposed architectures for learning port-Hamiltonian systems.

mathematical physicsanalysis of PDEsdifferential geometrydynamical systemsfunctional analysisnumerical analysisoptimization and controlspectral theory

Audience: researchers in the discipline

Port-Hamiltonian Seminar

Series comments: Slides and recordings can be found here: uni-wuppertal.sciebo.de/s/CQfBsXr9iOI17ZY