Publikation
Stable Port-Hamiltonian Neural Networks
Fabian Roth; Dominik K. Klein; Maximilian Kannapinn; Jan Peters; Oliver Weeger
In: Computing Research Repository eprint Journal (CoRR), Vol. abs/2502.02480, Pages 1-15, arXiv, 2025.
Zusammenfassung
In recent years, nonlinear dynamic system identifi-
cation using artificial neural networks has garnered
attention due to its manifold potential applications
in virtually all branches of science and engineering.
However, purely data-driven approaches often strug-
gle with extrapolation and may yield physically im-
plausible forecasts. Furthermore, the learned dy-
namics can exhibit instabilities, making it difficult to
apply such models safely and robustly. This article
proposes stable port-Hamiltonian neural networks,
a machine learning architecture that incorporates
the physical biases of energy conservation or dissi-
pation while guaranteeing global Lyapunov stabil-
ity of the learned dynamics. Evaluations with illus-
trative examples and real-world measurement data
demonstrate the model’s ability to generalize from
sparse data, outperforming purely data-driven ap-
proaches and avoiding instability issues. In addition,
the model’s potential for data-driven surrogate mod-
eling is highlighted in application to multi-physics
simulation data.