To main content

NAICNO/wp7-UC3-pseudo-hamiltonian-neural-networks: v1.0.0 — UC3 Pseudo-Hamiltonian Neural Networks

Abstract

Decomposes dynamical system dynamics into conservation, dissipation, and external force components using separate sub-networks with port-Hamiltonian structure. The approach outperforms standard neural networks on dynamical systems benchmarks and produces models that remain valid when external forces are modified. Based on the phlearn package from SINTEF.

Category

Dataset

Language

English

Affiliation

  • SINTEF Digital / Mathematics and Cybernetics

Year

2026

Publisher

SINTEF Digital

View this publication at Norwegian Research Information Repository