Data-driven machine learning approaches have gained traction for nonlinear system identification, yet standard models often fail to retain the underlying physical structure and can be difficult to interpret, particularly in the absence of an analytical model. In this context, port-Hamiltonian (pH) models offer a natural physics-informed representation. However, when parameterized with standard multilayer perceptrons (MLPs), the learned constitutive components often remain poorly interpretable. This paper proposes a structure-preserving identification framework for nonlinear port-Hamiltonian systems based on Kolmogorov-Arnold Networks (KANs). The proposed PH-KAN model parameterizes the interconnection matrix, dissipation matrix, Hamiltonian, and input mapping using dedicated KAN blocks while enforcing the port-Hamiltonian constraints by construction. This results in constitutive representations where the nonlinear functions defining the identified pH components can be explicitly inspected, leading to a more interpretable model than with standard MLP-based parameterizations.
Blogger's Review: The PH-KAN model significantly enhances the interpretability of port-Hamiltonian systems by integrating KAN structures, overcoming the limitations of traditional MLPs. This offers a novel perspective for nonlinear system identification and showcases the immense potential of combining physical insights with machine learning.