![[2602.15704] Controlled oscillation modeling using port-Hamiltonian neural networks](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.15704] Controlled oscillation modeling using port-Hamiltonian neural networks
Summary
This paper presents a novel approach to modeling controlled oscillations using port-Hamiltonian neural networks, emphasizing a second-order discrete gradient method for improved performance over traditional numerical methods.
Why It Matters
The research addresses challenges in learning dynamical systems through data-driven methods, particularly in preserving underlying conservation laws. By enhancing port-Hamiltonian neural networks with a discrete gradient method, the findings could lead to more accurate modeling in various applications, including robotics and control systems.
Key Takeaways
- Introduces a second-order discrete gradient method for modeling.
- Demonstrates improved performance over traditional Runge-Kutta methods.
- Explores various dynamical behaviors through specific oscillator models.
- Highlights the importance of power-preserving discretizations in neural networks.
- Analyzes the impact of Jacobian regularization during training.
Computer Science > Machine Learning arXiv:2602.15704 (cs) [Submitted on 17 Feb 2026] Title:Controlled oscillation modeling using port-Hamiltonian neural networks Authors:Maximino Linares, Guillaume Doras, Thomas Hélie View a PDF of the paper titled Controlled oscillation modeling using port-Hamiltonian neural networks, by Maximino Linares and Guillaume Doras and Thomas H\'elie View PDF HTML (experimental) Abstract:Learning dynamical systems through purely data-driven methods is challenging as they do not learn the underlying conservation laws that enable them to correctly generalize. Existing port-Hamiltonian neural network methods have recently been successfully applied for modeling mechanical systems. However, even though these methods are designed on power-balance principles, they usually do not consider power-preserving discretizations and often rely on Runge-Kutta numerical methods. In this work, we propose to use a second-order discrete gradient method embedded in the learning of dynamical systems with port-Hamiltonian neural networks. Numerical results are provided for three systems deliberately selected to span different ranges of dynamical behavior under control: a baseline harmonic oscillator with quadratic energy storage; a Duffing oscillator, with a non-quadratic Hamiltonian offering amplitude-dependent effects; and a self-sustained oscillator, which can stabilize in a controlled limit cycle through the incorporation of a nonlinear dissipation. We show how the ...
