Recent Results on Nonlinear System Identification Benchmarks

Description

This is an open invited track for the IFAC SYSID 2027 conference, taking place July 7-9 2027, in Lyon, France.

The objective of this invited track is to present novel data-driven modeling approaches for nonlinear systems. Solicited contributions should illustrate the developed methods and techniques on one or more nonlinear system datasets that are available on the nonlinearbenchmark.org platform. This open invited track aims, by promoting interaction through the benchmark setups, to develop a better understanding of and insight in state-of-the-art data-driven modeling approaches.

This open invited track is structured around the benchmark systems listed on the website: www.nonlinearbenchmark.org, featuring 15 nonlinear system datasets covering well-established benchmark challenges in the nonlinear system identification community (e.g. the Silverbox and the Wiener-Hammerstein datasets) as well as recent challenging datasets of real life systems (e.g. the NanoDrone and Fine Steering Mirror datasets). Solicited contributions should describe solutions to one or several of these benchmark problems. The nonlinear benchmark website functions as a platform providing detailed information regarding the benchmark problems, and making available numerical and experimental data sets together with identification and validation guidelines. For most of the datasets, this is formalized using a set of Python data loader functionalities and figure of merit calculation functions are provided through a GitHub repository. These scripts simplify the process of downloading, loading, and splitting various datasets available on the website.

Scope and Topics of Interest

The open invited track welcomes theoretical, methodological and experimental contributions that are applied or explore on or more of the benchmark datasets available on nonlinearbenchmark.org. Relevant topics include, but are not limited to:

• New nonlinear system identification algorithms that are validated on one or more of the featured datasets. This includes nonlinear, linear-parameter-varying, time-varying identification approaches, as well as linear identification based analysis methods.

• New data-driven model reduction, physics-informed and deep learning approaches for dynamical systems, validated on one or more of the featured datasets.

• Comparative studies of mulitple state-of-the-art and baseline system identification approaches on multiple benchmark datasets.

• New experiment design and active learning approaches validated on one or more of the benchmark systems, or models of the featured datasets.

Submitting a Contribution

Please inform the organizors if you are planning to submit a contribution to the open invited track!

Conference website: https://conferences.ifac-control.org/sysid2027/

Submission website: https://ifac.papercept.net/

Invited session identification code: TBA

Submission Deadline: Nov, 2nd, 2026

Like regular papers, OIT papers must present unpublished work and must not be currently under consideration for publication elsewhere. Initial submissions are at least 4 pages long and at most 8 pages long, with a strict 6 pages limit for final manuscripts to be included in the IFAC-PapersOnLine proceedings.

When submitting your open invited track contribution, choose the 'Open Invited Track Paper' option on the submission website. Once this option is chosen, you will have to specify the session identification code which is listed above.

In case you would like to contribute to the invited track, but submit a 'regular Paper with journal option', 'dissemination paper', or 'discussion paper' instead, please contact Maarten Schoukens (m.schoukens@tue.nl). Full details about these submission options can be found on the conference website. Note that these contribution options have different submission deadlines compared to the regular OIT paper contribution.

Organizors