Software, Data &
Background Material

Nonlinear System Identification Benchmark Platform

A list of hosted nonlinear dynamical system datasets can be found on nonlinearbenchmark.org. Over 12 system datasets are featured, examples include an industrial robot arm, an electro-magnetical positioning system, and a cascaded tanks system with overflow. We also offer a set of supporting scripts for data loading and results reporting for a selected set of benchmark systems. The website also includes an overview of results obtained on a subset of the benchmark systems, as submitted by the researchers. 

The website also acts as the platform for the yearly workshop and Ph.D. mini-course on nonlinear system identification.

DeepSI

DeepSI offers a powerful Python toolbox to perform Deep System Identification (DeepSI) with a wide range of tools and methods.

The DeepSI Python module aims to offer an intuitive machine learning environment for system identification without the need for deep expert knowledge. Implementing a system identification task often requires effectively no more than 10 lines of code. Coding examples can be found on the GitHub page of the toolbox (https://github.com/MaartenSchoukens/deepSI).

Multiple projects have been making use of this toolbox, some examples include:

1. Identification of nonlinear state-space models starting from input-output data (link)

2. System identification using video data (link)

3. Identification of steering dynamics in an autonomous car (link)

4. Identification of Koopman representations of nonlinear systems starting from input-output data only (link)

You can download the toolbox here: https://github.com/MaartenSchoukens/deepSI

Please cite this toolbox as: 

Gerben Beintema, Roland Toth, Maarten Schoukens. Deep subspace encoders for nonlinear system identification; Automatica, vol. 156, 111210, 2023. Published Version

Other Material

Below I include some links to other relevant material linked to (nonlinear) system identification. This material is not authored by me or my research group.

1. Yearly week-long Ph.D. Spring School on Spring School On Data-Driven Model Learning for Dynamic Systems, organized by Xavier Bombois (CNRS, Laboratoire Ampère), Hugues Garnier (Université de Lorraine, CRAN), Marion Gilson (Université de Lorraine, CRAN)  and Guillaume Mercère (Université de Poitiers, LIAS).