Identification of Dynamical Systems

lecturer: Ivan Markovsky, teaching assistant: Mehrad Sharabiany

Slides

Exercises

Reading materials

Mini-projects

Implementation of system identification methods

The objective is to implement and test subspace and maximum-likelihood system identification methods described in the course.

Benchmarking of system identification methods

The objective is to compare system identification methods from

on the benchmark examples from the DAISY dataset.

For a similar comparison, see Section 8 in

I. Markovsky. A software package for system identification in the behavioral setting. Control Eng. Practice, 21:1422–1436, 2013.

Extracting physical parameters from an identified black-box model

The identification methods taught in the course deliver "black-box" models, i.e., the models have no special structure. Structured models, also called "grey-box" and "white-box" models, may be interpretable. As an example consider a mass-spring-damper system. The identified model, given by a generic state-space or transfer function representation, does not make the physical parameter mass evident. The objective of the mini-project is to investigate this problem and propose possible solutions.

Reference:

G. Mercère, O. Prot and J. A. Ramos, "Identification of Parameterized Gray-Box State-Space Systems: From a Black-Box Linear Time-Invariant Representation to a Structured One," in IEEE Transactions on Automatic Control, vol. 59, no. 11, pp. 2873–2885

Comparison of direct and indirect data-driven control

This mini-project goes beyond the material taught in the course. The objective is to reproduce the results in

F. Dörfler and J. Coulson and I. Markovsky. Bridging direct & indirect data-driven control formulations via regularizations and relaxations, 2021