System Identification 2017

Lecturer: Lucian Busoniu, TAs: Zoltan Nagy, Marius Costandin

Navigation: [Versiunea romana|Schedule|Course material|Labs|Lab tests|Project|Contact]          [Back to Lucian Busoniu's webpage]

About this course

The course introduces nonparametric methods for system identification, as well as parametric methods including prediction error and instrumental variables techniques. Completing the control engineer's identification toolbox, input signals, online recursive methods, and model validation are also discussed. The material is described at an appropriate BSc level, and builds in a self-contained manner the required mathematical background. The course is based on the book by Soderstrom and Stoica.

This course is part of the Bachelor program of the Automation Department, UTCluj (3rd year 1st semester). Prerequisites: linear dynamical systems and linear algebra. The lecturer is Lucian Busoniu, while Zoltan Nagy and Marius Costandin teach lab classes.

Grading:

The two lab tests and an original project solution are required before being admitted to the exam. The final grade will be a rounded, weighted average of all the evaluations per the percentages above. Note that each student can accumulate up to 105%x10 = 10.5 points; a grade of 10.5 still corresponds to 10 in the catalog.

Schedule

Dates and locations:

Due to interdependencies between lectures, labs, and project classes, the actual schedule is slightly different from the official one. Please look at it carefully to determine exactly when you should be in class:

Image with schedule table

Everything takes place in the room and time slot indicated in the official schedule, so you don't need to worry about changes there. However, some things are skipped in particular weeks, and the projects are done by everyone in the same week. In the table, "odd" and "even" means that originally you would have come in the odd or even weeks respectively; you should instead come when indicated in the table.

Course material

The lecture slides are mandatory reading; they will be written down in detail to give a self-contained, complete picture of the topics. They are made available here in time for each lecture.

In addition to the slides, followers may optionally consult the following books:

Labs

Lab tests

Each student is assigned a randomly chosen technique studied in the labs, and a randomly chosen dataset; the student must then apply the technique to the dataset. The solution consists of Matlab code, which must be submitted at the end of the test, and will be run and verified by the lecturer. The test is 1 hour long, and the solution must be developed in the first 50 minutes; the last 10 minutes are reserved for discussing the solution with the lecturer. To ensure the test is solvable in the time slot allowed, some techniques can be applied in a simplified manner. These simplifications will be explicitly indicated in the test material.

You are free to use the course material, including lecture slides, lab descriptions and data, and the startup example code which was made available for some labs. All this will be available offline on the computer on which you will take the test. The complete Matlab documentation is of course also available. However, the internet connection is disabled on the computer and you are not allowed to use internet-connected devices, or to reuse the solutions you developed for the labs.

Project: Black-box nonlinear identification

The project aims to move beyond the linear-system case treated in the lectures. Nonlinearities characterize virtually every real system, and in certain circumstances they cannot be sufficienty approximated by linear dynamics, making nonlinear models necessary. This project deals with a nonlinear variant of the ARX method. More details, including the deadlines for the solutions, can be found in the project description (PDF). Please read it carefully.

Contact

Comments, suggestions, questions etc. related to this course or website are welcome; please contact either the lecturer or the TA at the addresses below (given as images for spambot protection).
Image with email addresses