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]
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.
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:
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.
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:
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.
For the first test, the slots are assigned as follows: Lab test 1 schedule (PDF), where the format is: day in the week of Nov 13-17 / hour. The test is during the normal lab slot. Please be on time; the one-hour interval cannot be exceeded since your colleagues are starting the test immediately after that. If for well-founded reasons you cannot make it on the assigned slot, please contact the lecturer as soon as possible with a request to change the slot. Reassignments cannot be made if they result in more than 8 students taking the test in the same time slot. Note that before test 1, everyone should have labs 2-6 submitted; if not please correct this ASAP.
For the second test, the slots are assigned as follows: Lab test 2 schedule (PDF), where the format is: day in the week of Jan 15-19 / hour. The rules are exactly the same as for the first test, but now the subjects are from labs 7-11.
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.
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).