System Identification 2022-2023

Lecturer: Lucian Busoniu, TAs: Zoltan Nagy, Tudor Santejudean, Bilal Yousuf

Navigation: [Versiunea romana|Evaluation rules|Schedule|Lectures|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. Input signals, online recursive methods, closed-loop identification, 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 System Identification by Soderstrom and Stoica.

This year, we will use a mix of offline teaching and online platforms: Microsoft Teams for interacting with students, and ClassMarker to take the quizzes. As a unique ID for each student, we will use use an email address that is associated to a ClassMarker account. Microsoft Teams may be associated with a different, Didatec email address (or you can of course use the Didatec email for everything). Details on how we are using these platforms can be found under the specific type of activity below.

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. Zoltan Nagy teaches the projects, while Tudor Santeajudean and Bilal Yousuf teach the lab classes.

Evaluation rules

Grading:

• 30%=2x15% two lab tests, see their section below for details.
• 10% lab quizzes: a short quiz with two questions, at the start of each lab except the first, from the material relevant to that lab.
• 30% final exam.
• 30% project (15% part 1 + 15% part 2), see below for details.
• 10% lecture quizzes: short quizzes during each lecture, from the material discussed during that particular lecture.

Eligibility conditions: Solutions to all labs except the first one; the two lab tests; and solutions to both parts of the project are all required before being admitted to the exam. More details about each component (labs, lab tests, project evaluation) will be found at the appropriate places in their separate descriptions below.

Schedule

Lectures are Wednesdays from 8AM, labs and projects on Mondays and Tuesdays. Please look at the schedule below carefully to determine exactly when you should be in class, and see the overall schedule on http://www.aut.utcluj.ro/ for when your time slots are allocated.

Note 4 Dec 2022: Schedule has been updated per the discussion on Teams.

Due to interdependencies between lectures, labs, and project classes; as well as lab tests and project deadlines, the actual schedule is slightly different from the official one. The following changes are performed:

• Since we need lecture 2 for lab 2, and labs come before lectures, we postpone lab 2 to week 3 (and we recover the delay in week 9).
• During lab test weeks, we may need to use project slots for the test as well, so that everyone has time to take the test.
• Since week 9 has two national holidays, we will not work on projects or lectures that week.
• For project session 4, the odd-week halfgroups take the project in an even week (number 8). The labels "odd" and "even" in the table refer to the time-slots in which you would come according to the official schedule.
• The last lecture slot is a backup for project presentations that do not fit in the regular project slots of the previous week.

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.

• Part 1: Introduction to System Identification (PDF).
• Part 2: Mathematical Background: Linear Regression. Probability Theory and Statistics (PDF).
• Part 3: Transient Analysis of Step and Impulse Responses (PDF).
• Part 4: Correlation Analysis (PDF).
• Part 5: ARX Identification (PDF).
• Part 6: Input Signals (PDF).
• Part 7: General Prediction Error, ARMAX, and OE Methods (PDF).
• Part 8: Instrumental Variable Methods. Closed-Loop Identification (PDF).
• Part 9: Recursive Identification (PDF).
• Part 10: Model Validation and Practical Issues (PDF).

At a random point during a lecture, a short quiz will be given from the material of that lecture, using the ClassMarker platform. A list of email addresses will be collected beforehand and you will be invited to ClassMarker. The quizzes are graded! So, you should pay attention during the lectures. If you answer all the questions in every lecture correctly, you get 1 point in your grade; lower scores scale linearly to between 0 and 1 points in the grade.

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

• T. Soderstrom and P. Stoica. System Identification. Prentice Hall, 1989. The full text of this book is available at: http://user.it.uu.se/~ts/bookinfo.html. This book forms the basis of the course.
• L. Ljung, System Identification: Theory for the User, 2nd ed., Prentice Hall, 1999. An advanced graduate-level textbook.

Logistics and platform

Labs will be developed using Matlab. PDF descriptions of the labs are given on this website.

Lab solutions are required starting from lab 2. It is not enough to be present at the lab in order to validate it; a complete, working, and original solution must be developed during the lab. Whether the solution is working will be verified by the teacher. Once this has been verified, you submit the solution via a dropbox link. Originality will then be verified as described below. At most two labs total can be recovered before the exam, including at most one copied lab; hence, accumulating three or more missing labs means you can no longer become eligibile for the exam.

Half-group boundaries are strict, for example you cannot access the quiz to the lab unless you are present during the lab slot of your halfgroup.

Lab quiz

Each lab except the first starts with a 2-minute quiz containing 2 short questions, which tests you on the lecture material relevant to that particular lab. If you answer everything correctly for all the labs, you get 1 point in your grade; lower scores scale linearly, as for the lecture quizzes. So, you should arrive prepared.

Deadline and on-time bonus

We very much prefer that students validate their lab as working and upload it to dropbox during their nominal 2-hour slot. To promote this, for each lab that satisfies this condition, the submitting student gets 0.1 points bonus in their final grade.

The final deadline to validate and submit the lab is on Wednesday on the week after the lab, at 8AM. For example, since lab 2 will be held on Monday the 17th and Tuesday the 18th of October, you are allowed to validate it with the teacher either during these days, or during your next lab on 24th or 25th, and can upload it on Wednesday 26th at 7:59AM at the latest.

Originality check

Once the upload link closes, the solutions to each lab are run through an automated plagiarism check. Each suspicious case is examined by the teachers, and if it becomes clear that the lab is copied, both the source and the destination lab are invalidated. In addition, we are imposing a "two strikes - you're out" rule. Namely, the first lab copied is only invalidated, and you can recover it at the end. The second lab copied invalidates the full set of solutions, you forfeit the discipline and have to take it next year.

Lab descriptions

• Lab 1: (Re)Introduction to Matlab (PDF) -- two short exercises in order to get re-acquainted with Matlab. For a brief intro to Matlab, have a look at the following document (with thanks to Paula Raica for allowing us to use it) ST_Lab1.pdf. You may also use the automated Matlab tutorial, with thanks to Tassos Natsakis who developed it.
• Lab 2: Linear Regression for Function Approximation (PDF). The data files are: #1, #2, #3, #4, #5, #6, #7, #8, #9, #10, #11, #12, #13, #12, #15, #16.
• Lab 3: Transient Analysis of Step Responses (PDF). The data files are: for first-order systems, #1, #2, #3, #4, #5, #6, #7, #8 (where # stands for index). For second-order systems, #1, #2, #3, #4, #5, #6, #7, #8.
• Lab 4: Transient Analysis of Impulse Responses (PDF). The data files are: for first-order systems, #1, #2, #3, #4, #5, #6, #7, #8. For second-order systems, #1, #2, #3, #4, #5, #6, #7, #8.
• Lab 5: Correlation analysis (PDF). The data files are: #1, #2, #3, #4, #5, #6, #7, #8.
• Lab 6: ARX Identification (PDF). The data files are: #1, #2, #3, #4, #5, #6, #7, #8.
• Lab 7: Pseudo-Random Binary Sequences (PDF). Download the system simulator: system_simulator.p, and the validation input: uval.mat.
• Lab 8: Identification of OE models with the Gauss-Newton method (PDF). The data files are: #1, #2, #3, #4, #5, #6, #7, #8.
• Lab 9: Instrumental Variable Methods (PDF). The data files are: #1, #2, #3, #4, #5, #6, #7, #8.
• Lab 10: Recursive Identification (PDF). The data files are: #1, #2, #3, #4, #5, #6, #7, #8.

• 6 points: code for redoing a (part of a) lab between 2 and 5, i.e., transient analysis, linear regression, or correlation analysis.
• 4 points: one-on-one question(s), both on the code and testing the higher-level insight about the methods in the labs.

Lab test 2: The same structure as for test 1, except now from labs 6 to 10.

The grade at each lab test is added with a 0.15 weight to the final grade, irrespective of its value.

Project

See the project description (PDF) for the topic, rules, and deadlines. Please read it carefully. A detailed description of the linear regression method needed in the first part of the project can be found in last year's lecture; we will talk about it in lecture 2. Alternately, here is a one-page (handwritten) summary.

Guidelines for writing your report and developing and giving your presentation are also available.

Comments, suggestions, questions etc. related to this course or website are welcome; please contact either the lecturer or the TAs via Teams or email.