System Identification - Fall 2011

Welcome to the course webpage of System Identification for fall 2011! The course comprises class-room teaching as well as computer exercises using Matlab.

General Information

Instructors: Roy Smith, Heinz Koeppl
Assistants: Stefan Almér, Preetam Nandy, Michael Unger
Time/Place: Wednesdays 10 am-12 noon, NO C 60

Schedule

Whenever slides were used in the classroom, they will be linked to this list. Moreover, any scribe of a lecture (see below) that becomes available will also be linked here.

* denotes preliminary scribes subject to revision by the scribers.

Lab Exercises

Exercises should be attempted before coming to the scheduled session.

Reference material (downloadable)

Material from the following papers has been discussed in the lectures. The papers are here so that you can read the details.

• Subspace-based multivariable system identification from frequency response data,
  Tomas McKelvey, Hüseyin Akçay and Lennart Ljung, IEEE Trans. Automatic Control, Vol. 41, No. 7, pp. 960-979, 1996.
• Closed-loop identification via the fractional representation: experiment design,
  Fred Hansen, Gene Franklin and Robert Kosut, Proc. American Control Conf., pp. 1422-1427, 1989.
• An indirect method for transfer function estimation from closed loop data,
  Paul M.J. Van den Hof and Ruud J.P. Schrama, Automatica Vol. 29, No. 6, pp. 1523-1527, 1993.
• Identification and control -- closed-loop issues,
  Paul M.J. Van den Hof and Ruud J.P. Schrama, Automatica, Vol. 31, No. 12, pp. 1751-1770, 1995.
• Lecture notes on IVM and 2-stage LS method Endogenous Regressors and Instrumental Variables by Prof John Rust, University of Maryland.

Functions (downloadable)

The following functions are provided to save you the time and trouble of coding them yourself. They are not optimized and so will not work well for very large data sets. You can assume that these functions will be present in the path when you submit your final. There is no need to include them in your subfunctions.

Midterm examination

The midterm results are available in the attached table. The grading scheme was as follows: points (0 or 1) were assigned for various features in your plots and these should be obvious from the labels. Your RMS error was evaluated on a scale of 0 to 6, and the assistants graded the quality of your explanations and presentation on a scale of 0 to 6. All of these were summed and then scaled to give a final 0 to 6 score. This could be taken as an estimate of your course grade, with the obvious proviso that the midterm counts for only 40%. For your interest, a histogram showing the grade distribution is also attached.

Final examination

NEWS: There will be a Question & Answer session on Dec 21, from 12:00-1:00pm in the lecture hall. No problem class will take place at that day.

The preliminary final results are shown in the attached table. The grading scheme is similar to the midterm and should be evident from the column headings. Please check the table for major discrepancies. You have until the 14th February to contact the grader to resolve any discrepancies. Note that grades were assigned on the basis of "demonstrated" features of your script. Several scripts crashed. The authors of those should contact the responsible grader and see if the problem is easily resolved.

Scribes

Every lecture should be scribed in LaTeX by a team of two students, based on their written lecture notes. The team will be different for each lecture and you may sign up for only one lecture. This serves as an exercise to reconsider the taught material and to provide lecture notes that can be used by the student of the class to prepare for their exam - for instance. Ambiguities or discovered mistakes need to be discussed with the instructor of the specific lecture.

The teams need to be formed and communicated together with the date for the desired lecture to Michael Unger. The scribes will be assigned on a first-come first-served basis. See scribe assignment above. Scribes will be part of the grading (see below). Please download the scribe template.

Grading

Grading is based on two Matlab/Simulink identification projects. You will have at least one week to work on these. For those of you who act as scribes there is the opportunity to earn a bonus grade of 15% (the maximum will be capped at 100% though). The weighting is as follows.

• Final project - 60%
• Midterm project - 40%
• Scribe - 15%

The overall course grades are given in the attached table. These were derived via the following formula:

percentage grade = min(100, (midterm/22)*40 + (final/25)*60 + scribe)

For your interest the grades are presented in a histogram.

Primary reference

• "System Identification; Theory for the User" Lennart Ljung, Prentice Hall (2nd Ed), 1999.

Secondary references

• "Dynamic system identification: Experimental design and data analysis", GC Goodwin and RL Payne, Academic Press, 1977.
• "Stochastic systems: estimation, identification and adaptive control", PR Kumar and P Varaiya, Prentice Hall, 1986. (pdf link).
• "System identification", Soederstroem and Stoica, Prentice Hall, 1989. (pdf link).