Schedule for System Identification
The course starts on Wednesday, April 9, 2014, at 13.15 in Algoritmen.
All lectures will be in Algoritmen
Date | Notes | Chapters | Problems |
---|---|---|---|
9/4 13.15 | Curve fitting | ... | ... |
16/4 13.15 | Linear Models | Kap 1,2,3 | 2G.4, 2E.2, 2E3, 3G.2, 3E.4 |
*Th 24/4 | Nonlinear Models | Kap 4,5 | 4G.4, 4G.5, 4E.1, 4E.3, 5G.1, 5E.2 |
30/4 13.15. | ... | 6,7 | 6G.1, 6G.3, 6G.5, 7G.1, 7G.4,7E.1, 7E.3 |
*We 7/5 8.15 | ... | 8,9 | 8G.2 8G.3, 8E.3, 9G.2, 9E.3, 9E.4 |
*To 15/5 15.15 | ... | Data Discussion | .. |
*We 21/5 8.15 | Practical Issues | Kap 10,11 | 10G.3,10E.1,10E.3,11E.4,11D.1,11E.3 |
28/5 13.15 | ... | 12,13,14 | 13G.2,13E.1,13E.3,13E.5.14G.2 |
*To 5/6 15.15 | ... | 15,16,17, Summary | 15E.1,16G.1,16E.2,16E.3 |
*To 12/6 15.15 | ... | Reports on micro-projects;** Questions | .. |
** The lecture on June 12 will be in Systemet
Note that I will give a summary the second half of the second last meeting. Part of the last meeting will be devoted to questions on the whole material.
The exam period will be from 11/6 to Midsummer.
The organization of the lectures is as follows: The material of the chapters will be discussed during the first half of the lecture on the indicated day. It is assumed that the participants have read and digested the material before that and also have solved the indicated problems. The participants should be prepared to take active part in the discussion of the material. The second half of the lecture will be used to introduce next week's material.
It is recommended that the participants meet an additional time per week, without the lecturer, to discuss the problems and prepare the discussion for the next lecture.
Data tests
Working with real data sets will be an essential part of the course. The participants should form teams of 2-3 persons and select a data set fromand work with it during the first few weeks. Preferably the work should be carried out using the System Identification Toolbox. (Matlab 8 or higher)
Project ideas:
1. Investigate what the MATLAB-package LOLIMOT can do for LOcal LInear
MOdels. How does it compare with Tree- NLARX models?
2. Investigate some new ideas to improve spectral estimated with
polynomial and rational interpolations. (Schoukens, Pintelon,
Hjalmarsson, McKelvey)
3. Investigate how LPV-models can be developed from Local Linear
Models.
4. If we use non-identifiable model structures, like fully
parameterized state-space models, in combination with
L1-regularization (Lasso), what parameterizations will we converge to?
5. Matlab has a "Global Optimization" Toolbox. Investigate how
such techniques can be used to avoid problems with local minima in
identification problems.
6. "Elastic Nets" are combinations of L1 and L2 regularization. The
Statistics Toolbox has mfiles for elastic nets for linear
regression. Investigate how that works for ARX identification
problems.
7. Arch and Garch models (ARX models with varying noise level) have
been very popular in Econometrics since the "Nobel Prize" in
2003. Investigate how those techniques will handle ARX-problems in
system identification, compared to ignoring the noise variation.
8. How would data tapering work when applied to ARX models?
Informationsansvarig: Lennart Ljung
Senast uppdaterad: 2015-12-09