Schedule

The course will be given during the periods 2018-10-17 -- 20181212 and 20190201-2019-03-15.

Part I: 2018-10-17 --- 2018-12-12

Basic Lectures

• Wednesdays 13.15-15.00  
• Lecture 3: Wednesday October 30, 13:15-15:00, Q36, Malvinas väg 6.
• Lecture 4: Wednesday November 7, 13:15-15:00, V23
• Lecture 5: Wednesday November 14, 13:15-15:00, Q36
• Lecture 6: Wednesday November 21, 13:15-15:00, Q21
• Lecture 7: Wednesday November 28, 13:15-15:00, V21, Teknikringen 72
• CANCELLED: Lecture 8: Wednesday December 5, 13:15-15:00, M23. Brinellvägen 64
• Lecture 8: Wednesday December 12, 13:15-15:00, M31, Brinellvägen 64

Extended Lectures (for FEL3202)

• TBD

Part II: 2019-02-01 -- 2019-03-15

Basic Lectures

In preparation for Lecture 9, read Chapters 2-5 in Ljung. Important in particular

• Chapter 2: Signal spectra (Section 2.3) including transformation  of signal spectra and spectral factorization.
• Chapter 3: Prediction (Section 3.2) including Lemma 3.1 and one-step ahead predictors, Observers (Section 3.3) , including a family of predictors.
• Chapter 4: A family of transfer function models (Section 4.2), in particular a general family of model structures. State-space models (Section 4.3), in particular innovations representation.  Formalia in Section 4.5. Section 4.4 is omitted. Section 4.6 (identifiability) will be covered in Lecture 9. 
• Chapter 5:  Read as an overview of general model structures. In particular, relate Ljung's view of a model (Section 5.4) to the pdf-model approach we have taken hitherto in the course (Subsection on "An other view of models").

• Lecture 9: Wednesday January 30,13:15-15:00. Q2
  • Identifiability (Section 4.6 )
  • The prediction error approach (Sections 8.1-8.5, 9.1-9.4)
  • The correlation approach (Sections 7.7, 8.6, 9.5)
• Lecture 10: Wednesday February 6,13:15-15:00. V11
  • Model structure selection (Chapter 16)
  • Model accuracy (Chapter 9, slides)
• Lecture 11: Wednesday February 13,13:15-15:00. Q2
  • Model accuracy continued (slides)
  • Computing the estimate (Chapter 10)
  • Iterative and multi-step methods, including subspace identification, and multi-step least-squares methods (parts of Chapter 7, Chapter 10, and lecture notes)
  • The Expectation-Maximization algorithm (Exercise 10G.3 in Ljung's book
• Lecture 12: Wednesday February 20, 13:15-15:00. M35
  • Nonlinear stochastic models (Guest lecture by Fredrik Lindsten Uppsala University)
• Lecture 13: Wednesday March 6, 13:15-15:00 Q22
  • Experiment design (Chapter 14 + lecture notes)
  • Biased estimation
  • Errors-in-variables estimation
  • Identification of dynamical networks
  • Concluding remarks

Extended Lectures (for FEL3202)

• TBD

Projects

A project should cover all (or most)  major steps of the system identification problem, refer to Figure 1.11 in Ljung's book. Preferrably real data should be used. Projects relating to the participants own research projects is encouraged.

A project group should consist of two participants of the course. 

A short report and a 5 minute presentation of the project is required to pass. Participants of FEL3202 are expected to make a more ambitious project. 

Schedule:

• Proposals due February 13. Send to hjalmars@kth.se with "FEL3201: Project proposal" in subject line.
• Report due March 31
• Presentations April, exact date TBD.

Exam

There will be a 48 hours take home exam. The exam period is March 13 - April 13. You will have to send me an email informing me before March 10 when you want to take the exam. The starting time is 09:00 for the chosen day.  You will need to send me a reminder the day of the exam before 08:30. The exam will then be sent to you by email. The solutions should be handed in by email. They can be scanned hand-written or type-set.