[go: up one dir, main page]
Scribd
Upload
47 views10 pages

SE SE SE SE513 513 513 513:::: System Identification System Identification System Identification System Identification

This document summarizes a lecture on system identification and subspace identification methods. It introduces state-space models and how they can represent discrete-time systems using matrices like A, B, C, and D. It also discusses how to estimate these matrices using least squares methods and the subspace identification algorithm. This involves choosing a well-conditioned basis for the observability matrix and then applying linear regression to estimate the state-space model parameters and noise statistics.

Uploaded by

Yamani Hurr
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
47 views10 pages

SE SE SE SE513 513 513 513:::: System Identification System Identification System Identification System Identification

This document summarizes a lecture on system identification and subspace identification methods. It introduces state-space models and how they can represent discrete-time systems using matrices like A, B, C, and D. It also discusses how to estimate these matrices using least squares methods and the subspace identification algorithm. This involves choosing a well-conditioned basis for the observability matrix and then applying linear regression to estimate the state-space model parameters and noise statistics.

Uploaded by

Yamani Hurr
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 10

5/28/2011

SE513
SE513:
513: System Identification

Lecture 11:
Subspace Identification

1
Dr. Sami El Ferik, KFUPM, term 102

Objectives
 State Space models and one step ahead prediction.
 Kalman Gain and Innovation equation
 Nonuniqueness of the state space representation
 Estimating the State-Space using LSM
 Subspace identification method.

2
Dr. Sami El Ferik, KFUPM, term 102

1
5/28/2011

State-Space Models
 Physically based models lead usually to a differential equation
of the form

 Or in discrete

 Where in case T as a sampling period and LTI system

3
Dr. Sami El Ferik, KFUPM, term 102

State-Space to Transfer function


 Using p for the differentiation operator

 Assume H as the measurement matrix and there is no noise


measurement

4
Dr. Sami El Ferik, KFUPM, term 102

2
5/28/2011

 Standard discrete form

 Which will lead

5
Dr. Sami El Ferik, KFUPM, term 102

Noise Representation
 Assume measurement noise

 And state process noise acting on the state such that

6
Dr. Sami El Ferik, KFUPM, term 102

3
5/28/2011

General Form

 Based on minimization problem of he quadratic error


between the state and its estimate (Kalman Filter)

7
Dr. Sami El Ferik, KFUPM, term 102

Kalman Gain

8
Dr. Sami El Ferik, KFUPM, term 102

4
5/28/2011

Innovation error

9
Dr. Sami El Ferik, KFUPM, term 102

Nonuniqueness of State-Space models


 Assume

 Where the random noises are assumed indep. White noises


 assume

 Where

10
Dr. Sami El Ferik, KFUPM, term 102

5
5/28/2011

Estimating state space models using


least square methods
 We have

 There are infinity many (A,B,C,D) that can generate the


same input-output relationship.
 There is a nee to fix the coordinate basis of the state space
realization or fix the structure of the different matrices.
 Assume the input-output relationship can be written as

11
Dr. Sami El Ferik, KFUPM, term 102

LS Regression
 Assume u, x, and y are measured.

12
Dr. Sami El Ferik, KFUPM, term 102

6
5/28/2011

How to obtain X(t)?


 K-step-ahead prediction of the output y(t)

 Define

 and

13
Dr. Sami El Ferik, KFUPM, term 102

 As
The system has an nth order

minimal state description if and only if the rank of is


equal to n for all
 The state vector of any minimal realization in innovations
form can be chosen as a linear combinations of that
form a row basis for

14
Dr. Sami El Ferik, KFUPM, term 102

7
5/28/2011

Approximation of k-step ahead


prediction
 The k-step ahead predictor knowing all the past up to t-1

 Therefore

 Or

 Dealing with r predictors

15
Dr. Sami El Ferik, KFUPM, term 102

 where

16
Dr. Sami El Ferik, KFUPM, term 102

8
5/28/2011

Remarks

 If the state space model equation does not include a noise
term

 then we select s1=0 .

17
Dr. Sami El Ferik, KFUPM, term 102

Subspace method steps


 Chose

 Estimate the rank n of and


determine L so that x(t) to a well
conditioned basis for it.
 Estimate A, B, C, D and the noise
covariance matrices by applying LS
method to the linear regression

18
Dr. Sami El Ferik, KFUPM, term 102

9
5/28/2011

Next Lecture
Subspace Methods for estimating
State Space Models

19
Dr. Sami El Ferik, KFUPM, term 102

10

You might also like