Scribd
Upload
59 views6 pages

FRT 041 System Identification

This document provides instructions for a laboratory exercise on system identification using MATLAB's System Identification Toolbox. The objectives are to: 1. Estimate impulse responses and transfer functions of an unknown process using correlation analysis and spectrum analysis. 2. Estimate parameters of Box-Jenkins models and validate model fits using residuals and cross-validation. 3. Design a controller for the identified process and simulate closed-loop performance.

Uploaded by

عبدالسلام المبروك ابوزناد
Copyright
© © All Rights Reserved
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
59 views6 pages

FRT 041 System Identification

This document provides instructions for a laboratory exercise on system identification using MATLAB's System Identification Toolbox. The objectives are to: 1. Estimate impulse responses and transfer functions of an unknown process using correlation analysis and spectrum analysis. 2. Estimate parameters of Box-Jenkins models and validate model fits using residuals and cross-validation. 3. Design a controller for the identified process and simulate closed-loop performance.

Uploaded by

عبدالسلام المبروك ابوزناد
Copyright
© © All Rights Reserved
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/ 6

FRT 041 System Identification

Laboratory Exercise 2

Lennart Andersson
Ulf Jönsson
Revised: Johan Bengtsson
Brad Schofield

Department of Automatic Control


Lund Institute of Technology
Lund University
October 1995

Revised: January 2006


1. Introduction
In this laboratory exercise you will learn how to use Matlab’s System Iden-
tification Toolbox. The toolbox provides a way to obtain discrete time models
based on discrete time data.

1.1 Preparation
Read "Manual for Practical System Identification" in the appendix. This
manual can also be useful during the projects. Section 6–10 are most impor-
tant for the exercises in this lab. We assume that you have some experience
from using Matlab. Work through the exercise below.

Exercise Write a sequence of Matlab commands, which solves the follow-


ing tasks. Assume that you have the data sequence zi=iddata(yi,ui,h)
for identification and another data sequence zv=iddata(yv,uv,h) for vali-
dation.
1. Estimate the impulse response by use of correlation analysis and plot
the step response.
2. Estimate the transfer function by use of spectrum analysis. Plot the
Bode diagram.
3. Estimate the parameters of a Box Jenkins model with model order
n B = 2, n C = 2, n D = 2, n F = 3 and n k = 1.
4. Compute and plot the residuals for the Box-Jenkins model obtained
in step 3.
5. Plot the Bode diagram and the noise spectrum for the Box Jenkins
model obtained in step 3.
6. Do a crossvalidation for the Box-Jenkins model obtained in step 3.

1
2. Process specification
The process in this laboratory exercise is a simulation model of a real
process with artificial measurement noise and a load disturbance. For the
real process the input signal and output signal must be between ±10V . To
get realistic results you should make sure that this is satisfied during your
simulations !

3. Identification
The purpose of this section is that you should go through the steps of system
identification for an unknown process. The result will be a mathematical
model of the process.

Login at the computer and start Matlab matlab-R12.1 -nodesktop.

Download the file simu.m from the course webpage, and put it in your
working directory. This file will be used to simulate the unkown process.
Use the Matlab command help frequently during the whole excercise !
Most of the time should be spent on obtaining parametric models and
validation. Suggested timeplan:
• Experiments 60min
• Spectrum analysis 45min
• Parametric models and Validation 105min

3.1 Exercise 1
n

u y
Process

Figure 1 Black box description of the process

Figure 1 shows the available process as a black box. The influence from
disturbances is illustrated by the signal n. The process can be simulated
using a special matlab command simu.

[y,u,t]=simu(u,h);

u process input
h sampling period
y process output
t time vector

2
Simulate the process with zero input. A zero input vector can be created
using

u=zeros(N,1);

N number of data points

Use the matlab command plot to display the simulation result. Explain
the result.

3.2 Exercise 2
Do some simple experiments such as step responses to gain some basic
information about the process.
A step input vector can be created using

u=[zeros(N1,1);A*ones(N2,1)];

N1 number of samples with zero input


N2 number of samples with input A
A amplitude

Check that the input signal and output signal are between ±10 ! Determine
dominant time constants and the stationary gain. Does the process have
continuous time minimum phase character?

Is the process linear and time invariant?

3.3 Exercise 3
The choice of input signal is crucial in system identification. Here we will
use a Psuedo-Random Binary Sequence (PRBS) input signal. A PRBS sig-
nal can be obtained as

prbs=A*kron(rand(N,1)>0.5,ones(n,1));

A Amplitude
N sequence length
n samples during each state
prbs column vector with n*N elements

3
where you have to choose A,N and n.
Generate identification data by simulating the system with a PRBS
input.
Examine the data and prepare them for identification. How should the
data be divided? It may be useful to use iddata objects to store your iden-
tification data.

3.4 Exercise 4
Use spectral analysis–e.g., spa or spectrum on the gathered data. What
is the effect of experiment duration on the spectral estimates? Investigate
the use of ‘windowing’ to improve the results. If desired, the Identification
toolbox GUI may be used in this exercise. It is started by typing ident in
Matlab.

3.5 Exercise 5
Estimate some parametric models. This step contains both model structure
selection, model order selection and validation. This exercise is the most
difficult and time consuming! Therefore, spend most of the time here. It
may be useful to use the Identification toolbox GUI in this exercise as well.

4. Controller design
Based on the parametric model obtained by identification you should now
design a controller.

4
n
r u y
T 1/ R process

−S

Figure 2 Closed-loop system

4.1 Exercise 6
Design a controller and simulate the closed-loop system. Write help ppbox
and in particular help rstd to see how the design can be performed (you
may need to type oldboxes to access these functions). The design should
satisfy the signal specification stated earlier, i.e., the input and output
should be between ±10V . The total structure can be seen in Figure 2. The
closed loop can be simulated using the command

[y,uc,t]=simu(yr,h,R,S,T);

yr reference value
h sampling period
R,S,T controller polynomials
y process output
uc process input
t time vector

You might also like