Scribd
Upload
57 views1 page

MICexam PDF

This document contains instructions for two problems involving system identification and Kalman filtering. Problem 1 involves identifying ARX, ARMAX, and OE models of different orders from input-output data to model a dynamic system. It asks to compare model performance on a separate test data set and analyze parameter uncertainty. Problem 2 involves designing a steady-state Kalman filter and one-step predictor for a linear time-invariant system with process and measurement noise. It asks to compare state estimates from the filter and predictor using root mean square error.

Uploaded by

Cesar Andres Sierra Pardo
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
57 views1 page

MICexam PDF

This document contains instructions for two problems involving system identification and Kalman filtering. Problem 1 involves identifying ARX, ARMAX, and OE models of different orders from input-output data to model a dynamic system. It asks to compare model performance on a separate test data set and analyze parameter uncertainty. Problem 2 involves designing a steady-state Kalman filter and one-step predictor for a linear time-invariant system with process and measurement noise. It asks to compare state estimates from the filter and predictor using root mean square error.

Uploaded by

Cesar Andres Sierra Pardo
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/ 1

01RKYQW - Estimation, filtering, and system identification

01PDDOV/QW - Identification and Control methodologies


Sample examination paper (exam duration: 3 hours)
SURNAME: NAME:

Remark: making use of MATLAB R2014A only, write two stand-alone m-files (named es1.m and
es2.m) to solve the problems, respectively. Write by hand a clear report that includes the reasoning
behind the computations, the main numerical results and their possible critical analysis.
Problem #1: the input-output measurements of a SISO dynamic system S1 to be modeled have
been collected in the MATLAB data1.mat file.
1) Identify ARX, ARMAX and OE models of different orders and delays, using only a part of the
experimental data and looking for models that guarantee satisfactory characteristics of whiteness of
the residuals associated to this first dataset.
2) Compare the identified models on a different set of data not useds for identification. To assess
1 N
the model quality, minimize the Root Mean Square Error RMSE = [y(t) − ŷ(t)]2 ,
P
N − N0 t=N0 +1
where y(t) = measured output, ŷ(t) = simulated (or predicted) output and N0 is a suitable time
instant after which the transient is past. Choose the best trade-off between RMSE and model
complexity: n = na + nb (for ARX models), n = na + nb + nc (for ARMAX models) and n = nf + nb
(for OE models), if the model is used for prediction purposes; n = na (for ARX and ARMAX models)
and n = nf (for OE models), if the model is used for simulation purposes.
3) Using all the experimental data, estimate the parameters of an ARX(3, 3, 1) model by means of
the standard Least-Squares algorithm and compare:
- the Estimate Uncertainty Intervals EUI 2 and, if possible, the Parameter Uncertainty Intervals
P UI 2 , assuming that the output measurements are corrupted by an energy-bounded noise whose
2-norm is less than 4;
- the Estimate Uncertainty Intervals EUI ∞ , assuming that the output measurements are corrupted
by an amplitude-bounded noise whose ∞-norm is less than 0.1.

Problem #2: consider the following LTI dynamic system S2 :


x(t + 1)= Ax(t) + Bu(t) + v1 (t)
y(t)= Cx(t) + v2 (t)
where
   
2.7258 −1.2424 0.7577 0.026 h i
A=

2 0 0 ,

B =  0 ,
 
C= 1 −0.45 −0.05
0 0.5 0 0
h iT
v1 (t) is a white noise with zero mean value and variance V1 = Bv1 BvT1 , Bv1 = 0.05 0.026 0 0 ,
v2 (t) is a white noise with zero mean value and variance V2 = 0.0004 and u(t) is a suitable input
signal whose values have been saved in the MATLAB data2.mat file. The noises v1 (t) and v2 (t)
are uncorrelated.h Assume thati the initial state x(1) is a random vector with zero mean value and
T
variance P1 = E x(1)x(1) = I3 .
1) Design the steady-state Kalman filter F∞ and the dynamic Kalman 1-step predictor in predictor-
corrector form Kpc .
2) Compare the state estimates provided by F∞ and Kpc by means of graphical representations and
evaluate the Root Mean Square Errors:
v
N ′
u
u 1
[xk (t) − x̂k (t)]2 ,
X
RMSEk = u
k = 1, . . . , 3
N − N0′
t ′
t=N0′ +1

where x̂k (t) is the estimate of the state xk (t) and N0′ is a suitable time instant after which the filter
or predictor transient is past.

You might also like