1 probabilistic uncertainty bounding in output error models with unmodelled dynamics 2006 american...
TRANSCRIPT
1
Probabilistic Uncertainty Bounding in Output Error Models with Unmodelled Dynamics
2006 American Control Conference, 14-16 June 2006, Minneapolis, Minnesota
Delft Center for Systems and Control
Sippe Douma and Paul Van den Hof
2
Uncertainty bounding in PE identification
3. For a given single realization determine a set offor which, within a probability level of holds that
Delft Center for Systems and Control
1. Analyse statistical properties of estimator(mapping data to parameter) through its pdf
2. Under (asymptotic) assumptions and no bias
with an analytical expression for covariance matrix P
4. This set is identical to the set of that forms an probability set of the pdf
3
Delft Center for Systems and Control
probability region for
4
Delft Center for Systems and Control
• Approximations to be made:• Employing asymptotic Gaussian distribution
of pdf• Assumption (no bias)• Obtaining P through Taylor approximation
(OE/BJ)• Replacing covariance matrix P by estimateThe
message • Estimator statistics are not necessary for obtaining probabilistic parameter uncertainty regions;There are alternatives with attractive properties
(Douma, Van den Hof, CDC/ECC-05)
Here:• Applied to Output Error models• Validity for finite-time• Options for
5
Contents
• Classical OE result• New OE approach• Extension towards finite time (?)• Options for extending to • Summary
Delft Center for Systems and Control
6
Delft Center for Systems and Control
OE modelling (classical)
1-step ahead predictor:
Identification criterion:
Based on Taylor approximation of
Conditioned on asymp. norm. of
7
Delft Center for Systems and Control
OE modelling (alternative)
Taylor approximation of :
which when substituted above, leads to
= linear regression type of expression for parameter error
known
8
Model uncertainty bounding requires:
• (asymptotic) normality of with covariance Q• Replacement of
by an estimate
leading to ellipsoid determined by
Delft Center for Systems and Control
Alternative
Same as before but conditions are relaxed
9
Delft Center for Systems and Control
ConclusionClassical results with approximated by sample estimates, requires
to become (asymptotically) Gaussian rather than
Benefit: relaxation of conditions for normality
10
Delft Center for Systems and Control
One step further (towards finite-time)
With svd: it follows that
Lemma:
If unitary and random, and e Gaussian with cov(e)=2 I,and and e independent, then is Gaussian with Cov = 2 I.
This would suggest that is Gaussian for any value of N.
11
Delft Center for Systems and Control
Corresponding uncertainty
Only problem: condition of independent and isnot satisfied, since is a function of
Starting from:
It follows that (for finite N):
with
Only needs to be replaced by an estimated expression
However: this does not appear devastating in practice!
12
Delft Center for Systems and Control
Simulation example:
First order OE system:
identified with 1st order OE model.
Compare empirical distributions of
and
for different values of N, on the basis of 5000 Monte Carlo simulations
13
Delft Center for Systems and Control
Component related to numerator parameter:
-50 0 50
(T
)-1
T e
-2 0 2
N = 2
VTe
-2 0 2
-2 0 2
N = 5
-0.5 0 0.5
-2 0 2
N = 25
-0.2 0 0.2
-2 0 2
N = 50
14
Delft Center for Systems and Control
Component related to denominator parameter:
15
Delft Center for Systems and Control
ConclusionNew analysis arrives at correct Gaussian distribution for uncertainty quantification, for a remarkably small number of N
16
Delft Center for Systems and Control
Options for extending to In analysis, replace by
becomes:
output noise unmodelled dynamics
Results to be quantified in frequency domain by usinga linearized mapping
17
Delft Center for Systems and Control
Summary
• There are alternatives for parameter uncertaintybounding, without constructing pdf of estimator
• Applicable to ARX, OE and also BJ models (see also Douma & VdHof, CDC/ECC-2005, SYSID2006)
• And with good options to be extended to the situation of approximate models
• Leading to simpler and less approximative expressions, remarkably robust w.r.t. finite time properties