eece 574 - adaptive control - recursive identification in closed-loop and adaptive...

43
EECE 574 - Adaptive Control Recursive Identification in Closed-Loop and Adaptive Control Guy Dumont Department of Electrical and Computer Engineering University of British Columbia January 2010 Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 1 / 43

Upload: others

Post on 23-Jun-2020

22 views

Category:

Documents


5 download

TRANSCRIPT

Page 1: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

EECE 574 - Adaptive ControlRecursive Identification in Closed-Loop and Adaptive Control

Guy Dumont

Department of Electrical and Computer EngineeringUniversity of British Columbia

January 2010

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 1 / 43

Page 2: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Tracking Time-Varying parameters

Tracking Time-Varying Parameters

All previous methods use the least-squares criterion

V(t) =1t

t

∑i=1

[y(i)− xT(i)θ ]2

and thus identify the average behaviour of the process.

For standard RLS, the estimation gain eventually converges to zero andadaptation stops.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 2 / 43

Page 3: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Tracking Time-Varying parameters Forgetting Factor

Forgetting Factor

When the parameters are time varying, it is desirable to base the identificationon the most recent data rather than on the old one, not representative of theprocess anymore. This can be achieved by exponential discounting of olddata, using the criterion

V(t) =1t

t

∑i=1

λt−i[y(i)− xT(i)θ ]2

where 0 < λ ≤ is called the forgetting factor.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 3 / 43

Page 4: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Tracking Time-Varying parameters Forgetting Factor

Forgetting Factor

The new criterion can also be written

V(t) = λV(t−1)+ [y(t)− xT(t)θ ]2

Then, it can be shown (Goodwin and Payne, 1977) that the RLS schemebecomes

RLS with Forgetting

θ(t +1) = θ(t)+K(t +1)[y(t +1)− xT(t +1)θ(t)]K(t +1) = P(t)x(t +1)/[λ + xT(t +1)P(t)x(t +1)]

P(t +1) ={

P(t)− P(t)x(t +1)xT(t +1)P(t)[λ + xT(t +1)P(t)x(t +1)]

}1λ

In choosing λ , one has to compromise between fast tracking and long termquality of the estimates. The use of the forgetting may give rise to problems.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 4 / 43

Page 5: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Tracking Time-Varying parameters Forgetting Factor

Forgetting Factor

The smaller λ is, the faster the algorithm can track, but the more theestimates will vary, even the true parameters are time-invariant.

A small λ may also cause blowup of the covariance matrix P, since inthe absence of excitation, covariance matrix update equation essentiallybecomes

P(t +1) =1λ

P(t)

in which case P grows exponentially, leading to wild fluctuations in theparameter estimates.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 5 / 43

Page 6: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Tracking Time-Varying parameters Forgetting Factor

Variable Forgetting Factor

One way around this is to vary the forgetting factor according to theprediction error ε as in

λ (t) = 1− kε2(t)

Then, in case of low excitation ε will be small and λ will be close to 1.In case of large prediction errors, λ will decrease.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 6 / 43

Page 7: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Tracking Time-Varying parameters EFRA

Exponential Forgetting and Resetting Algorithm

The following scheme1 is recommended:

EFRA Algorithm

ε(t +1) = y(t +1)− xT(t +1)θ(t)

θ(t +1) = θT(t)+

αP(t)x(t +1)λ + xT(t +1)P(t)x(k +1)

ε(t)

P(t +1) =1λ

[P(t)− P(t)x(t +1)xT(t +1)P(t)

λ + x(t +1)TP(t)x(t +1)

]+β I− γP(t)2

where I is the identity matrix, and α , β and γ are constants.

1M.E. Salgado, G.C. Goodwin, and R.H. Middleton, “Exponential Forgetting andResetting”, International Journal of Control, vol. 47, no. 2, pp. 477–485, 1988.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 7 / 43

Page 8: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Tracking Time-Varying parameters EFRA

Exponential Forgetting and Resetting Algorithm

With the EFRA, the covariance matrix is bounded on both sides:

σminI ≤ P(t)≤ σmaxI ∀t

where

σmin ≈β

α−ησmax ≈

η

γ+

β

η

with

η =1−λ

λ

With α = 0.5, β = γ = 0.005 and λ = 0.95, σmin = 0.01 and σmax = 10.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 8 / 43

Page 9: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop The Identifiability Problem

Identification in Closed Loop

The Identifiability ProblemLet the system be described by

y(t)+a · y(t−1) = b ·u(t−1)+ e(t)

withu(t) = g · y(t)

Let a and b be closed-loop estimates of a and b.

Then,the closed-loop system can be written as:

y(t)+(a− b ·g)y(t−1) = e(t)

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 9 / 43

Page 10: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop The Identifiability Problem

Identification in Closed Loop

Hence any estimates a and b that satisfy

a− b ·g = a−b ·g

will give the same value for the identification criterion.

All estimates such that

a = a+ k ·gb = b+ k

will give a good description of the process.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 10 / 43

Page 11: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop The Identifiability Problem

Identification in Closed Loop

If the identification is performed using two feedback gains g1 and g2 or ifthe parameter a is fixed, then the system becomes identifiable, becausewe have as many equations as unknowns.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 11 / 43

Page 12: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop Definitions

Definitions

Let the discrete plant be described by

y(t) = GD(q−1)u(t)+GN(q−1)e(t)

where GD and GN are linear rational transfer functions in the backwardshift operatorq−1 (i.e. q−1y(t) = y(t−1) that can be parameterized by avector θ , {e(t)}= N(0,σ). Let S denote this true system.

Let us assume that the identification is performed with the feedbackcontroller R such that u(t) = R · y(t).The problem is then to find θ , an estimate of θ such that the modelM(θ)given by

y(t) = GD(q−1)u(t)+ GN(q−1)e(t)

where GD(q−1) and GN(q−1) are parameterized by θ , describes thesystem S. Assume that an identification method denoted I is used toobtain the estimate θ .

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 12 / 43

Page 13: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop Definitions

Definitions

A loose and intuitive definition of identifiability is that M(θ) describes Sas the number of measurements N tends to infinity. Let us define

DT(S,M) = {θ |GD(q−1)≡ GD(q−1) and GN(q−1) = GN(q−1)∀q}

This is the set of desired estimates which corresponds to models M(θ)with the same plant and noise transfer functions as the actual systemS(θ).Note that this set does not depend on the regulator R nor on theidentification method I. Note also that the orders of the model transferfunctions can be greater than those of the system, in which case thereexist some pole-zero cancellations.

The actual estimates θ depend on the number of measurements, thesystem and its model, the regulator and the identification method and canbe written as θ(N;S,M, I,R).

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 13 / 43

Page 14: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop Definitions

Definitions

Definition 1: The system S is said to be system identifiable under M, I,R, i.e.SI(M, I,R)if

θ(N;S,M, I,R)−→ DT(S,M)w.p.1

as N→ ∞.

Definition 2: The system S is said to be strongly system identifiable under I and R, i.e.SSI(I,R) if it is SI(M, I,R) for all M s.t. DT(S,M) 6= φ where φ denotes the empty set.

Definition 3: The system S is said to be parameter identifiable under M, I and R,i.e.PI(M, I,R) if it is SI(M, I,R) and DT(S,M) consists of only one element.

According to those definitions, the system of Example 1 is neither SI nor PI for the class of

models (2). Moreover, since for the model (2) DT(S,M) 6= φ , the system is not SSI either.

However, when two regulators are used, the system becomes PI. It is also important to note

that when a system is SSI(I,R), the fact that the identification is performed under closed-loop

is irrelevant and the identification can then be performed as if the system were operating in

open-loop.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 14 / 43

Page 15: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop Definitions

Example

Consider the system

A(q−1)y(t) = B(q−1)q−ku(t)+C(q−1)e(t)

with the feedbackF(q−1)u(t) = G(q−1)y(t)

The closed-loop system can then be described as

(AF−q−kBG)y(t) = CFe(t)

It is then obvious that any estimates A and B such that

(AF−q−kBG) = (AF−q−kBG)

will describe the above system.Hence, if L(q−1) is an arbitrary polynomial, any Aand B s.t.{

A = A+LGB = B+qkLF

(1)

are possible estimates.This means that the true order of the system cannot be established from this type ofclosed-loop experiment and thus, has to be known à-priori.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 15 / 43

Page 16: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop Identifiability Conditions

Identifiability Conditions

Let the system S be described by the following ARMAX process:

A(q−1)y(t) = q−kB(q−1)u(t)+C(q−1)e(t)

where

A(q−1) = 1+a1q−1 + . . .+anaq−na

B(q−1) = b1q−1 + . . .+bnbq−nb

C(q−1) = 1+ c1q−1 + . . .+ cncq−nc

k ≥ 0 na ≥ 0 nb ≥ 1 nc ≥ 0

A, B, C are assumed to co-prime, i.e. have no common factors.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 16 / 43

Page 17: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop Identifiability Conditions

Identifiability Conditions

Let the regulator R be given by:

F(q−1)u(t) = G(q−1)y(t)

whereF(q−1) = 1+ f1g−1 + . . .+ fnf q

−nf

G(q−1) = g0 +g1q−1 + . . .+gngq−ng

F and G are co-prime and G may contain a delay, i.e. some of its leadingcoefficients may be zero.

Let the class of models M(θ) be

A(q−1)y(t) = q−kB(q−1)u(t)+ C(q−1)e(t)

with na ≥ 0, nc ≥ 0, k ≥ 0, nb ≥ 1

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 17 / 43

Page 18: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop Identifiability Conditions

Identifiability Conditions

Introduce

n∗ = min [(na−na),(nb + k−nb− k),(nc−nc)]

The system with its feedback can be written as

y =CF

AF−q−kBG

Let us define np be the number of common factors between C and(AF−q−kBG) and let the order of the polynomial AF−q−kBG bemax[(na +nf ),(k +nb +ng)]− r with r ≥ 0. Note that r > 0 meansna +nf = k +nb +ng and that some of highest order coefficients arezeros.

Assume that a direct identification, using the sequences {u(t)} and{y(t)}, is employed to estimates θ such that M(θ) describes S(θ).

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 18 / 43

Page 19: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop Identifiability Conditions

Identifiability Conditions

Theorem (Söderström, 1973)

A necessary and sufficient condition for the set of desired estimates DT(S,M)to be non-empty is

n∗ ≥ 0 and k∗ ≤ k

From the definition of n∗, it is seen that the condition simply means thatthe true system is contained in the model set, i.e. that the model order issufficient and that the dead-time is not overestimated.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 19 / 43

Page 20: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop Identifiability Conditions

Identifiability Conditions

A related well known sufficient identifiability condition is that the orderof the feedback law be greater than or equal to the order of the forwardpath, see Goodwin and Payne (1977).

The presence of dead-time k > 0 helps identifiability. This can beexplained by the fact that the feedback signal then contains a componentmore or less independent of the current output y(t).The previous condition unfortunately involve the true order of thesystem. If the true order isnot known à-priori, then these conditionscannot be checked à-posteriori, since the order cannot be determined inclosed-loop identification.

These conditions are thus of little use in a practical application. Thefollowing theorem provides a way to guarantee SSI without knowledgeof the systems order.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 20 / 43

Page 21: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop Identifiability Conditions

Identifiability in Closed Loop

Theorem (Söderström et al. (1975))Consider the system S:

y(t) = GD(q−1)u(t)+GN(q−1)e(t)

wheredim y = ny,dim u = nu,dim e = ny, e = N(O,Λ)

andGN(q−1) has all zeros outside unit circle

. Let the control signal be given by

u(t) = Fi(q−1)y(t)+ ki(q−1)v(t)

{v} is an external signal of dim v = nv and Fi and kiare rational functions and i = 1, . . .r.Assume, without loss of generality, that there is a dead-time either in the process or in theregulator, i.e. GD(0)Fi(0) = 0. Then, the system S is identifiable if and only if

n0{rankny{

nv︷︸︸︷k1(z) . . .kr(z)

ny︷︸︸︷Fi(z) . . .Fr(z)

0 . . .0 I . . . I

= ny +nu ∀z

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 21 / 43

Page 22: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop Identifiability Conditions

Practical Conditions for Closed-Loop Identifiability

A necessary condition for identifiability is

r(nv +ny)≥ ny +nu

orr ≥ (nu +ny)/(nv +ny)

If nv = nu, then closed-loop identifiability always satisfied

Without external signal (nv = 0),

r ≥ 1+nu/ny

which when ny = nu becomes r ≥ 2

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 22 / 43

Page 23: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop Identifiability Conditions

Consequences for Adaptive Control

Under time-varying feedback, identifiability should be garanteed.

Because in adaptive control, the controller parameters are constantlyupdated, identifiability should be garanteed (at least during the transientphase).

Nonlinearities in the controller should also help identifiability.

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 23 / 43

Page 24: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Identification in Closed Loop Relevant Loops

Identification for Control

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 24 / 43

Page 25: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Effect of Under-modelling PEM and Undermodelling

Prediction Error Identification

(Gevers, 2005)

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 25 / 43

Page 26: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Effect of Under-modelling PEM and Undermodelling

Properties of the Identified Model

(Gevers, 2005)

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 26 / 43

Page 27: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Effect of Under-modelling Bias and Variance Errors

Bias and Variance Errors

Ljung (1985) was the first to formalize the situation when the true plantdoes not belong to the model setModelling error consists of two distinct components:

The bias error arises when the model structure is unable to represent thetrue systemThe variance error is caused by the noise and the finiteness of the data set

The variance error goes asymptotically to zero as the number of datagoes to zero.

If the model is not exact, then its quality should reflect its intended use

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 27 / 43

Page 28: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Effect of Under-modelling Bias and Variance Errors

Bias and Variance Errors

(Gevers, 2005)

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 28 / 43

Page 29: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Effect of Under-modelling Bias and Variance Errors

Bias and Variance Errors

Introduce θ ∗ as the convergence point of the prediction error estimate:

θ∗ = lim

N→∞θN = arg min

θ∈Dθ

V(θ)

As defined by Ljung (1985), the bias and variance of the transferfunction estimates are:

Definition: Bias and Variance Errors

G0(ejω)− G(ejω , θN) = G0(ejω)− G(ejω ,θ ∗)︸ ︷︷ ︸Bias

+ G(ejω ,θ ∗)− G(ejω , θN)︸ ︷︷ ︸Variance

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 29 / 43

Page 30: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Effect of Under-modelling Bias and Variance Errors

Quantifying the Variance

DefineVar(GN(ejω)) = E[|G(ejω , θN)−E[G(ejω , θN)]|2]

With Φu and Φv the power spectral densities of u and H0e0, Ljung (1985)showed that

limn→∞

limN→∞

= Var(GN(ejω)) =nN

Φv(ω)Φu(ω)

where n is the model order

Recently, exact expressions for finite-order models have been derived

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 30 / 43

Page 31: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Input Design PRBS

Pseudo Random Binary Sequences (PRBS)

(Landau, Lozano, M’Saad, Adaptive Control, Springer-Verlag,1998, pp. 179-182)

A PRBS is a sequence of rectangular pulses of constant magnitude,modulated in width that approximates a white noise, and thus is rich infrequencies.

A PRBS of length 2N−1 is generated by means of a shift register of Nstages:

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 31 / 43

Page 32: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Input Design PRBS

Pseudo Random Binary Sequences (PRBS)

Maximum duration of a PRBS rectangular pulse is dmax = NTs

For estimation of static gain, the duration of at least one pulse must begreater than the rise time of the process, dmax = NTs > tR

Typically, duration L of test is taken s.t. L = (2N−1)Ts

The condition NTs > tR might result in too long an experiment

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 32 / 43

Page 33: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Input Design PRBS

Pseudo Random Binary Sequences (PRBS)

By choosing the PRBS clock frequency s.t.

fPRBS =fsp

p = 1,2,3, · · ·

the condition becomes dmax = pNTs > tRIf p is the PRBS frequency divider, then

dmax = pNTs L′ = pL p = 1,2,3 · · ·

If N is increased by p−1, then

dmax = (N +p−1)Ts L′ = 2(p−1)L p = 1,2,3 · · ·

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 33 / 43

Page 34: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Input Design PRBS

Pseudo Random Binary Sequences (PRBS)

Dividing the PRBS clock frequency reduces high frequencies while augmentinglow frequencies

Recommended to choose p < 4

For example for N = 8 and p = 1,2,3

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 34 / 43

Page 35: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Input Design Optimal Input Design

Optimal Input Design

Optimal design requires knowledge of the true system!This advocates iterative method...

First use a PRBSThen, using estimated model, calculate an input optimal for that model

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 35 / 43

Page 36: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Input Design Persistent Excitation

Persistent Excitation

A signal u is called persistently exciting of order n if the matrix Cn

defined below is positive definite

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 36 / 43

Page 37: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Input Design Persistent Excitation

Persistent Excitation

A step is PE of order 1(q−1)u(t) = 0

A sinusoid is PE of order 2

q2−2qcosωh+1)u(t) = 0

To identify a transfer function of order n one needs n sinusoids

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 37 / 43

Page 38: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Examples

Example 1

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 38 / 43

Page 39: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Examples

Example 1: Excitation

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 39 / 43

Page 40: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Examples

Example 1: Feedback

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 40 / 43

Page 41: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Examples

Example 1: Forgetting Factor

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 41 / 43

Page 42: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Examples

Example 2: Model Structure

Consider the process

y(t)−0.8y(t−1) = 0.5u(t−1)+ e(t)−0.5e(t−1)

First use the estimation model with RLS

y(t)+ay(t−1) = bu(t−1)+ e(t)

Then use the estimation model with RELS

y(t)+ay(t−1) = bu(t−1)+ e(t)+ ce(t−1)

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 42 / 43

Page 43: EECE 574 - Adaptive Control - Recursive Identification in Closed-Loop and Adaptive Controlcourses.ece.ubc.ca/574/EECE574-5_RecursiveID_closed-loop.pdf · 2010-01-19 · EECE 574 -

Examples

Example 2: Model Structure

Guy Dumont (UBC EECE) EECE 574 - Adaptive Control January 2010 43 / 43