[ieee 2008 fifth international conference on fuzzy systems and knowledge discovery (fskd) - jinan...
TRANSCRIPT
Stability of Linear Dynamical Systems with Fuzzy Parameters
Le Hung Lan Department of Cybernetics, Faculty of Electrical –Electronic Engineering
University of Transport and Communications Hanoi, Vietnam
E-mail: [email protected]
Abstract
In the paper the problem of stability analysis of
uncertain linear systems with parameters described by fuzzy functions is studied. Each coefficient is considered as a variable interval parametrized by the degree of confidence in validity of the corresponding model. Then the system stability can be solved by using some extended robust stability criteriums for nonsymmetric intervals. 1. Introduction
It is well-known that it is not possible to obtain an accurate model of system, some uncertainty is always present. A system has parametric uncertainty if there exists a mathematical model of it, but the values of some of its parameters are not exactly known. Many times, an estimation of the uncertain parameter values are given by an expert. In this case, the uncertain parameter can be represented by a fuzzy number
iq~ ,
with membership function ]1,0[)( ∈= iqμα . The membership value α can be interpreted as the confidence degree in that the value of the parameter equals its nominal value. Thus, a value 1=α indicates precise knowledge ( )ker(
~
ii qq ∈ ) whereas 01=α represents maximum
uncertainty ( ∈iq supp(iq
~ ). By using α -cuts concept
the coefficient iq~
can be represented as varying interval.
[ ])(),()( iiii
def
i qqq ααα +−= where iα is the confidence degree in that coefficient
(.)−iq and (.)+
iq are strictly increasing and decreasing functions, respectively, so that
++−− == iiii qqqq )0(,)0( . (1)
Let us consider an uncertain linear dynamical continuous system with single input and single output described by the differential equation.
)(...)()(...)()( 0~
)(~
0~
)1(1
~)( tubtubtyatyaty m
mn
nn ++=+++ −
− (2)
where ji ba~~
, are fuzzy numbers. This linear system with fuzzy parametric
uncertainty can be represented by the transfer function
0~
11
~0
~1
1~~
~
~~~~
...
...
),(
),(),,(asas
bsbsb
bsU
asYabsGn
nn
mm
mm
+++
++==−
−
−− (3)
Given a confidence degrees bj
ai αα , for the
parameters ji ba~~
, , the confidence degree in the model can be defined as the confidences vector
),...,,...,,( 110bm
ban
a ααααα −= . System (2) can then be represented by the family of transfer functions
α
αα
α⎥⎦⎤
⎢⎣⎡++⎥⎦
⎤⎢⎣⎡+
⎥⎦⎤
⎢⎣⎡++⎥⎦
⎤⎢⎣⎡
=⎥⎦⎤
⎢⎣⎡
−− 0
~1
1~
0~~
~~
...
...),(
asas
bsbqsG
nn
n
mm
(4)
where [ ] [ ])(),(,)(),(~~
bjj
bjjj
aii
aiii bbbaaa αααα
αα
+−+− =⎥⎦⎤
⎢⎣⎡=⎥⎦
⎤⎢⎣⎡ ,
i=1, …, n, j=0, … , m are interval components, corresponding to the α -cuts of the uncertain
parameters ji ba~~
, . For a given confidence degree, the transfer function (3) corresponds to a model with interval parameters. Thus, a fuzzy parametric uncertain system is a generalization of an interval system [1,2,5,6] and the robust stability theory can be usefull for study of this system class. Firstly, we extend some robust stability criteriums and next show that how can apply them for fuzzy dynamical system stability analysis.
Fifth International Conference on Fuzzy Systems and Knowledge Discovery
978-0-7695-3305-6/08 $25.00 © 2008 IEEE
DOI 10.1109/FSKD.2008.164
115
Fifth International Conference on Fuzzy Systems and Knowledge Discovery
978-0-7695-3305-6/08 $25.00 © 2008 IEEE
DOI 10.1109/FSKD.2008.164
115
2. Extended robust stability criteriums of interval systems 2.1. Interval characteristic polynomial
Tsypkin and Polyak [5] introduced a simple graphical robust stability criterium for a family of interval polynomials, where there is symmetric parametric perturbation
01
1 ...)( asasasp nn
nn +++= −
− (5)
iii aaa Δ≤− γ0 . (6)
The extended Tsypkin-Polyak involves the case of nonsymmetric parametric perturbation
+− Δ≤−≤Δ− iiii aaaa γγ 0 (7) Denote the nominal polynomial
00
101
00 ...)( asasasp nn
nn +++= −
− (8) and
},...,0,),({)( 0 niaaaajp iiii =Δ≤−≤Δ−=Π +− γγωωγ (9) a set on complex plan, ∞<≤ ω0 .
The set )(ωγΠ here is a rectangle:
...,)(
...,)(
...,)(
...,)(
),())()(Im()(
),())()(Re()(
67
45
231
67
45
231
66
44
220
66
44
220
0
0
+Δ+Δ+Δ+Δ=
+Δ+Δ+Δ+Δ=
+Δ+Δ+Δ+Δ=
+Δ+Δ+Δ+Δ=
≤−≤−
≤−≤−
−+−+−
+−+−−
−+−+−
+−+−−
+−
+−
ωωωωωωωωωωωωωωωω
ωγωωωγωγωωωγ
aaaaS
aaaaT
aaaaS
aaaaS
TjpjpT
SjpjpS
(10)
hence the condition )(0 ωγΠ∉ means that the inequalities
)()(Im)(),()(Re)( 00 ωγωωγωγωωγ +−+− ≤≤−≤≤− TjpTSjpS (11) can not be satisfied simultaneously for any ∞<≤ ω0 . If we introduce
[ ]{ }[ ]{ }
)(/)()(),(/)()(
,)()()()()(5.0)(
,)()()()()(5.0)(
...)(Im)(
...)(Re)(
00
0
0
405
203
01
010
404
202
00
00
ωωωωωωωωωωωωωωωωωω
ωωωωωωωωω
TVySUx
TTsignVTTT
SSsignUSSS
aaajpV
aaajpU
==
−++=
−++=
−+−==
−+−==
+−+−
+−+−
−
(12)
then the condition (13) can be written as { } γωω >)(,)(max yx (for 0>ω ). As )(ωx and
)(ωy have the same signs as )(),( 00 ωω VU , the plot )()()( ωωω jyxz += goes through the same quadrants
as the plot )(0 ωjp . If we apply Mikhailov’s test for
stability of a fixed polynomial )(0 sp with regard to
conditions 0>na (it give γ>Δ −nn aa /0 ) and for
0=ω , we arrive to the following Extended Tsypkin-Polyak criterion [6].
Theorem 1. The following conditions are necessary and sufficient for robust stability of the family (7), (9):
a) The plot )()()( ωωω jyxz += , ∞<≤ ω0 goes through n quadrants in counterclockwise direction.
b) It does not intersect γ -square (i.e. the set γγ ≤≤ yx , ).
c) Boundary conditions hold: γγγ >∞>∞> )(,)(,)0( yxx .
2.2. Closed-loop systems with interval plant
Let us consider the closed-loop control system with fixed controller
)()()(
sCsDsR = , (13)
and interval plant
nmsasaasbsbb
asAbsBbasP n
n
mm <
++++++== ,
......
),(),(),,(
10
10 (14)
,...,,1,0,,...,,1,0, 00 niabbmibbb iiiiii =Δ≤−=Δ≤− γγ (15)
where 0,0 ≥≥ ii βα are ranges of bounded coefficients perturbations, 0≥γ - margin of
perturbations, 00 , ii ab - nominal coefficients. In [7] the author proposed the simple graphical
criterium for system robust stability test and simultaneously computing stability margin maxγ .
Note that the coefficient interval perturbations given by (15) are symmetric, next we extend the criterium for the case of nonsymmetric interval perturbations. It means that
,...,,0,,...,,0, 00 niaaaamibbbb iiiiiiii =Δ≤−≤Δ=Δ≤−≤Δ +−+− γγγγ (16)
where −− ΔΔ ii ab , - left range of perturbations, ++ ΔΔ ii ab , - right range of perturbations.
Let the nominal system with plant
nn
mm
sasaasbsbb
sAsBs 00
100
001
00
0
00 ...
...)()()(P
++++++==
is stable. If denote
116116
...
...,
...,
...,
,)(
67
45
231
67
45
231
66
44
220
66
44
220
000
+Δ+Δ+Δ+Δ=Δ
+Δ+Δ+Δ+Δ=Δ
+Δ+Δ+Δ+Δ=Δ
+Δ+Δ+Δ+Δ=Δ
+=
−+−++
+−+−−
−+−++
+−+−−
ωωωωωωωωωωωω
ωω
aaaaa
aaaaa
aaaaa
aaaaa
vjujA
odd
odd
even
even
AA
(17)
then the set ),( ajA ω is rectangle in complex plan
with center )(0 ωjA , left width −Δ evenaγ , right width +Δ evenaγ , above height +Δ oddaγω , and under heght
−Δ oddaγω . Similarly we have the values
,,),(0−+ ΔΔ eveneven bbjB γγω −+ ΔΔ oddodd bb γωγω , for the
rectangle ),(B bjω .
Denote RRj vjuejR ωρω ϕ +==)(
( ) ( )RA
RA
A
RA
RA
A
vuuvvvvuuu 002002
0201;1 −−=+−=
−−
ρω
ρ,
( ) ( )RB
RB
B
RB
RB
B
uvvuvvvuuu 00002
00 ; +−=−−=−−
ω , (18)
,,,, 00000000
ABB
ABB
BAA
BAA uucvvdvvduuc
−−−−−=−=−=−=
( ) ( ) ( )[ ]
( )( ) ( )
( )( ) ( )
( ) ( ) ( )[ ],5,0
},)]([
)](){[(5,0
},)]([
)](){[(5,0
,5,0
22222
2
2
2
2
1
AAAAAA
RRoddoddoddodd
RRevenevenevenevenA
RRoddoddoddodd
RRevenevenevenevenA
Aeveneveneveneven
A
csignccccc
vvsignbbbb
uusignbbbbc
vvsignbbbb
uusignbbbbc
csignaaaac
−+−+
+−+−
+−+−−
+−+−
+−+−+
−+−+
−++=
Δ−Δ+Δ+Δ+
+Δ−Δ−Δ+Δ=
Δ−Δ−Δ+Δ+
+Δ−Δ+Δ+Δ=
Δ−Δ−Δ+Δ=
ω
ρ
ω
( ) ( ) ( )[ ]
( )( )( )
( ) },)]()[(
)](){[(5,0
,})]()[(
)](){[(5,0
,5,0
2
2
1
RRoddoddoddodd
RRevenevenevenevenA
RRoddoddoddodd
RRevenevenevenevenA
Aoddoddoddodd
A
uusignbbbb
vvsignbbbbd
uusignbbbb
vvsignbbbbd
dsignaaaad
+−+−
+−+−−
+−+−
+−+−+
−+−+
Δ−Δ−Δ+Δ+
+Δ−Δ−Δ+Δ=
Δ−Δ+Δ+Δ+
+Δ−Δ+Δ+Δ=
Δ−Δ−Δ+Δ=
ρ
( ) ( ) ( )[ ]( ) ( ) ( )[ ]
( )( ) ( )
( )( ) ( )
( ) ( ) ( )[ ],5,0
,/})]([
)](){[(5,0
,/})]([
)](){[(5,0
,5,0
,5,0
22222
22
2
22
2
1
22222
BBBBBB
RRoddoddoddodd
RRevenevenevenevenB
RRoddoddoddodd
RRevenevenevenevenB
Beveneveneveneven
B
AAAAAA
csignccccc
vvsignaaaa
uusignaaaac
vvsignaaaa
uusignaaaac
csignbbbbc
dsignddddd
−+−+
+−+−
+−+−−
+−+−
+−+−+
−+−+
−+−+
−−+=
Δ−Δ−Δ+Δ+
+Δ−Δ−Δ+Δ=
Δ−Δ+Δ+Δ+
+Δ−Δ+Δ+Δ=
Δ−Δ+Δ+Δ=
−++=
ρω
ρω
( ) ( ) ( )[ ],5,01B
oddoddoddoddB dsignbbbbd −+−+ Δ−Δ+Δ+Δ=
( )
( ) ,/})]()[(
)](){[(5,02
2
ρRRoddoddoddodd
RRevenevenevenevenB
uusignaaaa
vvsignaaaad+−+−
+−+−+
Δ−Δ+Δ+Δ+
+Δ−Δ−Δ+Δ=
( )( )
( ) ( ) ( )[ ]BBBBBB
RRoddoddoddodd
RRevenevenevenevenB
dsignddddd
uusignaaaa
vvsignaaaad
−+−+
+−+−
+−+−−
−−+=
Δ−Δ−Δ+Δ+
+Δ−Δ+Δ+Δ=
22222
2
2
5,0
,/})]()[(
)](){[(5,0
ρ
BB
B
LBB
B
LAA
A
LAA
A
L dddv
cccu
dddv
cccu
21
2
21
2
21
1
21
1 ,,,+
=+
=+
=+
=
Construct plot
( ) ∞<≤−−+++
+−−++=
ω
ω
0,)()()(
)()()()(212121
212121
LLLLLL
LLLLLL
vvsignvvvvj
uusignuuuuL (19)
Theorem 2. Necessary and sufficient conditions for robust stability of the interval system (13),(14),(16) are as follows:
(a) Nominal system is stable, (b) The plot )(ωL does not intersect the γ -square
(i.e the square with vertices ),( γγ ±± . Proof.
The closed-loop system characteristic polynomial corresponds to the set of polynomials with interval parameters
),()(),()(),,( asAsCbsBsDabsH += . (20) Following the Zero Exclusion Principle the entire
set is stable if and only if a) nominal polynomial is stable and b) ),,(0 abjH ω∉ . But ),,( abjH ω is the sum of two rectangles: ),()(1 bjBjD ωω=Π and
),()(2 ajAjC ωω=Π . So the condition ),,(0 abjH ω∉ means that the rectangles 12 , Π−Π do not intersect. The condition b) in theorem is implied from the simple note: two rectangles do not intersect iff all vertices of each rectangle do not belong to another one [7].
Remark. The main advantage of the two above robust stability criteriums is that by constructing plot
)(ωz or )(ωL it is not only possible to verify the system stability with given value 0≥γ , but also find maximal value maxγ , where maxγ - dimension of the bigest square touched them. The system stability remains preserved for all maxγγ < , in other word, maxγ is system stability margin. 3. Stability of systems with fuzzy coefficients
117117
Consider fuzzy system described by fuzzy
characteristic polynomial
,...)( 0~
11
~~~asasasp n
nn
n +++= −− (21)
or closed-loop system with fixed controller (13) and fuzzy plant
nmsasaa
sbsbb
asA
bsBbasPn
n
mm <
+++
+++== ,...
...
),(
),(),,( ~
1
~
0
~
~1
~~
~
~~~~
0 , (22)
where the coefficients are fuzzy numbers with triangular or trapezoidal membership functions. The α -cut represents in a compact way the family of characteristic polynomials
)(...)()()( 001
11 αααα asasasp nnn
nnn
def+++= −
−− , (23)
[ ] niaaa iiiiii ...,,0,)(),()( =∈ +− ααα , or transfer functions
)(...)()()(...)(
001
11
00ana
nnna
nn
bmbmm
asasabsbP
αααααα
+++++= −
−−
, (24)
[ ] niaaa aii
aii
aii ...,,0,)(),()( =∈ +− ααα ,
[ ] mibbb bii
bii
bii ...,,0,)(),()( =∈ +− ααα ,
parameterized by the confidence degree in the model. Note that their coefficient are nonsymmetrical
functional intervals with centers ioi aa
~ker)1( ∈ or
ioi bb
~ker)1( ∈ . So the desired theorems give suitable
tools for system stability analysis. Two main problems can be raised in system
stability analysis here: firstly, given different confidence levels iα (or b
iai αα , ), the answer of
system stability can be obtained by examining plot )(ωz or )(ωL with 1=γ ; furthmore, in the case of
common confidence level α and triangular membership functions, using those graphical criteriums we can find stability margin minα (the system stability remain preserved for all minαα > ), where
maxmin 1 γα −= . (25) It is not so simple for the case of trapezoidal
membership functions. Here for fuzzy number iq~
we must define
.)0(),0(
,1)1()0(
)0()1()0(
)0(
,)0()0()1()1(
)0()1()0()1(
00
00
0
iiiiii
ii
ii
ii
iii
iiii
iiiii
qqqqqq
qqqq
qqqqm
qqqqqqqqq
−=Δ−=Δ
≥−
−=−
−=
−+−−=
++−−
++
+
−−
−
+−−+
+−−+
(26)
Then we can apply the above two theorems to compute maxγ . Besides this case is different with the above triangular case: each coefficient has own stability margin
),1( maxmin γα −= ii m (27)
The maximal value of them can be considered as the overall system stability margin i
minmin max αα = . 4. Conclusion
We have considered above the problem of fuzzy system stability that can be solved effectively by the robust stability approach. The proposed methods are based on extended robust stability criteriums for the case of nonsymmetric interval and α -cuts concepts. They can be used for linear continuous dynamical system with coefficients being described by triangular or trapezoidal fuzzy functions. 5. References [1] Bondia J., Pico J. “Analysis of systems with variable parametric uncertainty using fuzzy functions”, Proc. European control conference ECC’99, Karlsruhe, Germany, 1999. [2] Bondia J., Pico J. Analysis of linear systems with fuzzy parametric uncertainty, Fuzzy Sets and Systems. 2003. V.135. p. 81-121. [3] Argoun M.B. “Frequency domain conditions for the stability of perturbed polynomials”, IEEE Trans. Automat. Control. 1987. AC-32. V.10. p.913-916. [4] Husek P., Pytelkova R., “Analysis of systems with parametric uncertainty described by fuzzy functions”, Proc. 10th Mediterranean Conference of Control and Automation – MED2002. Lisbon. Portugal, July 9-12, 2002. [5] Tsypkin Ya.Z., Polyak B.T, Frequency domain criteria for lp-robust stability of continuous linear systems, IEEE Trans. Aut. Control, 1991. N.1. 1464-1469. [6] Le Hung Lan, “Analiz robastnoj ustoichivosti system s nechetkimi parametrami. Avtomatika I Telemechanhika.” N.4, 98-109. /in Russian/ = “Robust stability of fuzzy-parameter systems. Automation and Remote Control,” 2005, N.4, 645-661 /in English/. [7] Le Hung Lan, “Modifisirovanui chastotnui criterii robastnoi ustoichivosti zamknutux system. Avtomatika I Telemexhanika”. N.8.119-130. 1993. /in Russian/
118118