Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
tracking-based AFC,
31
were devel-
oped to synchronize various types of chaotic systems.
Moreover, a state feedback AFC and an output feedback
AFC were also addressed in Refs. 32 and 33, respectively,
to control the uncertain GLS. However, there exist two
controllers and the synchronization problem is not consid-
ered in Refs. 32 and 33. Moreover, all the aforementioned
approaches can only achieve uniformly ultimately bounded
(UUB) stability.
This study focuses on asymptotically synchronizing two
uncertain GLSs involving unstructured uncertainties, param-
eter variations, and external disturbances via a single contin-
uous controller. Based on the work of Refs. 3537, a novel
CAFC, which utilizes both tracking and modeling error feed-
backs and has the potential to obtain smooth and quick pa-
rameter adaptation,
38
is developed for the synchronization
problem. The design procedure of the proposed controller is
as follows: First, the synchronization problem is transformed
into the stabilization problem by feedback linearization to
facilitate controller design; second, the CAFC structure is
established by introducing a series-parallel identication
model with a low-pass lter into an indirect AFC; third, a
key property of the optimal fuzzy approximation error (FAE)
is exploited by the Mean Value Theorem to facilitate asymp-
totic stability analysis; nally, the composite adaptive laws
are derived by the Lyapunov synthesis, where the closed-
loop system achieves asymptotic stability in the sense that
all involving signals are UUB and both tracking and model-
ing errors converge to zero.
The structure of this paper is organized as follows. The
dynamic model of the GLS, transformation of the synchroniza-
tion problem, and control objective are formulated in Sec. II.
The design procedure of the proposed approach is given in
Sec. III. Simulation results are shown in Sec. VI. Concluding
remarks are given in Sec. V.
II. PROBLEM FORMULATION
A. System transformation
The model of the GLS is as follows:
39
_ x
1
= (25a 10)(x
2
x
1
)
_ x
2
= (28 35a)x
1
x
1
x
3
(29a 1)x
2
_ x
3
= x
1
x
2
(8 a)x
3
=3;
_
_
_
(1)
where x = [x
1
; x
2
; x
3
[
T
R
3
is a state vector and a [0; 1[ is
an unknown system parameter. When a [0; 0:8), a = 0:8
and a (0:8; 1[, Eq. (1) belongs to the classical Lorenz sys-
tem, Lu system, and Chen system, respectively. Eq. (1) has
three unstable equilibria \a [0; 1[ shown as follows:
40
S
0
= (0; 0; 0)
S
= (
(8 a)(9 2a)
_
;
(8 a)(9 2a)
_
; 27 6a)
S
= (
(8 a)(9 2a)
_
;
(8 a)(9 2a)
_
; 27 6a):
_
_
(2)
Let Eq. (1) be the master system and dene the slave system
as the perturbed and controlled GLS:
21
_ y
1
= (25a 10)(y
2
y
1
) d
1
(t)
_ y
2
= (28 35a)y
1
y
1
y
3
(29a 1)y
2
d
2
(t) u
_ y
3
= y
1
y
2
(8 a)y
3
=3
;
_
_
_
(3)
where y = [y
1
; y
2
; y
3
[
T
R
3
is a state vector, u is a control
input, and d
1
and d
2
are mismatched and matched external
disturbances, respectively.
Dene the original tracking error vector e
o
:= y x =
[ e
o1
; e
o2
; e
o3
[
T
. Subtracting Eq. (1) from Eq. (3) yields the
tracking error dynamics in the afne nonlinear form:
_ e
o
= f (e
o
; y) b
0
u d(t)
y = h(e
o
);
_
(4)
where f , b
0
, d, and h are given by Eq. (11). Let L
f
h and L
g
h
denote the Lie derivatives of h with respect to f and g,
respectively. Since
L
g
h(e
o
) = 0
L
g
L
f
h(e
o
) =
0
1
0
_
_
_
_
T
(25a 10)
25a 10
0
_
_
_
_
= 25a 10 ,= 0;
_
_
(5)
the strong relative degree of Eq. (4) is equal to two.
38
There-
fore, there exists a diffeomorphism:
T(e
o
) =
e
g
_ _
=
e
o1
(25a 10)(e
o2
e
o1
)
e
o3
_
_
_
_
(6)
that transforms Eq. (4) into the perturbed controllable canon-
ical form:
41
_ e
1
= e
2
_ e
2
= f (e) g(e)u d(t)
_ g = q(e; g)
;
_
_
_
(7)
where e is a tracking error vector dened as e := [e
1
; e
2
[
T
= [e
1
; _ e
1
[
T
, f, g, d, and q are given by Eq. (12). From Eq.
(12), one knows that the system in Eq. (7) has the following
property as in Ref. 42.
Property 1: There exist a unknown function
f (e) and
unknown constants g, g, and
d such that [f (e)[ _
f (e),
0 < g _ [g(e)[ _ g, and [d(t)[ _
d, \e R
n
. Moreover,
f (e) satises f (0) = 0.
From Eq. (12), one also obtains the zero dynamic of the
system as follows:
_ g = q(0; g) = (8 a)g=3 (8)
which is asymptotically stable with a [0; 1[. Thus, one can
only consider the rst two equations of Eq. (7) during con-
troller design.
B. Control objective statement
Choose a gain vector k = [k
2
; k
1
[
T
R
2
so that h(s) =
k
1
s k
2
is a Hurwitz polynomial, where s is a complex
023144-2 Pan, Er, and Sun Chaos 22, 023144 (2012)
Downloaded 05 Jul 2012 to 155.69.4.4. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
variable. If f and g are known and d = 0 in Eq. (7), one can
apply the following ideal control law:
u
+
= (k
T
e f (e))=g(e) (9)
to Eq. (7) for obtaining the tracking error dynamics:
_ e = Ae (10)
where A = [0; 1; k
2
; k
1
[. From the selection of k, one
knows A is a stable matrix and lim
t
|e(t)|= 0. However,
since f and g are unknown and d ,= 0 in Eq. (7), u
+
in Eq. (9)
is unrealizable.
f (e
o
; y) =
(25a 10)(e
o2
e
o1
)
(28 35a)e
o1
e
o1
e
o3
e
o1
y
3
e
o3
y
1
(29a 1)e
o2
e
o1
e
o2
e
o1
y
2
e
o2
y
1
(8 a)e
o3
=3
_
_
_
_
;
b
0
= [0; 1; 0[
T
; d(t) = [d
1
(t); d
2
(t); 0[
T
; h(e
o
) = e
o1
:
_
_
(11)
f (e) = (25a 10)(27e
1
6ae
1
e
1
g e
1
y
3
gy
1
) (4a 11)e
2
;
g(e) = 25a 10; d(t) = L
2
f gud
L
d
e
1
2
k=1
L
k1
f gud
L
d
L
3k
f
e
1
;
q(e; g) = e
1
(e
1
e
2
=(25a 10)) e
1
y
2
(e
1
e
2
=(25a 10))y
1
(8 a)g=3:
_
_
_
(12)
Accordingly, the objective of this study is to synchron-
ize the slave system in Eq. (3) with the master system in
Eq. (1) by a single continuous AFC such that the closed-loop
system achieves asymptotic stability in the sense that all
involving signals are UUB and e converges to zero.
Remark 1: From the property of the diffeomorphism,
38
one knows that if e converges to zero or a small neighbor-
hood of zero as t , then e
o
also tents to zero or a small
value, which implies that the slave system in Eq. (3) is syn-
chronous with the master system in Eq. (1).
III. COMPOSITE ADAPTIVE FUZZY CONTROL
A. Indirect adaptive control structure
From the argument in Ref. 43, one knows that d can also
be estimated by a function approximator. Thus, to construct
a certain control law, one introduces the following FLS:
23
^
f (e[
^
h
f
) =
^
h
T
f
n(e) (13)
^ g(e[
^
h
g
) =
^
h
T
g
n(e) (14)
to approximate f
L
(e) := f (e) d and g(e), respectively,
where
^
h
f
,
^
h
g
R
M
are adjustable parameter vectors, and
n(e) = [n
1
(e); ; n
M
(e)[
T
R
M
is a fuzzy basic function
vector. The elements in n(e) are dened as follows:
n
l
1
l
2
(e) :=
2
i=1
l
A
l
i
i
(e
i
)
p
l
1
=1
p
l
2
=1
(
2
i=1
l
A
l
i
i
(e
i
))
; (15)
where A
l
i
i
is the linguistic variable of e
i
, l
A
l
i
i
is the membership
function of A
l
i
i
, p is the number of fuzzy partitions, M = p
2
is the number of fuzzy rules, l
1
l
2
= j, l
i
= 1; ; p, and
j = 1; ; M. Since f (0) = 0 and g(0) > 0 in Eq. (12),
^
f and ^ g
should be designed to satisfy
^
f (0[
^
h
f
) = 0 and ^ g(e[
^
h
g
) > 0.
Dene compact sets T := e[ |e|_ M
e
, X
f
:=
^
h
f
[
|
^
h
f
|_ M
f
and X
g
:=
^
h
g
[ |
^
h
g
|_ M
g
, where T is
the fuzzy approximation region, and M
e
, M
f
, M
g
R
. Therefore, combining
with Property 2, one has the following property as in Ref. 23.
Property 2: There exists a nite constant w R
such
that w = sup
\eT
[w[.
B. Identification model
Being motivated by Ref. 23, one introduces the series-
parallel identication model with a low-pass lter:
023144-3 Pan, Er, and Sun Chaos 22, 023144 (2012)
Downloaded 05 Jul 2012 to 155.69.4.4. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
_
^ e
1
= ^ e
2
_
^ e
2
= a^ e
2
ae
2
^
f (e[
^
h
f
) ^ g(e[
^
h
g
)u v
_
(21)
to achieve composite adaptation, where ^ e
i
is the estimation
of e
i
, i = 1; 2, a is a user-dened lter parameter, and v is a
modeling compensation term. Dene the modied modeling
error as follows:
35
e := e
2
^ e
2
: (22)
Then v in Eq. (21) can be given by
v = b sgn(e) (23)
in which b R
~
h
T
f
_
~
h
f
=c
f
~
h
T
g
_
~
h
g
=c
g
= e
T
Qe=2 c
e
ae
2
e
T
Pbw c
e
e(w v)
~
h
T
f
(e
T
Pbn(e) c
e
en(e)
_
~
h
f
=c
f
)
~
h
T
g
(e
T
Pbn(e)u c
e
en(e)u
_
~
h
g
=c
g
):
Using Eqs. (22), (32), and (33), one obtains
_
V
L
= e
T
Qe=2 c
e
ae
2
e
T
Pbw c
e
e(w v)
_ e
T
Qe=2 c
e
ae
2
e
T
Pbw:
Applying Eq. (25) to the above expression leads to
_
V
L
_ e
T
Qe=2 c
e
ae
2
|Pb| q(|e|) |e|
2
_ (k
min
(Q)=2 |Pb| q(|e|)) |e|
2
c
e
ae
2
:
Let k
Q
:= k
min
(Q)=2 |Pb| q(|e|). Noting Eq. (34), one
gets k
Q
R
, lim
t
|e(t)|= 0 and
lim
t
|e(t)|= 0. From Eq. (19), it is easy to obtain
u L