Professional Documents
Culture Documents
PII: S0016-0032(17)30429-5
DOI: 10.1016/j.jfranklin.2017.08.035
Reference: FI 3118
Please cite this article as: Haoyi Que, Zheng-Guang Wu, Hongye Su, Globally Exponential Synchro-
nization for Dynamical Networks with Discrete-Time Communications, Journal of the Franklin Institute
(2017), doi: 10.1016/j.jfranklin.2017.08.035
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service
to our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and
all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
T
AbstractWe present a new synchronization criterion in been used to derive an upper bound for communication inter-
discrete-time communications for CDNs(complex dynamical net- vals in synchronization criterion, which could be considered as
IP
works) in this paper, which global exponential convergence is a pioneer in linear systems [18] to investigate the sampled-data
achieved. Between every two different nodes, the communica-
tions are discrete-time, at the same time between each two control, at the same time improved [19].
Regrettably, with some useful terms neglecting in the
CR
of the communication instants, the intervals are time-varying
and assumed to be bounded. We use time-dependent Lyapunov employed Lyapunov functional, we found that the available
functional approach to give globally exponential conditions under information for actual discrete-time communication pattern
the criterion for synchronization of CDNs, with less conservatism has not been fully applied in [26]. Besides, the free-weighting
and decision variables compared with some other existing results,
matrix method learning from [27], it is naturally used in the
such that the available information could be take full advantage
of in the actual discrete-time communication pattern. Some
simulations results are given at the end to show the effectiveness.
Index TermsComplex dynamical networks, discrete-time
communication, globally exponential synchronization, sampled-
US conditions of [26]. However, rather than reducing the conser-
vatism, some of them not only complex the synchronization
criteria, but also make a higher demand on computation. Thus,
it is important and necessary for further study that the problem
AN
data system of synchronization CDNs under discrete-time communications
should achieve a less conservative synchronization criterion
and less decision variables, which also motivates the current
I. I NTRODUCTION
work in this paper.
Complex dynamical networks (CDNs) could be established
M
decade, synchronization, as one of the basic and significant discrete-time communication pattern. The criterion given in
characteristics of CDNs, due to the physics and potential this paper decreases the number of decision variables com-
engineering applications, continued to be active, and thus pared with [26], meanwhile has less conservatism. Numerical
fruitful results have been reported in this field, e.g., [2][16] to examples and correspondent simulations are provided to show
PT
name a few. Sampled-data control/filtering problem, as another effectiveness of the theoretic results.
research frontier, still remains open and reports numerous
results. For example, [17][23] and the references therein. II. P RELIMINARIES
With the bloom of computer science, faster calculation speed
CE
( [24], [25] and the references therein). Different from CDNs t [tk , tk+1 ), i = 1, 2, . . . , N (1)
in [26] which is continuous-time, communication model based
where tk satisfy
on sampled-data control techniques in discrete-time situation,
has been introduced. Between each two of the communication 0 = t0 < t1 < < tk < < lim tk = +
k+
instants, the intervals are time-varying and assumed to be
bounded. Time-dependent Lyapunov functional approach has k = 0, 1, 2, . . . (2)
are the communication instants. N is the number of nodes.
This work was partially supported by the National Natural Science Foun-
dation of China (61174029, 61174058, 61573322), the National Key Basic f : Rn R+ Rn is a nonlinear vector-valued function.
T
Research Program, China (2012CB215202), and the 111 Project (B12018). xi (t) = [xi1 (t), xi2 (t), . . . , xin (t)] Rn is the state vector
Haoyi Que Z.-G. Wu and H. Su are with the National Laboratory of Indus- of node i. c is the coupling strength which is always a constant.
trial Control Technology, Institute of Cyber-Systems and Control, Zhejiang
University, Yuquan Campus, Hangzhou Zhejiang, 310027, PR China (e-mail: is a inner-coupling matrix of appropriate dimensions. G =
fatqcrab@163.com;nashwzhg@gmail.com;hysu@iipc.zju.edu.cn). (gij )N N is the coupling configuration matrix, where gij > 0
ACCEPTED MANUSCRIPT
represents different two node i and j P is connected, otherwise 0 0 0 0
N 1
without connection gij = 0. gii = j=1,j6=i gij define the 0 0 0
diagonal elements in G. Besides, , G and A are constant ..
0 1 0 .
matrices.
The distance that we mentioned in this paper, which =
.. .. RN (N 1)
0 0 . . 0
is between any two consecutive communication instants, is
. ..
assumed to be bounded. Specifically, it is assumed that .. . 1 0
tk+1 tk = hk 6 h, k 0, where h is the a allowable
0 0 0 1
communication interval and is a positive scalar.
Let
III. M AIN R ESULTS
x1 (t) f (x1 (t))
" 1: For
Theorem # a given scalar h, if there exist matrices
T
x (t) f(x (t))
2 2 U1 U2
x(t) =
..
RN n , g(t) =
..
RN n (3)
P > 0, > 0, X, X1 , H = H1 H2 H3 , and a
IP
. . U3
xN (t) f(xN (t)) diagonal matrix > 0, under the Assumption 1 we have
X + XT
CR
Therefore (1) could be shown as P + h + U 1 hX + hX 1 U1 U 2
2
x(t) = (IN A)x(t) + g(t) + c(G )x(tk ), t [tk , tk+1 ) U 2 >0
(4) U3
(8)
We introduce the following definition for CDN (1).
Definition 1 [3]: CDN (1) is said to be globally exponen-
tially synchronized if there exist constants and > 0, such US
that for any initial values xi (0) (i = 1, 2, . . . , N ) and sufficient
11
1 =
12
22
13
H2T
<0
(9)
AN
large T > 0, then 33
||xi (t) xj (t)|| 6 et (5)
11 12 13 hH1T
holds for all t > T and any i, j = 1, 2, . . . , N .
22 H2T hH2T
M
and then the nonlinear part of the node system satisfies the + ( In )(IN 1 ) + (IN 1 )( In )
following condition: + H1 + H1T
(f(x1 , t)f(x2 , t)) (f(x1 , t)f(x2 , t)) 6 (x1 x2 ) (x1 x2 ) 12
T T = P + H2
AC
(6) 13 = cP M (G ) U2 + X X1 + H3 H1T
Lemma 1 [26]: G RN N is a matrix that satisfy the same = 2( In )
22
sum for each row. The following equation holds:
X + XT
33 = U2 + U2T + X1 + X1T H3 H3T
M G = M GM (7) 2
X + XT X + XT
where 11 =h (IN 1 A) + h(IN 1 A)T
2 2
1 1 0 0 + h(IN 1 A)T U1 (IN 1 A)
1 0 1 0 0
X + XT
12 = h + h(IN 1 A)T U1
. .. .. 2
M = .. . . R(N 1)N
X + XT
1 0 0 1 0 13 = h(IN 1 A)T U2 + ch M (G )
2
T
1 0 0 1 + h(IN 1 A) (X + X1 )
ACCEPTED MANUSCRIPT
T
1
Then CDN (1) is globally exponentially synchronized. 1 U1 U2 U3 U2T U1 + U2 U31 U2T
U= (17)
Proof 1: Choose a scalar > 0 small enough, then h U1 U2 U31 U2T
IP
3 (h) = And jensens inequality is applied, hence,
" # " #
CR
11 + 11 12 + 12 13 + 13 y(t) y(t)
V (t) > (18)
22 + h(U1 + I) H2T + 23
< 0 (12) y(tk ) y(tk )
33 + 33 where
" #
where
and denote
ij = ij |U1 =U1 +I US =
t k+1
Meanwhile,
hk
t
( + hk (H + U)) +
t
hk
t k
, =
P
0
0
0
AN
hk h hk
x1 (t) x2 (t) + hk (H + U) = ( + h(H + U)) + (19)
x (t) x (t) h h
1 3
y(t) =
.
(13) Learning from (8), we could use Schur complement to get
.
.
M
Then we get
y(t) = (IN 1 A)y(t) + f (t) + cM (G )y(tk )
t [tk , tk+1 ) (14) V (t) > ||y(t)||2 (21)
where f (t) = M g(t).For system (14), Consider the following Now we get V (t) is an available Lyapunov function for
PT
Lyapunov functional: system (14) from the definition (15). Then comes the following
derivative equality and inequality:
V (t) = V1 (t) + V2 (t) + V3 (t), t [tk , tk+1 ) (15)
V1 (t) = 2y(t)T P y(t)
CE
T
and
where
V3 (t)
IP
" #T " # 11 12 13
y(t) y(t) 22 H2T
= H (hk ) =
y(tk ) y(tk )
CR
33 hk U3
" #T " # T T T
y(t) y(t) H1 H1
+ 2(tk+1 t) H
y(tk ) 0 H T 1 H T
+ hk 2 U3 2
" #T " #
H3T H3T
=
y(t)
y(tk )
+ 2(tk+1 t)y(t)T
H
X +X
y(t)
y(tk )
(IN 1 A)y(t)
T
US On the other hand, it can be shown that
3 (hk ) =
hk
3 (h) +
h hk
1 (29)
AN
2 h h
T
X +X and
+ 2(tk+1 t)y(t)T f (t) hk h hk
2 (hk ) =
(h) + 1 (30)
X +X T h h
+ 2c(tk+1 t)y(t)T M (G )y(tk )
2 Also from Schur complement and inequality (10), we have
M
Zt
V (t) 6 ||y(t)||2 ||y(tk )||2 ||y(s)||2 ds (33)
Zt " #T " #" #
(t) H T U11 H H T (t) tk
ds > 0 (25)
CE
>0
U1 where
tk+1 t t tk
From (25), we can immediately have that = ( + hk H) +
hk hk
Zt
And
y(s)T U1 y(s) ds 6 (t tk )(t)T H T U11 H(t)
hk h hk
tk + hk H = ( + hH) +
h h
+ 2(t)T H T (y(t) y(tk )) (26) hk h hk
6 ( + hH + U ) + (35)
Furthermore, by Assumption 1, for a diagonal matrix > 0, h h
the following inequality holds where
" #
U1 U1
2y(t) ( In )(IN 1 )y(t) 2f (t)( In )f (t) > 0
T
U = >0
(27) U1
ACCEPTED MANUSCRIPT
From (8), there exists a scalar 1 > 0 holds discrete-time communication pattern, which implies Theorem
1 proposed here is less conservative than that of [26].
+ hH + U < 1 I, P < 1 I (36) Remark 4: It should be pointed out that the total number of
scalar decision variables involved in [26] is 15m2 +2m+N 1,
Thus,
where m = (N 1)n, while the conditions in Theorem 1
V1 (t) + V3 (t) < 1 ||y(t)||2 + 1 ||y(tk )||2 (37) have only 7.5m2 + 1.5m + N 1 scalar decision variables.
More decision variables for granted increase the computational
Furthermore, another scalar 2 > 0 exists, accordingly burden. Surely our results would cost less time compared with
Zt [26] under the same condition. Especially with the increasing
of the number of nodes N and the dimensions of isolated node
V2 (t) < 2 ||y(s)||2 ds + 2 ||y(tk )||2 (38)
n, the computational superiority of our results becomes more
tk obvious.
T
> 0 is sufficiently small that we have
IV. N UMERICAL E XAMPLES AND S IMULATIONS
IP
1 6 0, 1 + 2 6 0, 2 6 0 (39) We choose Chuas circuit here, which could be considered
as the isolated nodes of the dynamical network.Then the
Then
following model holds:
CR
V (t) + V (t) 6 0 (40)
x1 = 1 (x1 + x2 u(x1 ))
For t [tk , tk+1 ), integrating (40) holds x2 = x1 x2 + x3
x3 = 2 x2
V (t) 6 e(ttk ) V (tk ) 6 e(ttk1 ) V (tk1 )
6 6 et V (0) = et y(0)T P y(0)
which combining (21) further implies that
US
(41)
where 1 = 10, 2 = 14.87, and u(x1 ) = 0.68x1 +
0.5(1.27 + 0.68)(|x1 + 1| |x1 1|). With initial condition
[0.7, 0.1, 0.05], the state trajectory could be shown in Fig.1.
AN
||y(t)||2 6 V (t) = O(et ) (42)
5
Thus, the globally exponential synchronization of CDN (1) is
achieved. Proof ends. 0
x3
M
2.5
Synchronization error
1.5
0.5
T
IP
0
0 1 2 3 4 5 6
Time(s)
CR
Fig. 2. Complex network topology Fig. 4. Synchronization error with topology in Fig.2
TABLE I
1 and the method in [26] are listed in Table I. Obtaining A LLOWABLE COMMUNICATION INTERVALS h
the trajectories of all the states with c = 1.53 tend to be
synchronized in Fig. 3.
US c
[26]
Theorem 1
1.53
0.008
0.025
1.55
0.014
0.044
2.00 3.00
0.045 0.033
0.052 0.037
TABLE II
4.00
0.026
0.029
5.00
0.022
0.024
AN
2
0
A LLOWABLE COMMUNICATION INTERVALS h
x1
-2
0 0.5 1 1.5 2 2.5 3 3.5 c 2.15 2.20 3.00 4.00 5.00 6.00
Time(s) [26] 0.025 0.030 0.060 0.063 0.058 0.053
0.5 Theorem 1 0.071 0.082 0.104 0.087 0.074 0.064
0
x2
-0.5
0 0.5 1 1.5 2 2.5 3 3.5
2
Time(s)
Theorem 1 only involves 624 decision variables. Thus, our
0
criterion uses much less decision variables, but this is not at the
x3
-2
0 0.5 1 1.5 2 2.5 3 3.5
Time(s) intervals, that is, compared with [26], Theorem 1 derived here
not only has less conservatism, but also has computational
Fig. 3. Trajectories of the state of complex network
advantages.
Remark 5: It should be pointed out that when the algorithm
PT
3 1 1 1 V. C ONCLUSION
1 3 1 1 The problem of globally exponential synchronization for
G= continuous-time CDNs with discrete-time communications is
1 1 3 1
considered. Based on the time-dependent Lyapunov func-
1 1 1 3
tional approach, an exponential synchronization criterion has
which means that the network is fully connected. For a given been established for the considered CDNs, which could take
coupling strength c, Table II gives the allowable communica- more advantage of the available information under the actual
tion intervals h by Theorem 1 and the method in [26]. discrete-time communication pattern. It has been shown that
Also it can be found from Tables I and II that for the the condition presented is less conservative with less compu-
same coupling strength c, our condition gives larger allowable tational burden than the existing results. The effectiveness and
communication intervals h than that in [26]. Furthermore, it superiority has been shown through numerical examples and
can be easily calculated that for this example the condition in simulations. Future works may be done trying to apply this
[26] involves 1236 decision variables, meanwhile in our work theorem into some applications.
ACCEPTED MANUSCRIPT
R EFERENCES [27] M. Wu, Y. He, J. She, and G. Liu, Delay-dependent criteria for robust
stability of time-varying delay systems, Automatica, vol. 40, pp. 1435
[1] X. Wang and G. Chen, Complex networks: Small-world, scale-free and 1439, 2004.
beyond, IEEE Circuits and Systems Magazine, vol. 3, pp. 620, 2003.
[2] J. Lu and G. Chen, A time-varying complex dynamical network
model and its controlled synchronization criteria, IEEE Transactions
on Automatic Control, vol. 50, pp. 841846, 2005.
[3] J. Lu, D. W. C. Ho, and J. Cao, A unified synchronization criterion for
impulsive dynamical networks, Automatica, vol. 46, pp. 12151221,
2010.
[4] Wu Z G, Park J H, Su H, et al. Exponential synchronization for com-
plex dynamical networks with sampled-data[J]. Journal of the Franklin
Institute, 2012, 349(9): 2735-2749.
[5] Choi H D, Ahn C K, Karimi H R, et al. Filtering of Discrete-Time
Switched Neural Networks Ensuring Exponential Dissipative and l-l
Performances[J]. IEEE Transactions on Cybernetics, 2017.
T
[6] Ahn C K, Shi P, Basin M V. Two-dimensional dissipative control and
filtering for Roesser model[J]. IEEE Transactions on Automatic Control,
IP
2015, 60(7): 1745-1759.
[7] Ahn C K, Shi P, Basin M V. Deadbeat dissipative FIR filtering[J]. IEEE
Transactions on Circuits and Systems I: Regular Papers, 2016, 63(8):
1210-1221.
CR
[8] J. Zhou, G. Chen, and J. Lu, Adaptive synchronization of an uncertain
complex dynamical network, IEEE Transactions on Automatic Control,
vol. 51, pp. 652656, 2006.
[9] Li X, Ding C, Zhu Q. Synchronization of stochastic perturbed chaotic
neural networks with mixed delays[J]. Journal of the Franklin Institute,
2010, 347(7): 1266-1280.
[10] Wang X, Fang J, Dai A, et al. Mean square exponential synchronization
for two classes of Markovian switching complex networks under feed-
US
back control from synchronization control cost viewpoint[J]. Journal of
the Franklin Institute, 2015, 352(8): 3221-3242.
[11] Shang-Guan X C, He Y, Lin W J, et al. Improved Synchronization
AN
of Chaotic Lure Systems with Time Delay Using Sampled-Data Con-
trol[J]. Journal of the Franklin Institute, 2016.
[12] J. Zhao, D. J. Hill, and T. Liu, Stability of dynamical networks
with non-identical nodes: A multiple V -Lyapunov function method,
Automatica, vol. 47, pp. 26152625, 2011.
[13] B. Liu, W. Lu, and T. Chen, Synchronization in complex networks
M
2002.
[15] Y.-W. Wang, H. O. Wang, J. Xiao, and Z. Guan, Synchronization of
complex dynamical networks under recoverable attacks, Automatica,
vol. 55, pp. 569576, 2010.
[16] W. Yu, P. DeLellis, G. Chen, M. Bernardo, and J. Kurths, Distributed
PT