See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/389989399

An Improved Topology Identification Method of Complex Dynamical Networks

Article in IEEE Transactions on Cybernetics · May 2025

DOI: 10.1109/TCYB.2025.3547772

CITATIONS READS
0 91

5 authors, including:

Yi Zheng Xiaoqun Wu
Shenzhen University Shenzhen University
4 PUBLICATIONS 32 CITATIONS 176 PUBLICATIONS 5,540 CITATIONS

SEE PROFILE SEE PROFILE

Ziye Fan Jinhu Lu

Wuhan University Academy of Mathematics and Systems Science, Chinese Academy of Sciences
11 PUBLICATIONS 65 CITATIONS 301 PUBLICATIONS 23,444 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Yi Zheng on 26 April 2025.

The user has requested enhancement of the downloaded file.

IEEE TRANSACTIONS ON CYBERNETICS, VOL. 55, NO. 5, MAY 2025 2165

An Improved Topology Identification Method of

Complex Dynamical Networks
Yi Zheng , Xiaoqun Wu , Member, IEEE, Ziye Fan , Kebin Chen , and Jinhu Lü , Fellow, IEEE

Abstract—Over the past decade, numerous synchronization- been made in the study of complex networks. Synchronization
based identification methods have been proposed to address the has yielded significant advances in multiple research direc-
challenge of identifying unknown network topologies. The linear tions, as a focal research topic in complex networks, including
independence condition (LIC) is an essential requirement in these
methods, however, there are issues with this condition. In this consensus [4], [5], [6], [7], [8], network synchronization [9],
article, we propose an improved LIC-free synchronization-based [10], [11], [12], [13], and finite- or fixed-time synchro-
identification method to address above issues. Specifically, a drive nization [14], [15], [16], [17] superdiffusion [18]. Network
network consisting of isolated nodes that satisfy specific condi- topology has a significant impact on the dynamic behavior of
tions is constructed, and the network containing an unknown complex networks [19], leading to the topology identification
topology is defined as the response network. Through the design
of appropriate controllers and update laws, the drive network of complex networks becoming a research hotspot [20], [21],
and the response network achieve synchronization, while the [22], [23], [24], [25], [26], [27].
estimation matrix accurately identifies the unknown topology A variety of methods have been proposed to identify
matrix. Our method is proven to be a generalized form of network topology, with the synchronization-based identifica-
the existing LIC-free identification methods. Furthermore, we tion method being the focus of this article. In 2008, Wu [28]
introduce a novel proof framework to theoretically demonstrate
the effectiveness of our method. Finally, two simulation examples presented a synchronization-based identification method to
demonstrate the effectiveness of the proposed method. obtain the network topology. Subsequently, several identifi-
cation methods based on generalized synchronization have
Index Terms—Complex networks, linear independence
condition (LIC), synchronization, topology identification. been proposed. In 2014, Zhang et al. [29] proposed an
identification method based on generalized synchronization.
In 2020, Wang et al. [30] identified the topology of complex
networks by constructing an auxiliary network to achieve
I. I NTRODUCTION generalized synchronization. Extensive literature has also been
OMPLEX networks are widely present in nature, such published on the topology identification of various types
C as epidemic networks [1], information networks [2], and
brain neural networks [3]. In recent years, advancements have
of networks, including fractional-order networks [31], [32],
multiplex networks [33], and networks with random perturba-
tions [34], [35].
Received 29 September 2024; revised 27 February 2025; accepted 28 The above methods share one similarity: the linear inde-
February 2025. Date of publication 19 March 2025; date of current version pendence condition (LIC) serves as an important condition
24 April 2025. This work was supported in part by the National Natural
Science Foundation of China under Grant 92267101 and Grant 62141604; to ensure successful identification. In recent years, several
in part by the Shenzhen Science and Technology Program under Grant methods that do not require the LIC have been proposed.
KJZD20240903100022028; and in part by the Startup Grant from Shenzhen In 2021, Zhu et al. [36] presented a LIC-free identifica-
University. This article was recommended by Associate Editor G. Zhu.
(Corresponding author: Xiaoqun Wu.) tion method based on a regulation mechanism. In 2022,
Yi Zheng is with the College of Computer Science and Software Liu et al. [33] proposed a LIC-free identification method
Engineering, Shenzhen University, Shenzhen 518060, China, and also with the for multilink complex networks via chaotic exosignals. In
College of Information and Communication, National University of Defense
Technology, Wuhan 430019, China (e-mail: [email protected]). 2022, Fan and Wu [37] focused on identifying the partial
Xiaoqun Wu is with the College of Computer Science and Software topology of simplicial complexes and provided a rigorous
Engineering, Shenzhen University, Shenzhen 518060, China (e-mail: xqwu@ definition for the first time. In 2024, Li and Shi [38] proposed a
whu.edu.cn).
Ziye Fan is with the College of Computer Science and Software synchronization-based topology identification method for mul-
Engineering, Shenzhen University, Shenzhen 518060, China, and also with tilink hypergraphs, along with a verifiable linear independence
the School of Mathematics and Statistics, Wuhan University, Wuhan 430072, definition.
China (e-mail: [email protected]).
Kebin Chen is with the College of Information and Communication, Most of the aforementioned studies rely on the LIC. Three
National University of Defense Technology, Wuhan 430019, China (e-mail: issues with the LIC need to be highlighted. First, the LIC
[email protected]). is an essential condition to ensure the success of most
Jinhu Lü is with the School of Automation Science and Electrical
Engineering, Beihang University, Beijing 100191, China, and also synchronization-based identification methods. However, the
with Zhongguancun Laboratory, Beijing 100094, China (e-mail: jhlu@ detailed explanation of the LIC is omitted. Second, although
iss.ac.cn). it has been previously observed that the LIC may be satisfied
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TCYB.2025.3547772. if nodes in a network exhibit distinct dynamics or the network
Digital Object Identifier 10.1109/TCYB.2025.3547772 is subject to random perturbations, there is still a lack of an

2168-2267 
c 2025 IEEE. All rights reserved, including rights for text and data mining, and training of artificial intelligence
and similar technologies. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: SHENZHEN UNIVERSITY. Downloaded on April 25,2025 at 03:09:16 UTC from IEEE Xplore. Restrictions apply.
2166 IEEE TRANSACTIONS ON CYBERNETICS, VOL. 55, NO. 5, MAY 2025

effective approach to confirm whether a network satisfies the LIC-free methods have been proposed for identifying the
LIC. Third, global or partial synchronization of nodes in a topology of complex networks [33], [36]. The equations of
network has been reported to cause identification failure [39], an n-dimensional complex network, as described in [28], are
since the LIC does not hold. Therefore, the application scope represented as
of the synchronization-based identification method is limited
  
N
by the LIC. ẏi (t) = f i t, yi (t) +  cij Hyj (t − τ (t)) (1)
Based on the aforementioned discussion, this article presents j=1
an improved identification method that addresses the above
three issues. In our method, the drive network consists of iso- where yi (t) ∈ Rn is the state vector, i = 1, 2, . . . , N. f i : R+ ×
lated nodes, and the network with unknown topology is used Rn → Rn represents the self-dynamics function.  ∈ R+
as the response network. Synchronization between the two describes the coupling strength as a constant. C ∈ RN×N is the
networks is achieved by employing the proposed controllers unknown coupling matrix representing the network topology.
and update laws. Our method avoids the identification failure When i = j, if there is an edge connecting nodei to node j,
caused by network synchronization and does not require the then cij > 0, otherwise, cij = 0. Define cii = − N j=1 cij , i =
LIC. The main contributions are as follows. 1, 2, . . . , N. H ∈ RN×N represents the inner coupling matrix.
1) Our method does not require the LIC and its effec- τ (t) represents the transmission delay.
tiveness is guaranteed by a condition that is convenient In most synchronization-based identification meth-
to check. When the drive network satisfies a spe- ods [28], [29], [30], a network is constructed as follows:
cific assumption, the unknown topology matrix of
  
N
the response network will be successfully identified. ˙ŷ (t) = f t, ŷ (t) +  ĉij (t)H ŷj (t − τ (t)) + ūi (t). (2)
i i i
Furthermore, this assumption explicitly states the con- j=1
ditions that the drive network should satisfy to ensure
successful identification. This aspect is not addressed in ŷi (t) ∈Rn is the state vector, ĉij (t) is the estimated value of
the synchronization-based identification methods based cij , ūi (t) = −di (t)ēi (t) is the controller used to synchronize
on LIC. network (1) and network (2), and ēi (t) = ŷi (t) − yi (t),
2) Our method exhibits a broader scope of effectiveness i = 1, 2, . . . , N. The update laws are designed as ĉ˙ ij (t) =
compared to existing synchronization-based identifica- −δij ē i (t)H ŷj (t − τ (t)), and ḋi (t) = ki ēi  , i = 1, 2, . . . , N.
2

tion methods based on LIC. When existing identification Then, one has
methods fail, our method can still achieve successful 
N
identification, as demonstrated in the provided numerical lim c̃ij (t)Hyj (t − τ (t)) = 0 (3)
examples. t→+∞
j=1
3) In [36], the condition for ensuring successful iden-
tification requires the functions to be periodic. Our where c̃ij (t) = ĉij (t)−cij , i, j = 1, 2, . . . , N. One indispensable
method extends the requirement of periodic functions condition to ensure successful identification is the LIC, as
to nonperiodic ones, which significantly broadens the described as follows.
applicability. Moreover, our method is demonstrated to Assumption 1 [33, Assumption 2]: (Linear Independence
be more general compared to existing LIC-free methods. Condition) Suppose that {Hyi (t − τ (t))}N i=1 are linearly inde-
This article is organized as follows. In Section II, some pendent on the outer synchronization manifold {yi (t) =
useful preliminaries are proposed. In Section III, the improved ŷi (t)}N
i=1 .
identification method is introduced. In Section IV, two simula- Based on Assumption 1, it can be concluded that
tion examples are presented. Section V concludes this article. lim ĉij (t) = cij , i, j = 1, 2, . . . , N. (4)
Appendix provides proofs of lemmas. t→+∞

In [28], Assumption 1 is omitted, allowing (4) is directly

II. P RELIMINARIES obtained from (3). In [22] and [30], the LIC was proposed to
Notations: N, R+ , Rn , and Rn×m denote the set of positive ensure the validity of derivation. A limitation of Assumption 1
integers, the set of positive real numbers, the n-dimensional is that its description lacks sufficient detail, which makes it
Euclidean space, and the set of all n × m-dimensional difficult to state the role of the LIC in theoretical derivation.
real matrices, respectively. x = (x x)(1/2) represents the
Euclidean norm of a vector x, where x ∈ Rn . A and x rep- B. Problem Formulation
resent the transpose of matrix A and vector x, respectively. L2 To identify the topology matrix C in network (1), we
denotes the square-integrable function space. det(A) represents construct a drive network described as
the determinant of the matrix A.
xi (t) = vgi (t) (5)
A. Review of Existing Methods where xi (t) ∈ Rn is the state vector and constant vector
In 2008, a synchronization-based identification method was v ∈ Rn satisfies Hv = 0, i = 1, 2, . . . , N. The continuously
initially proposed to identify the unknown topology of com- differentiable function gi (t) : R+ → R is bounded and has
plex networks with time-varying delay [28]. Subsequently, bounded derivative.

Authorized licensed use limited to: SHENZHEN UNIVERSITY. Downloaded on April 25,2025 at 03:09:16 UTC from IEEE Xplore. Restrictions apply.
ZHENG et al.: IMPROVED TOPOLOGY IDENTIFICATION METHOD OF COMPLEX DYNAMICAL NETWORKS 2167

In addition, we add controllers to network (1) to obtain the If lim a(t) exists, ḣ2 (t) is bounded, and lim h1 (t) = 0,
t→+∞ t→+∞
following response network: then
  
N
lim ȧ(t) = 0.
ẏi (t) = f i t, yi (t) +  cij Hyj (t − τ (t)) + ui (t) (6) t→+∞
j=1
Lemma 2: Suppose a(t) : R → R is a differentiable
where ui (t) is the controller to be designed, i = 1, 2, . . . , N. In function such that
the remainder of this article, network (5) is referred to as the
lim ȧ(t) = 0.
drive network, and network (6) is referred to as the response t→+∞
network. The main objective of this article is to identify the
unknown topology matrix C = (cij )N×N of network (6) by Then, for ∀ tn , tn ∈ [Tn , Tn+1 )
designing appropriate controllers and update laws.     
lim a tn − a tn = 0
n→+∞

C. Mathematical Preliminaries with {Tn } defined in Assumption 2.

Some necessary assumptions and lemmas are introduced Lemma 3: Let ak (t) : R → R be a differentiable function
before presenting the main results. The corresponding proofs such that
can be found in the Appendix.
lim ȧk (t) = 0
Assumption 2: For an increasing sequence {Tn } satisfying: t→+∞
1) T1 ≥ 0;
and
2) lim Tn = +∞;
n→+∞
3) Tmin < Tn+1 − Tn < Tmax 
N
lim ak (t)gk (t) = 0 (8)
where Tmin and Tmax are positive constants, there exist at least t→+∞
N1 time points tn1 , tn2 , . . . , tnN , . . . , tnN1 ∈ [Tn , Tn+1 ), with N1 ≥ k=1

N, and matrix Bn ∈ RN×N , such that where gk (t) is defined in (5), and gk (t) satisfies |gk (t)| ≤ M,
M > 0, t ∈ [0, +∞), k = 1, 2, . . . , N. Then, for ∀ tn , tn ∈
|det(Bn )| >  [Tn , Tn+1 )
where 
N
   
⎛       ⎞ lim ak tn gk tn = 0
g1 tn1  g2 tn1  · · · gN tn1  n→+∞
k=1
⎜ g1 tn2 g2 tn2 · · · gN tn2 ⎟
⎜ ⎟
⎜ .. .. .. .. ⎟ with {Tn } defined in Assumption 2.
⎜ . .  ⎟
⎜ .   .   ⎟ Lemma 4: Suppose Assumption 2 holds. Let ak (t) : R → R
Gn = ⎜ g1 tN g2 tN · · · gN tN ⎟ (7)
⎜ n n n ⎟ be a differentiable function such that
⎜ .. .. .. .. ⎟
⎜ . . . . ⎟
⎝ ⎠ lim ȧk (t) = 0
g1 tnN1 g2 tnN1 · · · gN tnN1 t→+∞

and
with gk (t) defined in (5), k = 1, 2, . . . , N, and  is a positive
constant. The matrix Bn consists of N row vectors from Gn or 
N
lim ak (t)gk (t) = 0
N row vectors of the matrix resulting from several row sum t→+∞
k=1
transformations over Gn .
Assumption 3 [36, Assumption 2]: Let τ (t) be a differen- where gk is defined in (5), k = 1, 2, . . . , N. Then
tiable function satisfying lim τ̇ (t) ≤ ξ < 1, where ξ is a
t→+∞ lim ak (t) = 0, k = 1, 2, . . . , N. (9)
constant. t→+∞
Assumption 4: Nonlinear functions f i satisfy
Lemma 5: Suppose Assumptions 2–3 hold. Let ak (t) : R →
f i (t, y(t)) − f i (t, x(t)) ≤ Ly(t) − x(t). R be a differentiable function such that

i = 1, 2, . . . , N, for ∀ x(t), y(t) ∈ Rn , where L is a positive lim ȧk (t) = 0

t→+∞
constant.
Note that functions that satisfy Assumption 4 are common, and
such as chaotic systems and functions with bounded deriva- 
N
tives. lim ak (t)gk (t − τ (t)) = 0 (10)
t→+∞
Lemma 1 [36, Lemma 4]: The differentiable functions a(t), k=1
h2 (t) : [0, +∞) → Rm and the function h1 (t) : [0, +∞) → where gk is defined in (5), k = 1, 2, . . . , N. Then
Rm , m ∈ N, satisfy
lim ak (t) = 0, k = 1, 2, . . . , N. (11)
ȧ(t) = h1 (t) + h2 (t). t→+∞

Authorized licensed use limited to: SHENZHEN UNIVERSITY. Downloaded on April 25,2025 at 03:09:16 UTC from IEEE Xplore. Restrictions apply.
2168 IEEE TRANSACTIONS ON CYBERNETICS, VOL. 55, NO. 5, MAY 2025

III. M AIN R ESULTS Furthermore, one obtains

 t  t
The error between the ith node xi (t) in network (5) and the
ith node yi (t) in network (6) is defined as e(s) ds ≤ −
2
V̇(s)ds = V(0) − V(t) ≤ V(0).
0 0
ei (t) = xi (t) − yi (t), i = 1, 2, . . . , N. (12) Thus e(t) ∈ L2 .
From (19), it follows that V(t) is bounded.
The controllers to be designed in network (6) are According to (18), ei (t),  cij (t) and di (t) are bounded. Notice
that gi (t) is bounded, xi (t) and yi (t) are bounded. Since ei (t),

N
cij (t), di (t) and yi (t) are bounded, from (15) and (17), ė(t) is

ui (t) = ẋi (t) − f i (t, xi (t)) −  ĉij (t)Hyj (t − τ (t))
bounded. With ė(t) bounded and e(t) ∈ L2 , according to the
j=1
Barbǎlat lemma [40], it can be deduced that
+ di (t)ei (t) (13)
lim e(t) = 0. (20)
where ĉij (t) is the estimated value of element cij of the t→+∞
unknown topology matrix C, i, j = 1, 2, . . . , N. From (12), the From (16) and (20), one has
error system is
lim ĉ˙ ij (t) = lim e
i (t)Hyj (t − τ (t)) = 0. (21)
ėi (t) = ẋi (t) − ẏi (t), i = 1, 2, . . . , N. (14) t→+∞ t→+∞

i, j = 1, 2, . . . , N. Denote ėi (t) = hi1 (t) + hi2 (t), i =

Combining (6), (13), and (14), one has
1, 2, . . . , N, where
  
N  
ėi (t) = f i (t, xi (t)) − f i t, yi (t) +  
cij (t)Hyj (t − τ (t)) hi1 (t) = f i (t, xi (t)) − f i t, yi (t) ,
j=1 
N
− di (t)ei (t) (15) hi2 (t) = 
cij (t)Hyj (t − τ (t)) − di (t)ei (t). (22)
j=1
with c̃ij (t) = ĉij (t) − cij , i, j = 1, 2, . . . , N.
Theorem 1: Suppose Assumptions 2–4 hold. Then, the Since limt→+∞ hi1 (t) ≤ limt→+∞ Lei (t) = 0, one can
unknown topology matrix C in network (6) will be success- obtain
fully identified by the estimation matrix Ĉ(t) = (ĉij (t))N×N lim hi1 (t) = 0.
with controllers (13) and update laws t→+∞

ĉ˙ ij (t) = −δij e
i (t)Hyj (t − τ (t))
Calculating the derivative of (22), one has
 (16)
ḋi (t) = ki ei (t)ei (t) 
N
ḣi2 (t) = c̃˙ ij (t)Hyj (t − τ (t)) − ḋi (t)ei (t) − di (t)ėi (t)
where ki and δij are positive constants, i, j = 1, 2, . . . , N.
j=1
Proof: Since Assumption 4 holds, one obtains
  
N
f i (t, xi (t)) − f i t, yi (t)  ≤ Lei (t). (17) + (1 − τ̇ (t)) c̃ij (t)H ẏj (t − τ (t)).
j=1
i = 1, 2, . . . , N. The constructed Lyapunov function is
Since yj (t), c̃ij (t), di (t) and ėi (t) are bounded,
1  1  1 2
N N N
V(t) = ei (t)ei (t) + c̃ (t) from (12), (16), (21) and Assumption 3, one gets
2
i=1
2 δij ij
i=1 j=1 ei (t), c̃˙ ij (t), ḋi (t), τ̇ (t) are bounded. Since ġj (t) is bounded,
ẋj (t) is bounded. Thus ẏi (t) = ẋi (t) − ėi (t) is bounded. One
1  (di (t) − d∗ )2
N
+ (18) obtains ḣi2 (t) is bounded. This indicates lim ei (t) = 0,
t→+∞
2 ki
i=1 lim hi1 (t) = 0, and ḣi2 (t) is bounded. From Lemma 1 one
t→+∞
where d∗ = L + 1. With (15)–(17), one gets gets

N 
N 
N
lim ėi (t) = 0. (23)
V̇(t) ≤ L ei (t)2 +  c̃ij (t)e
i (t)Hyj (t − τ (t))
t→+∞
i=1 i=1 j=1 Further, according to (12), (15), and (23), one gets

N 
N
  
N
− di (t)ei (t)2 + di (t) − d∗ ei (t)2 lim c̃ij (t)Hyj (t − τ (t)) = 0, i = 1, 2, . . . , N. (24)
i=1 i=1 t→+∞
j=1

N 
N
−ε c̃ij (t)e
i (t)Hyj (t − τ (t)) From Assumption 3, one can obtain
i=1 j=1
lim (t − τ (t)) = +∞

N t→+∞
= − ei (t)2 . thus
i=1
lim ei (t − τ (t)) = lim ei (t − τ (t))
Defining e(t) = (e   
1 (t), e2 (t), . . . , eN (t)) ∈ R , one has
nN t→+∞ (t−τ (t))→+∞
 
= lim xi (t − τ (t)) − yi (t − τ (t)) = 0. (25)
V̇(t) ≤ −e(t)2 . (19) (t−τ (t))→+∞

Authorized licensed use limited to: SHENZHEN UNIVERSITY. Downloaded on April 25,2025 at 03:09:16 UTC from IEEE Xplore. Restrictions apply.
ZHENG et al.: IMPROVED TOPOLOGY IDENTIFICATION METHOD OF COMPLEX DYNAMICAL NETWORKS 2169

Considering (24), (25) and c̃ij (t) is bounded, one gets Remark 1: In 2021, Zhu et al. [36, Assumption 3] proposed
that gk (t) in network (5) satisfies the following condi-

N (p) (p)
tion. Suppose that there exist t1 , . . . , tp ∈ [0, T), p =
lim c̃ij (t)Hyj (t − τ (t)) 1, 2, . . . , N, satisfying
t→+∞
j=1

p 

N
  (p) δp = 0, k = p
= lim c̃ij (t)H xj (t − τ (t)) − ej (t − τ (t)) gk tl =
t→+∞ 0, 1≤k<p
j=1 l=1


N
where T is the common period of gk (t). Note that the
= lim c̃ij (t)Hvgj (t − τ (t)) = 0. (p) (p)
t→+∞ number of time points t1 , . . . , tp ∈ [0, T), p = 1, 2, . . . , N,
j=1
is [N(N + 1)/2]. Set Tn = nT, and based on the above
Notice that Hv = 0, then assumption, one obtains the following matrix:
⎛ ⎞

N
g1 t1(1) g2 t1(1) g3 t1(1) · · · gN t1(1)
lim c̃ij (t)gj (t − τ (t)) = 0. (26) ⎜ ⎟
t→+∞ ⎜ g t(2) g t(2) g t(2) · · · g t(2) ⎟
j=1 ⎜ 1 1 2 1 3 1 N 1 ⎟
⎜ ⎟
⎜ g t(2) g t(2) g t(2) · · · g t(2) ⎟
Since c̃ij (t) = ĉij (t) − cij , (21) indicates ⎜ 1 2 2 2 3 2 N 2 ⎟
⎜ ⎟
⎜ g t(3) g t(3) g t(3) · · · g t(3) ⎟
⎜ 1 1 2 1 3 1 N 1 ⎟
lim c̃˙ ij (t) = 0. (27) ⎜
⎜ .. .. .. .
⎟
⎟
t→+∞ ⎜ . . . ⎟.
⎜ . . . . . ⎟
⎜ (N) (N) (N) (N) ⎟
According to Assumption 2, Assumption 3, (26), (27), and ⎜ g1 t1 g2 t1 g3 t1 · · · gN t1 ⎟
Lemma 5, one has ⎜ ⎟
⎜ (N) (N) (N) (N) ⎟
⎜ g1 t2 g2 t2 g3 t2 · · · gN t2 ⎟
  ⎜ ⎟
lim c̃ij (t) = lim ĉij (t) − cij = 0. (28) ⎜ . . . . ⎟
t→+∞ t→+∞ ⎜ .. .. .. .. ⎟
⎝ ⎠
(N) (N) (N) (N)
Therefore, (28) indicates that the unknown topology matrix C g1 tN g2 tN g3 tN · · · gN tN
is successfully identified by the estimation matrix Ĉ(t).
Theorem 2: Suppose Assumptions 3 and 4 hold, and For the above matrix, add the rows after the (i(i − 1) + 2)/2th
row to the (i(i − 1) + 2)/2th row, i = 1, 2, . . . , N. Then, one
gk (t) = sin 2k−1 t , k = 1, 2, . . . , N. obtains
⎛ ⎞
δ1 ∗ ∗ ··· ∗
Then, the estimation matrix Ĉ(t) = (ĉij (t))N×N will asymptot- ⎜ 0 δ2 ∗ ··· ∗ ⎟
⎜ ⎟
ically approach the unknown topology matrix of the response ⎜ g t(2) g t(2) g t(2) · · · g t(2) ⎟
⎜ 1 2 2 2 3 2 N 2 ⎟
network (6) under the action of the drive network (5), con- ⎜ ⎟
⎜ 0 0 δ · · · ∗ ⎟
trollers (13), and update laws (16). ⎜ 3 ⎟
Proof: There exist time points ti = π/2i ∈ [0, 2π ), i = ⎜ .. .. .. . .. .
.. ⎟
⎜ . . . ⎟
1, 2, . . . , N, such that | det(G) |> 1/2, where GN = ⎜ ⎟
⎜ 0 0 0 · · · δ ⎟
⎜ N ⎟
⎛ ⎞ ⎜ (N) (N) (N) (N) ⎟
g1 (t1 ) g2 (t1 ) g3 (t1 ) · · · gN (t1 ) ⎜ g1 t2 g2 t2 g3 t2 · · · gN t2 ⎟
⎜ g1 (t2 ) g2 (t2 ) g3 (t2 ) · · · gN (t2 ) ⎟ ⎜ ⎟
⎜ ⎟ ⎜ . . . . ⎟
⎜ .. .. .. .. ⎟
⎜ ⎟ ⎝ ⎠
G = ⎜ g1 (t3 ) g2 (t3 ) g3 (t3 ) · · · gN (t3 ) ⎟ (N) (N) (N) (N)
⎜ .. .. .. . .. .. ⎟
. g1 tN g2 tN g3 tN · · · gN tN
⎝ . . . ⎠
g1 (tN ) g2 (tN ) g3 (tN ) · · · gN (tN ) where ∗ represents a constant that does not affect the sub-
⎛ ⎞
1 0 0 ··· 0 sequent derivations. Notice that the N rows of matrix GN
⎜∗ 1 0 ··· 0⎟ containing δi = 0, i = 1, 2, . . . , N, can form the matrix
⎜ ⎟
⎜ ⎟
= ⎜∗ ∗ 1 ··· 0⎟ ⎛ ⎞
⎜ .. .. .. . . .. ⎟ δ1 ∗ ∗ · · · ∗
⎝. . . . .⎠ ⎜ 0 δ2 ∗ · · · ∗ ⎟
∗ ∗ ∗ ··· 1 ⎜ ⎟
BN = ⎜ . . . . .. ⎟
. .
⎝ . . . . . ⎠. .
where ∗ represents a constant that does not affect the sub- 0 0 0 · · · δN
sequent derivations. Notice that 2π is the common period of
gk (t), k = 1, 2, . . . , N. Set Tn = 2nπ , tni = π/2i + Tn ∈ such that |det(BN )| > , where  is a positive constant.
(p) (p)
[Tn , Tn+1 ), i = 1, 2, . . . , N, and then Assumption 2 holds. This indicates that there exist time points t1 , . . . , tp ∈
From Theorem 1, one gets [Tn , Tn+1 ), p = 1, 2, . . . , N, satisfying |det(BN )| > , where
BN is composed of N row vectors of matrix GN . Therefore,
lim ĉij (t) = cij , i, j = 1, 2, . . . , N. [36, Assumption 3] is a special case of Assumption 2.
t→+∞

Authorized licensed use limited to: SHENZHEN UNIVERSITY. Downloaded on April 25,2025 at 03:09:16 UTC from IEEE Xplore. Restrictions apply.
2170 IEEE TRANSACTIONS ON CYBERNETICS, VOL. 55, NO. 5, MAY 2025

-1

-2

-3

-4
0 1 2 3 4 5 6 7 8 9 10
t

Fig. 1. Identification result of network (29) based on Theorem 2: The Fig. 3. Identification result of network (29) based on the method in [28]: Time
evolution of the off-diagonal elements ĉij (t), i = 1, 2, . . . , 6, i  = j. Fig. 1 evolution of first-row elements ĉ1i (t), i = 1, 2, . . . , 6. Fig. 3 indicates that the
indicates that the estimated values tend to be the true values. This indicates estimated values do not tend to the true values, which means identification
that the estimated values tend to be the true values. failure.

1
1

0
0

-1
-1
-2

-2
-3

-4 -3

-5 -4

-6
0 5 10 15 20 -5
0 1 2 3 4 5 6 7 8 9 10
t
t

Fig. 2. Identification result of network (29) based on Theorem 2: The Fig. 4. Identification result of network (29) based on the method in [28]: Time
evolution of diagonal elements ĉii (t), i = 1, 2, . . . , 6. Fig. 2 indicates that the evolution of diagonal elements ĉii (t), i = 1, 2, . . . , 6. Fig. 4 indicates that the
estimated values tend to be the true values. This indicates that the estimated estimated values do not tend to the true values, which means identification
values tend to be the true values. failure.

IV. N UMERICAL S IMULATIONS The initial states are set as yi (0) = (1 + 0.2i, 2 + 0.3i, 3 +
In the following two examples, consider the following 0.4i) , di (0) = 10, and ĉij (0) = 0, i, j = 1, 2, . . . , 6. Other
network: parameters are taken as ki = 10,  = 100, and δij = 0.01,
  
6 i, j = 1, 2, . . . , 6.
ẏi (t) = f i t, yi (t) +  cij Hyj (t − τ (t)) (29) Example 2: Based on the method in [28], the following
j=1 network is obtained:
where i = 1, 2, . . . , 6, H = I3 , τ (t) = 0.2, and   
6
⎛ ⎞ ˙ŷ (t) = f t, ŷ (t) +  ĉij (t)H ŷj (t − τ (t)) − di (t)ei (t)
−5 1 1 1 1 1 i i i
⎜ 1 −5 1 1 1 1 ⎟ j=1
⎜ ⎟
⎜ 1 1 −5 1 1 1 ⎟
C=⎜ ⎟ with ei (t) = ŷi (t) − yi (t), ĉ˙ ij (t) = −δij ei (t)H ŷj (t − τ (t)),
⎜ 1 1 1 −5 1 1 ⎟.
⎜ ⎟ and ḋi (t) = ki ei 2 , i = 1, 2, . . . , 6. The same node dynamics
⎝ 1 1 1 1 −5 1 ⎠
and fully connected mode might lead to identification failure.
1 1 1 1 1 −5
Figs. 3 and 4 show that the estimated value curves do not
The dynamics of all nodes are defined as f i (t, w) = −w, where converge to the true values, indicating that the identification
w ∈ R3 . It is certain that Assumptions 3 and 4 hold. method in [28] has failed. Therefore, the identification method
Example 1: Based on Theorem 2, C will be identified, and presented in this article is demonstrated to be superior.
the numerical results are shown in Figs. 1 and 2. Example 1 The initial states are set as yi (0) = (1 + 0.2i, 2 + 0.3i, 3 +
verifies the validity of our method. 0.4i) , ŷi (0) = (3 + 0.1i, 2 + 0.4i, 1 + 0.6i) , di (0) = 10, and

Authorized licensed use limited to: SHENZHEN UNIVERSITY. Downloaded on April 25,2025 at 03:09:16 UTC from IEEE Xplore. Restrictions apply.
ZHENG et al.: IMPROVED TOPOLOGY IDENTIFICATION METHOD OF COMPLEX DYNAMICAL NETWORKS 2171

ĉij (0) = 1, i, j = 1, 2, . . . , 6. Other parameters are taken as when Tn > T 2 , tn ∈ [Tn , Tn+1 ). Since
ki = 1,  = 1, and δij = 1, i, j = 1, 2, . . . , 6.
lim ȧk (t) = 0
t→+∞
V. C ONCLUSION and ak (t) is a differentiable function, k = 1, 2, . . . , N, from
We propose an improved identification method, based on Lemma 2, one obtains
existing identification methods, that does not require the linear     
lim ak tn − ak tn = 0, k = 1, 2, . . . , N
independence condition (LIC). In this article, we introduce an n→+∞
assumption satisfied by the drive network to ensure successful
for ∀ tn , tn ∈ [Tn , Tn+1 ). For ε/2MN > 0, ∃ T 3 > 0 ∀ Tn > T 3
identification and prove that our method is a generalized
such that
form of existing methods. Discovering additional function      ε

sequences that satisfy Assumption 2 is an open and chal- ak tn − ak tn  < , k = 1, 2, . . . , N. (32)
lenging problem. Numerical examples show that our method 2MN
can accurately identify the unknown topology matrix of com- For the ε/2 in (31) and ε/2MN in (32), ∃ T 4 = max{T 2 , T 3 }
plex networks, even the identification method based on LIC such that
fails. This work provides a theoretical method for identifying  
 N    
unknown parameters in networks, including power grids,  a t g t 
 k n k n 
circuit systems, neural networks, and other systems. Existing k=1
synchronization-based identification methods primarily use  
 N         
analog signals, whereas digital signals are more commonly 
= ak tn − ak tn + ak tn gk tn 
used in real-world applications. Therefore, extending this k=1
   N 
method to discrete systems is a significant work.  N             

≤  
ak tn − ak tn gk tn  +  ak tn gk tn 
A PPENDIX k=1 k=1
ε ε
Proof of Lemma 2: When tn = tn , one has < MN +
2MN 2
    = ε.
a tn − a tn = 0.
∀ Tn > T 4 , ∀ tn , tn ∈ [Tn , Tn+1 ). Then one has
When tn = tn , without loss of generality, let tn < tn ∈
[Tn , Tn+1 ), one has 
N
   
      lim ak tn gk tn = 0
n→+∞
|a tn − a tn | = |ȧ(ξ ) tn − tn | < |ȧ(ξ )|Tmax k=1

with ξ ∈ (tn , tn ). Note that for ∀ t , t ∈ [Tn , Tn+1 ).

Proof of Lemma 4: Suppose the conclusion is incorrect.
lim ȧ(t) = 0 There is at least one of ak (t), k = 1, 2, . . . , N, that does not
t→+∞
satisfy (9). Without loss of generality, suppose a1 (t) satisfies
holds, then for ∀ (/T) > 0, ∃ T 1 > 0 such that for t > T 1 , lim a1 (t) = 0. This means that there exists ε0 > 0 and an
|ȧ(t)| < ε/Tmax . For the above ε/Tmax , when ξ > T 1 , one has t→+∞
increasing time series {t(m) ∈ [Tm , Tm+1 ), m ∈ N} such that
ε   
|ȧ(ξ )| < . (30) a1 t(m)  > ε0 (33)
Tmax
For the ε in (30), ∃ Tn > T 1 such that where t(m) → +∞, m → +∞. Since ȧk (t) → 0, t → +∞,
    ε from Lemma 2, one has
|a tn − a tn | < |ȧ(ξ )|Tmax < Tmax = ε.   
Tmax lim a1 (Tm ) − a1 t(m) = 0, k = 1, 2, . . . , N. (34)
m→+∞
This implies
     According to Lemma 3, one has
lim a tn − a tn = 0
n→+∞ 
N
 
lim ak (Tm )gk tm = 0 (35)
for ∀tn , tn ∈ [Tn , Tn+1 ), tn = tn . m→+∞
k=1
Thus, one obtains
for ∀ tm ∈ [Tm , Tm+1 ), k = 1, 2, . . . , N. According to the
     1 , t2 , . . . , tN , . . . , tN1 .
lim a tn − a tn = 0 arbitrariness of tm , (35) holds for tm = tm m m m
n→+∞
It follows that
for ∀tn , tn ∈ [Tn , Tn+1 ). 
N
i
Proof of Lemma 3: From (8), for ∀ ε/2 > 0, ∃ T 2 > 0 such lim ak (Tm )gk tm = 0, i = 1, 2, . . . , N1 .
m→+∞
that k=1
 N 
      ε Since gk (tm
i ) satisfy (7) and
 ak tn gk tn  < (31)
 2
k=1 |det(Bm )| > .

Authorized licensed use limited to: SHENZHEN UNIVERSITY. Downloaded on April 25,2025 at 03:09:16 UTC from IEEE Xplore. Restrictions apply.
2172 IEEE TRANSACTIONS ON CYBERNETICS, VOL. 55, NO. 5, MAY 2025

k = 1, 2, . . . , N, i = 1, 2, . . . , N1 , for ∀ ak (Tm ), k = R EFERENCES

1, 2, . . . , N, [1] R. Pastor-Satorras and A. Vespignani, “Epidemic spreading in scale-free
⎛  1  1  1 ⎞ ⎛ ⎞
networks,” Phys. Rev. Lett., vol. 86, pp. 3200–3203, Apr. 2001.
g1 tm  g2 tm  ··· gN tm  0 [2] B. Ahlgren, C. Dannewitz, C. Imbrenda, D. Kutscher, and B. Ohlman,
⎜ g1 tm 2 g2 tm 2 ··· gN tm 2 ⎟⎛ ⎞ ⎜0⎟ “A survey of information-centric networking,” IEEE Commun. Mag.,
⎜ ⎟ a1 (Tm ) ⎜ ⎟ vol. 50, no. 7, pp. 26–36, Jul. 2012.
⎜ .. .. .. .. ⎟
⎜ . ⎟⎜ a2 (Tm ) ⎟ ⎜ .. ⎟
⎜ .  . . N  ⎟ ⎜ .⎟
[3] D. S. Bassett and O. Sporns, “Network neuroscience,” Nat. Neurosci.,
⎟⎜
lim ⎜ g1 tN g2 tN  
··· ⎟⎜ .. ⎟ = ⎜ ⎟ vol. 20, no. 3, pp. 353–364, 2017.
m→+∞⎜ m m gN tm ⎟⎝ . ⎠ ⎜0⎟
⎜
⎟ [4] X. Wu, B. Mao, X. Wu, and J. Lü, “Dynamic event-triggered leader-
⎜ .. .. .. .. ⎟ ⎜ .. ⎟
⎜ . . . . ⎟ aN (Tm ) ⎝.⎠
follower consensus control for MultiAgent systems,” SIAM J. Control
⎝ ⎠ Optim., vol. 60, no. 1, pp. 189–209, 2022.
N1 N1 N1
g1 tm g2 tm · · · gN tm 0 [5] H. Su, H. Wu, and J. Lam, “Positive edge-consensus for nodal networks
via output feedback,” IEEE Trans. Autom. Control, vol. 64, no. 3,
pp. 1244–1249, Mar. 2019.
holds, if and only if [6] B. Mao, X. Wu, J. Lü, and G. Chen, “Predefined-time bounded
consensus of multiagent systems with unknown nonlinearity via dis-
tributed adaptive fuzzy control,” IEEE Trans. Cybern., vol. 53, no. 4,
lim ak (Tm ) = 0, k = 1, 2, . . . , N. (36) pp. 2622–2635, Apr. 2023.
m→+∞
[7] B. Mao, X. Wu, H. Liu, Y. Xu, and J. Lü, “Adaptive fuzzy tracking
control with global prescribed-time prescribed performance for uncertain
When k = 1, combining (36) and (34), for ∀ tm ∈ [Tm , Tm+1 ) strict-feedback nonlinear systems,” IEEE Trans. Cybern., vol. 54, no. 9,
pp. 5217–5230, Sep. 2024.
  [8] B. Mao, X. Wu, Z. Fan, J. Lü, and G. Chen, “Performance-guaranteed
lim a1 tm = 0. (37) finite-time tracking of strict-feedback systems with unknown control
m→+∞
directions: A novel switching mechanism,” IEEE Trans. Autom. Control,
early access, Jan. 13, 2025, doi: 10.1109/TAC.2025.3528344.
It is concluded that (37) contradicts (33), and thus
[9] X. Wu, X. Zhou, J. Liu, Q. Yang, and Y. Chen, “Synchronizability of
lim a1 (t) = 0 is not true. time-varying structured duplex dynamical networks with different intra-
t→+∞
Therefore, one obtains layer rewiring mechanisms,” Sci. China Technol. Sci., vol. 65, no. 2,
pp. 375–385, 2022.
[10] W. Yu, P. DeLellis, G. Chen, M. Di Bernardo, and J. Kurths, “Distributed
lim ak (t) = 0, k = 1, 2, . . . , N. adaptive control of synchronization in complex networks,” IEEE Trans.
t→+∞ Autom. Control, vol. 57, no. 8, pp. 2153–2158, Aug. 2012.
[11] R. Berner, S. Vock, E. Schöll, and S. Yanchuk, “Desynchronization
Proof of Lemma 5: From Assumption 3, ∃ t0 > 0 such that transitions in adaptive networks,” Phys. Rev. Lett., vol. 126, Jan. 2021,
τ̇ (t) < 1, for ∀t > t0 . Taking θ (t) = t − τ (t), it means that Art. no. 28301.
[12] X. Wu et al., “Synchronization in multiplex networks,” Phys.
θ̇ (t) > 0 on (t0 , +∞). Since Rep., vol. 1060, pp. 1–54, Apr. 2024. [Online]. Available:
 t  t  https://doi.org/10.1016/j.physrep.2024.01.005
1+ξ [13] W. Zhou, K. Wang, and W. Zhu, “Synchronization for discrete coupled
θ̇ (s)ds = θ (t) − θ (t0 ) ≥ 1− ds fuzzy neural networks with uncertain information via observer-based
t0 t0 2
impulsive control,” Math. Model. Control, vol. 4, no. 1, pp. 17–31, 2024.
1−ξ [14] Y. Xu, X. Wu, B. Mao, J. Lü, and C. Xie, “Fixed-time Synchronization
= (t − t0 ) in the pth moment for time-varying delay stochastic multilayer
2
networks,” IEEE Trans. Syst., Man, Cybern., Syst., vol. 52, no. 2,
pp. 1135–1144, Feb. 2022.
holds for ∀t > t0 , it means that lim θ (t) = +∞. The inverse
t→+∞ [15] X. Liu and T. Chen, “Finite-time and fixed-time cluster synchronization
function t(θ ) : (t0 − τ (t0 ), +∞) → (t0 , +∞) of θ (t) exists with or without pinning control,” IEEE Trans. Cybern., vol. 48, no. 1,
and satisfies lim t(θ ) = +∞. Define ωk (θ ) = ak (t(θ )), k = pp. 240–252, Jan. 2018.
θ→+∞ [16] T. Shi, C. Hu, J. Yu, Q. Zhu, and T. Huang, “Internal/boundary control-
1, 2, . . . , N. According to (10), one obtains based fixed-time Synchronization for spatiotemporal networks,” IEEE
Trans. Cybern., vol. 54, no. 11, pp. 6693–6706, Nov. 2024.
[17] Y. Chu, X. Han, and R. Rakkiyappan, “Finite-time lag synchronization

N
for two-layer complex networks with impulsive effects,” Math. Model.
lim ωk (θ )gk (θ ) = 0. Control, vol. 4, no. 1, pp. 71–85, 2024.
θ→+∞
k=1 [18] J. Wei, X. Wu, J.-A. Lu, J. Lü, and G. Chen, “A topological mechanism
of superdiffusion on duplex networks,” IEEE Trans. Control Netw. Syst.,
Then one has vol. 10, no. 2, pp. 556–563, Jun. 2023.
[19] E. Sovrano, “How to construct complex dynamics? A note on a
 dω (θ )   da (t(θ )) dt  topological approach,” Int. J. Bifurc. Chaos, vol. 30, no. 2, 2020,
 k   k 
lim   = lim   Art. no. 2050034.
θ→+∞ dθ θ→+∞ dt dθ [20] S. Boccaletti et al., “The structure and dynamics of multilayer
 da (t(θ ))   
 k   1  networks,” Phys. Rep., vol. 544, no. 1, pp. 1–122, 2014.
≤ lim   lim   [21] S. Boccaletti, V. Latora, Y. Moreno, M. Chavez, and D.-U. Hwang,
θ→+∞ dt θ→+∞ 1 − τ̇ (t)
 da (t)  “Complex networks: Structure and dynamics,” Phys. Rep., vol. 424,
 k  nos. 4–5, pp. 175–308, 2006.
≤ lim   = 0, k = 1, 2, . . . , N. [22] H. Liu, J.-A. Lu, J. Lü, and D. J. Hill, “Structure identification of uncer-
t→+∞ dt
tain general complex dynamical networks with time delay,” Automatica,
vol. 45, no. 8, pp. 1799–1807, 2009.
This indicates lim |[dωk (θ )/dθ ]| = 0. From Lemma 4, [23] G. Mei, X. Wu, Y. Wang, M. Hu, J.-A. Lu, and G. Chen, “Compressive-
θ→+∞
one can obtain lim ωk (θ ) = 0. Therefore sensing-based structure identification for multilayer networks,” IEEE
θ→+∞ Trans. Cybern., vol. 48, no. 2, pp. 754–764, Feb. 2018.
[24] H. J. van Waarde, P. Tesi, and M. K. Camlibel, “Topology
lim ak (t) = lim ak (t(θ )) = 0. identification of heterogeneous networks: Identifiability and reconstruc-
t→+∞ r→+∞ tion,” Automatica, vol. 123, Jan. 2021, Art. no. 109331.

Authorized licensed use limited to: SHENZHEN UNIVERSITY. Downloaded on April 25,2025 at 03:09:16 UTC from IEEE Xplore. Restrictions apply.
ZHENG et al.: IMPROVED TOPOLOGY IDENTIFICATION METHOD OF COMPLEX DYNAMICAL NETWORKS 2173

[25] H. J. Van Waarde, P. Tesi, and M. K. Camlibel, “Topology reconstruction Xiaoqun Wu (Member, IEEE) received the B.Sc.
of dynamical networks via constrained Lyapunov equations,” IEEE degree in applied mathematics and the Ph.D.
Trans. Autom. Control, vol. 64, no. 10, pp. 4300–4306, Oct. 2019. degree in computational mathematics from Wuhan
[26] Z. Fan, X. Wu, B. Mao, and J. Lü, “Output discernibility of topological University, Wuhan, China, in 2000 and 2005,
variations in linear dynamical networks,” IEEE Trans. Autom. Control, respectively.
vol. 69, no. 8, pp. 5524–5530, Aug. 2024. She currently serves as a Distinguished Professor
[27] Y. Zang, Z. Fan, Z. Wang, Y. Zheng, L. Ding, and X. Wu, “Stepwise with the College of Computer Science and Software
reconstruction of higher-order networks from dynamics,” Chaos, Engineering, Shenzhen University, Shenzhen, China.
Interdiscip. J. Nonlin. Sci., vol. 34, no. 7, 2024, Art. no. 73156. Prior to this, she was a Professor with the School
[28] X. Wu, “Synchronization-based topology identification of weighted of Mathematics and Statistics, Wuhan University,
general complex dynamical networks with time-varying coupling and held several visiting positions in Hong Kong,
delay,” Physica A, Stat. Mech. Appl., vol. 387, no. 4, pp. 997–1008, Australia, and USA. Her current research interests include parameter identi-
2008. fication and control of networked systems. She has published over 140 SCI
[29] S. Zhang, X. Wu, J.-A. Lu, H. Feng, and J. Lü, “Recovering structures journal papers in these areas.
of complex dynamical networks based on generalized outer synchro- Prof. Wu was a recipient of the Second Prize of the Natural Science Award
nization,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 61, no. 11, from the Hubei Province, China, in 2006, the First Prize of the Natural
pp. 3216–3224, Nov. 2014. Science Award from the Ministry of Education of China in 2007, and the
[30] Y. Wang, X. Wu, J. Lü, J.-A. Lu, and R. M. D’Souza, “Topology First Prize of the Natural Science Award from the Hubei Province, China, in
identification in two-layer complex dynamical networks,” IEEE Trans. 2013. In 2017, she was awarded the 14th Chinese Young Women Scientists
Netw. Sci. Eng., vol. 7, no. 1, pp. 538–548, Jan.–Mar. 2020. Fellowship and the Natural Science Fund for Distinguished Young Scholars
[31] Y. Zheng, X. Wu, G. He, and W. Wang, “Topology identification of of Hubei Province. She has been serving as an Associate Editor for IEEE
fractional-order complex dynamical networks based on auxiliary-system T RANSACTIONS ON C IRCUITS AND S YSTEMS II: E XPRESS B RIEFS since
approach,” Chaos, Interdiscip. J. Nonlin. Sci., vol. 31, no. 4, 2021, 2018.
Art. no. 43125.
[32] Y. Zheng, X. Wu, Z. Fan, and W. Wang, “Identifying topol-
ogy and system parameters of fractional-order complex dynamical
networks,” Appl. Math. Comput., vol. 414, Feb. 2022, Art. no. 126666. Ziye Fan received the B.Sc. degree in applied
[33] H. Liu, Y. Li, Z. Li, J. Lü, and J.-A. Lu, “Topology identification of mul- mathematics from Wuhan University, Wuhan, China,
tilink complex dynamical networks via adaptive observers incorporating in 2021, where she is currently pursuing the Ph.D.
chaotic exosignals,” IEEE Trans. Cybern., vol. 52, no. 7, pp. 6255–6268, degree in applied mathematics with the School of
Jul. 2022. Mathematics and Statistics.
[34] X. Wu, X. Zhao, J. Lü, L. Tang, and J.-A. Lu, “Identifying topologies Her current research interests include complex
of complex dynamical networks with stochastic perturbations,” IEEE networks and nonlinear systems.
Trans. Control Netw. Syst., vol. 3, no. 4, pp. 379–389, Dec. 2016.
[35] X. Wei, S. Chen, J.-A. Lu, and D. Ning, “Reconstruction of complex
networks with delays and noise perturbation based on generalized outer
synchronization,” J. Phys. A, Math. Theor., vol. 49, no. 22, 2016,
Art. no. 225101.
[36] S. Zhu, J. Zhou, G. Chen, and J.-A. Lu, “A new method for topology Kebin Chen was born in Turpan, Xinjiang, China,
identification of complex dynamical networks,” IEEE Trans. Cybern., in 1987. He received the B.S. and M.S. degrees from
vol. 51, no. 4, pp. 2224–2231, Apr. 2021. Xi’an Jiaotong University, Xi’an, China, in 2011 and
[37] Z. Fan and X. Wu, “Identifying partial topology of simplicial com- 2013, respectively, and the Ph.D. degree from the
plexes,” Chaos, Interdiscip. J. Nonlin. Sci., vol. 32, no. 11, 2022, National University of Defense Technology, Wuhan,
Art. no. 113128. China, in 2022.
[38] K. Li and C. Shi, “Synchronization-based topology identification of He is currently a Lecturer with the National
multilink hypergraphs: A verifiable linear independence,” Nonlin. Dyn., University of Defense Technology. His research
vol. 112, no. 4, pp. 2781–2794, 2024. interests include combat system-of-systems,
[39] L. Tang, J.-A. Lu, X. Wu, and J. Lü, “Impact of node dynamics parame- complex networks and systems engineering.
ters on topology identification of complex dynamical networks,” Nonlin.
Dyn., vol. 73, no. 1, pp. 1081–1097, 2013.
[40] G. Tao, “A simple alternative to the barbalat lemma,” IEEE Trans.
Autom. Control, vol. 42, no. 5, p. 698, May 1997.
Jinhu Lü (Fellow, IEEE) received the Ph.D.
degree in applied mathematics from the Academy
of Mathematics and Systems Science, Chinese
Academy of Sciences, Beijing, China, in 2002.
He was a Professor with RMIT University,
Melbourne, VIC, Australia, and a Visiting Fellow
with Princeton University, Princeton, NJ, USA. He
is currently a Vice President of Scientific Research
with Beihang University, Beijing. He is/was the
Yi Zheng received the B.Sc. degree in mathemat- Chief Scientist of the National Key Research and
ics and applied mathematics and the M.Sc. degree Development Program of China and the Leading
in quantitative ecology from the Department of Scientist of the Innovative Research Groups, National Natural Science
Mathematics, Shihezi University, Shihezi, China, Foundation of China. His current research interests include cooperation
in 2013 and 2018, respectively, and the Ph.D. control, industrial Internet, complex networks, and big data.
degree in applied mathematics from the School Prof. Lü was a recipient of the prestigious Ho Leung Ho Lee Foundation
of Mathematics and Statistics, Wuhan University, Award in 2015, the National Innovation Competition Award in 2020, the
Wuhan, China, in 2022. State Natural Science Award three times from the Chinese Government
From October 2022 to May 2024, he served as in 2008, 2012, and 2016, respectively, the Australian Research Council
a Lecturer with the College of Information and Future Fellowships Award in 2009, the National Natural Science Fund for
Communication, National University of Defense Distinguished Young Scholars Award, and the Highly Cited Researcher Award
Technology, Wuhan. Since July 2024, he serves as a Postdoctoral Researcher in engineering from 2014 to 2020, and 2022 to 2024. He was the General
with the College of Computer Science and Software Engineering, Shenzhen Co-Chair of the 43rd Annual Conference of the IEEE Industrial Electronics
University, Shenzhen, China. He received the China National Scholarship for Society in 2017. He is/was an Editor in various ranks for 16 SCI journals,
graduate students in 2021. His current research interests include complex including the Co-Editor-in-Chief of IEEE T RANSACTIONS ON I NDUSTRIAL
networks and nonlinear systems. I NFORMATICS. He is the Fellow of CAA/ORSC/CICC/CSIAM.

Authorized licensed use limited to: SHENZHEN UNIVERSITY. Downloaded on April 25,2025 at 03:09:16 UTC from IEEE Xplore. Restrictions apply.

View publication stats