IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 47, NO.

7, JULY 2017 1535

Finite-Time Consensus of Switched

Multiagent Systems
Xue Lin and Yuanshi Zheng

Abstract—This paper focuses on the finite-time consensus Consensus as a basic problem, aims to design appropriate
(FTC) problem of switched multiagent system (MAS) which protocol to guarantee agents to achieve agreement on certain
is composed of continuous-time and discrete-time subsystems. quantity. From the viewpoint of dynamical behaviors of agents,
Different from the existing results, each agent of this system is
controlled by switching control method. To achieve consensus in the research results of MASs consist of single-integrator
finite time for the switched MAS, two types of consensus protocols dynamics [17], double-integrator dynamics [18]–[21], nonlin-
(the FTC protocol and the fixed-time consensus (Fd TC) proto- ear dynamics [22], [23], etc. On the analysis of consensus
col) are proposed. By using algebraic graph theory, Lyapunov problems, the convergence rate plays a significant role, which
theory and matrix theory, it is proved that the FTC problem reflects the performance of consensus protocol. In order
in strongly connected network and leader-following network can
be solved, respectively. When the initial states of agents are not to accelerate the convergence speed, researchers established
available, the Fd TC protocol is applied to solve the FTC prob- and proposed some methods [24]. However, these conclu-
lem. Simulations are provided to illustrate the effectiveness of sions focus on asymptotic convergence, namely, agents cannot
our theoretical results. achieve consensus in finite time. Actually, requiring an arbi-
Index Terms—Consensus, finite-time, fixed-time, switched trary long time to reach consensus is often unacceptable in
multiagent systems (MASs). some practical situations. Therefore, some researchers were
attracted to explore the finite-time consensus (FTC) prob-
I. I NTRODUCTION
lem. For continuous-time MASs, a variety of FTC protocols
VER the past few decades, various control prob-
O lems for different systems have been investigated, for
instance, neutral-type neural networks [1], [2], nonlinear sys-
have been presented [25]–[28]. In [29], two protocols were
established to solve the FTC problem under time-invariant
undirected topology. By employing finite-time semistability
tems [3]–[6], Markovian jump systems [7]. In recent years, the theory, the finite-time rendezvous problem was developed
distributed coordination control of multiagent systems (MASs) in [30]. Whereas the protocols mentioned in [29] and [30]
has attracted much attention in control field. This is mainly involve discontinuous dynamics, which may lead to a variety
due to its superiority in engineering field, such as the flexi- of negative effects [25]. Therefore, continuous-time finite-time
bility of controller design, strong robustness, and so on. Take protocols were presented. Wang and Xiao [25] proposed a
large scale systems as an example, centralized controller is continuous-time protocol to solve the FTC problem under
not easy to achieve the goal because of the system’s complex- both the bidirectional and unidirectional interaction cases. The
ity. However, large scale systems can be divided into multiple FTC problem of MASs with respect to a monotonic function
subsystems, then distributed control is applied for the system. and heterogeneous MASs were investigated in [31] and [32],
For solving different problems motivated by various objec- respectively. In [33], FTC problem of second-order MASs
tives, lots of research results have been provided, for instance, without velocity measurements was investigated under undi-
consensus problem [8], [9], topology selection [10], con- rected connected graph. Liu et al. [34] proved that FTC of
trollability [11]–[14], optimal control [15], event-time driven MASs with a switching protocol can be achieved under a
control [16], etc. strongly connected and detail-balanced topology. It is note-
Manuscript received July 1, 2016; revised September 10, 2016 and worthy that the estimated bound of convergence time under
October 10, 2016; accepted November 14, 2016. Date of publication the aforementioned FTC protocols is associate with initial
December 8, 2016; date of current version June 22, 2017. This work was states of agents, which may limit some applications. The
supported in part by the National Natural Science Foundation of China
under Grant 61375120 and Grant 61304160, in part by the Fundamental fixed-time consensus (Fd TC) protocol was presented so as
Research Funds for the Central Universities under Grant JB160419, and to deal with this problem. For solving the Fd TC problem,
in part by the Shaanxi Provincial Natural Science Foundation of China Parsegov et al. [35] proposed an Fd TC protocol and obtained
under Grant 2014JQ-5029. This paper was recommended by Associate Editor
S. Tong. (Corresponding author: Yuanshi Zheng.) that the estimated bound of settling time is independent of
The authors are with the Key Laboratory of Electronic Equipment the initial states of agents. Leader–follower Fd TC of MAS
Structure Design of Ministry of Education, School of Mechano-electronic was investigated under undirected connected network in [36].
Engineering, Xidian University, Xi’an 710071, China, and also with the
Center for Complex Systems, School of Mechano-electronic Engineering, Fu and Wang [37] further considered the tracking problem
Xidian University, Xi’an 710071, China (e-mail: xuelinxd@yeah.net; for second-order MASs with bounded input uncertainties and
zhengyuanshi2005@163.com). proved that the fixed-time tracking problem can be solved
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. under interaction network among the followers is undirected
Digital Object Identifier 10.1109/TSMC.2016.2631659 connected. In addition, there exist some other research topics
2168-2216 c 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Indian Institute Of Technology (IIT) Mandi. Downloaded on August 01,2024 at 05:31:37 UTC from IEEE Xplore. Restrictions apply.
1536 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 47, NO. 7, JULY 2017

concerning finite-time problems, such as finite-time forma- Notation: Let R be the set of real numbers, Rn denotes the
tion control [27], finite-time containment control [28], and so n-dimensional real vector space and In be identity matrix. 1 =
on. For discrete-time MASs, the finite-time distributed averag- [1, . . . , 1]T ∈ Rn , ω = [ω1 , ω2 , . . . , ωn ]T ∈ Rn and ω > 0,
ing algorithm with fully-connected topology was first briefly In = {1, 2, . . . , n}, and d̄ = maxi∈In {dii } denotes maximum
discussed in [38]. By using a matrix factorization approach, degree of agent. AT denotes the transpose of matrix A or vector
the finite-time average consensus was solved under distance A. Rn×n and Cn×n denote n × n real matrix and n × n complex
regular graph in [39]. matrix, respectively. B = [bij ] ∈ Rn×n , B ≥ 0 if all bij ≥
A switched system is composed of a finite number of 0. We say that B is a non-negative matrix if B ≥ 0. Both
subsystems, and a switching law managing the switching W ∈ Rn×n and W̄ ∈ R(n−1)×(n−1) are the diagonal matrices
among these subsystems [40]–[43]. For MASs, when switch- with wi > 0 as the (i, i) entry. b̄ = [b1 , b2 , . . . , bn−1 ]T =
ing behaviors occur in the topologies, it can be found that [a1n , a2n , . . . , a(n−1)n ]T ≥ 0 where there is at least one bi >
the system matrix changes with topology switches [17], [44]. 0, and b̃ = diag(b̄) = diag{b1 , b2 , . . . , bn−1 }. J{a1 , . . . , an }
Therefore, the MAS with switching topologies is regarded as represents Jordan matrix with diagonal entry ai and sign(·) is
a kind of switched system which consists of only continuous- the sign function. | · | represents the absolute value. x is the
time or discrete-time subsystems. Zheng and Wang [45] largest integer not greater than x. x̄k (t) = maxi∈V xi (t), xk (t) =
considered the switched MAS consisting of continuous-time mini∈V xi (t) and α0 = maxij αij . span(1) = {ξ ∈ Rn : ξ =
and discrete-time subsystems, and investigated the consensus r1, r ∈ R}.
problem for different topologies. Since realistic communica-
tion among agents may change, the consensus problem of II. P RELIMINARIES AND P ROBLEM F ORMULATION
the switched MAS under random networks was also solved A. Preliminaries
in [46]. Moreover, containment control of such switched
MAS was considered in [47]. In reality, such switched MAS The communication relationship between agents is
exists widely in applications. For instance, an MAS is con- described by graph G (A) = (V, E, A) with vertex set
trolled either by a physically implemented regulator or by V = {ν1 , ν2 , . . . , νn }, edge set E = {eij } ⊆ V × V and
a digitally implemented one with a switching rule between non-negative matrix A = [aij ]n×n . If (νj , νi ) ∈ E, agents i and
them synchronously, i.e., the switched MAS is composed of j are adjacent and aij > 0. The  degree matrix D = [dij ]n×n
continuous-time and discrete-time subsystems. However, so is a diagonal matrix with dii = j∈Ni aij and the Laplacian
far, results concerning FTC were concerned with the MAS matrix of the graph is defined as L = [lij ]n×n = D − A. For
consisting of only continuous-time subsystem or discrete-time Laplacian matrix L, λi (L) denotes the ith eigenvalue of L
subsystem. and 0 = λ1 (L) ≤ λ2 (L) ≤ . . . ≤ λn (L). The directed graph
This paper aims at designing appropriate consensus proto- G (A) is said to satisfy the detail-balanced condition if there
cols to solve the FTC problem of the switched MAS proposed exist some scalars ωi > 0 such that ωi aij = ωj aji for all
in [45]. Zheng and Wang [45] studied the asymptotic consen- i, j ∈ In . For more details, please refer to [32]. For MASs,
sus problem of switched MAS. Different from [45], we mainly agent n is a leader, i.e., an1 = an2 = · · · =  an(n−1) = 0 and 
LFF − b̄
study the FTC problem, namely, the consensus of agents can ā = [a1n , a2n , . . . , a(n−1)n ] ≥ 0. L =
T
be achieved in finite time. The main problem we need to 01×(n−1) 01×1
solve is how to design an effective consensus protocol for the where LFF ∈ R(n−1)×(n−1) , and b̄ ∈ R(n−1)×1 . Followers’
switched MAS to solve the FTC problem. Difficulty comes interaction subgraph is G (Ā).
partly from how to cope with the different dynamical behav-
iors and switching behaviors. It is difficult to apply the idea B. Problem Formulation
of FTC of continuous-time MASs (discrete-time MASs) to We consider an MAS which consists of n agents, and each
the present case. Motivated by Zheng and Wang [45], two agent is controlled by switching control method (continuous-
types of distributed consensus protocols are designed for the time control and sampled-data control). Thus, the agent takes
switched MAS in this paper. The main contribution of this the switched dynamics, it switches between continuous-time
paper includes the following. dynamics and discrete-time dynamics.
1) An FTC protocol is designed for the switched MAS. The continuous-time dynamics is
By using this consensus protocol, we show that if the
ẋi (t) = ui (t), i ∈ In (1a)
sum of time intervals T0 , over which the continuous-time
subsystem is activated, is larger than a finite constant, and discrete-time dynamics is
the FTC problem can be solved under strongly connected
xi (t + 1) = xi (t) + hui (t), i ∈ In (1b)
network.
2) When the initial states of agents are not available, we where xi (t) ∈ R and ui (t) ∈ R are the state and control input of
propose an Fd TC protocol to solve the Fd TC prob- agent i, respectively. Sampling period h > 0 and initial value
lem. We obtain that the sum of time intervals T0 is x0 = [x1 (0), . . . , xn (0)]T . Assume that the synchronous switch
uncorrelated with the initial states of agents. is applied for each agent.
3) By using these two protocols, the FTC problem and the Throughout this paper, there is a sequence of time instants
Fd TC problem under leader-following network are also 0 ≤ t1 < t̄1 < t2 < t̄2 < · · · < t̄k−1 < tk < t̄k < · · · for
investigated, respectively. system (1) that satisfies the following assumption.

Authorized licensed use limited to: Indian Institute Of Technology (IIT) Mandi. Downloaded on August 01,2024 at 05:31:37 UTC from IEEE Xplore. Restrictions apply.
LIN AND ZHENG: FTC OF SWITCHED MASs 1537

Assumption 1: are located in the following region:

1) Continuous-time subsystem (1a) is activated when t ∈ ⎧ ⎫
⎪
⎪ ⎪
⎪
(tk , t̄k ].  n ⎨ n
  ⎬
2) Discrete-time subsystem (1b) is activated when t ∈ z ∈ C : |z − (ω̃ii − ωi )| ≤  
ω̃ij = |ω̃ii − ωi |
⎪ ⎪
(t̄k−1 , tk ]. i=1⎪⎩ j=1 ⎪
⎭
3) t̄k − tk ≥ τ ∗ where τ ∗ is constant. i=j

Note that Assumption 1 guarantees system (1) to be it is easy to see that all the eigenvalues λi ≤ 0.
always composed of only the continuous-time subsystem (or Next, we prove that zero eigenvalue is algebraically simple.
continuous-time and discrete-time subsystems). There does Let λ1 (L), . . . , λn (L) denote the eigenvalues of L, we have
not exist the case that system (1) is composed of only the P−1 LP = J{λ1 (L), . . . , λn (L)}. Then, we have P−1 (−2hL +
discrete-time subsystem. Moreover, system (1) can be viewed h2 L2 )P = J{−2hλ1 (L) + h2 λ21 (L), . . . , −2hλn (L) + h2 λ2n (L)}.
as a system consisting of continuous-time subsystem and From Gersgorin Disk theorem, it is easy to obtain 0 < h <
discrete-time subsystem at different time interval. (1/d̄) ≤ (2/λn (L)). Since G (A) is strongly connected, we
Definition 1 [32]: System (1) reaches FTC if for any initial obtain rank(−2hL + h2 L2 ) = n − 1. By Sylvester inequal-
state x0 ∈ Rn and i, j ∈ In , there is a settling time T such that ity [48], rank((I − hL)T W(I − hL) − W) = rank(W(−2hL +
limt→T (xj (t) − xi (t)) = 0 and xj (t) = xi (t), ∀t ≥ T. h2 L2 )) = n − 1. Hence, λi < 0, i = 2, . . . , n and λ1 = 0 is
Lemma 1 [48]: A complex matrix B ∈ Cn×n is nonsingular algebraically simple.
if B is irreducibly diagonally dominant. Lemma 5: Suppose that followers’ interaction subgraph
Lemma 2 [25]: Let π1 , π2 , . . . , πn ≥ 0 and 0 < κ ≤ 1, G (Ā) is strongly connected and satisfies the detailed balance
then condition, and 0 < h < (1/d̄ + bi ), i ∈ In−1 . Then, for all
 n κ i ∈ In−1 , it holds that λ̄i < 0 where λ̄i is the ith eigenvalue
n 
κ of matrix (I − hLFF )T W̄(I − hLFF ) − W̄.
πi ≥ πi .
i=1 i=1 Proof: Since 0 < h < (1/d̄ + bi ) and matrix Ā is a non-
negative matrix, we have that
Lemma 3 (Comparison Theorem [49]): Let U(t, u) be con-      
tinuous on an open (t, u)-set E and u = u0 (t) the maximal I − hLFF = I − h D̄ + b̃ − Ā = I − h D̄ + b̃ + hĀ
solution of (du/dt) = U(t, u), u(t0 ) = u0 . Let v(t) be a
continuous function on [t0 , t0 + a] satisfying the conditions is a non-negative matrix. Similar to Lemma 4, we have W̄ Ā =
v(t0 ) ≤ u0 , (t, v(t)) ∈ E, and v(t) has a right derivative ĀT W̄ and W̄LFF = LFF T W̄. Due to L 1 = b̄, we have (I −
FF
DR (v(t)) on t0 ≤ t < t0 + a such that DR (v(t)) ≤ U(t, v(t)). hLFF ) W̄(I − hLFF )1 = W̄(1 − 2hb̄ + h2 l̄), l̄ = [l11 b1 + l12 b2 +
T
Then, on a common interval of existence of u0 (t) and v(t), · · ·+l1n−1 bn−1 , . . . , ln−11 b1 +ln−22 b2 +· · ·+ln−1n−1 bn−1 ]T and
v(t) ≤ u0 (t). lji ≤ 0, j = 1, . . . , n−1, j = i. Owing to 0 < h < [1/(d̄ + bi )],
Lemma 4: Suppose that graph G (A) is strongly connected we have (I − hLFF )T W̄(I − hLFF )1 ≤ W̄1. By Gersgorin Disk
and satisfies the detailed balance condition, and 0 < h < theorem [48], all eigenvalues of (I − hLFF )T W̄(I − hLFF ) − W̄
(1/d̄). Then, for all i ∈ In , it holds that λi ≤ 0 and λi = 0 is are nonpositive.
algebraically simple, where λi is the ith eigenvalue of matrix Because graph G (Ā) is strongly connected and there is at
(I − hL)T W(I − hL) − W. least one agent connected  to the leader, there is at least one
Proof: From the definition of L and 0 < h < (1/d̄), we lii such that |lii | > j=i |lij |, i, j ∈ In−1 . By Lemma 1, it
have that follows that LFF is nonsingular, i.e., there is no zero eigenvalue
for matrix LFF . By Gersgorin Disk theorem, we have 0 <
I − hL = I − h(D − A) = (I − hD) + hA
h < [1/(d̄ + bi )] ≤ (2/λn−1 (LFF )). According to Lemma 4,
is a non-negative matrix and its row sums are 1. Hence, we can get rank(−2hLFF + h2 LFF 2 ) = n − 1. Therefore, (I −

matrix (I −hL)T W(I −hL) is also non-negative. Because graph hLFF ) W̄(I − hLFF ) − W̄ = W̄(−2hLFF + h2 LFF
T 2 ) is full rank.

G (A) satisfies the detailed balance condition, there exist some Consequently, there is no zero eigenvalue and λ̄i < 0, i =
scalars ωi > 0 such that ωi aij = ωj aji for all i, j ∈ In . 1, . . . , n − 1.
Thus, we get WA = AT W and WL = LT W. It follows from The asymptotic consensus problem for switched MAS has
WL = LT W and L1 = 0 that (I − hL)T W(I − hL)1 = been investigated in [45]. In this paper, two types of consen-
W(I − hL)1 = ω. Consequently, there must exist a zero sus protocols which guarantee the switched MAS to achieve
eigenvalue corresponding to a right eigenvector 1 for matrix consensus in finite time are designed.
(I − hL)T W(I − hL) − W.
Let (I − hL)T W(I − hL) = W̃, then III. FTC P ROTOCOL FOR THE S WITCHED MAS
⎡ ⎤ This section proposed FTC protocol for system (1) to solve
ω̃11 − ω1 . . . ω̃1n
⎢ .. .. .. ⎥ FTC problem. For continuous-time and discrete-time subsys-
W̃ − W = ⎣ . . . ⎦ tems, we design different consensus protocols, respectively. It
ω̃n1 ...
ω̃nn − ωn is presented as follows:
n  n  α  
it is obvious that ω̃ii − ωi ≤ 0 and j=1 ω̃ij − ωi = 0. By aij sig xj (t) − xi(t) ij , t ∈ tk ,t̄k
ui (t) = j=1n  (2)
Gersgorin Disk theorem [48], all the eigenvalues of W̃ − W j=1 aij xj (t) − xi (t) , t ∈ t̄k−1 , tk

Authorized licensed use limited to: Indian Institute Of Technology (IIT) Mandi. Downloaded on August 01,2024 at 05:31:37 UTC from IEEE Xplore. Restrictions apply.
1538 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 47, NO. 7, JULY 2017

where 0 < αij < 1 and sig(μ)αij = sign(μ)|μ|αij . Note that and
sig(μ)αij is a continuous function. When αij = 1, it will 
n
1
become the consensus protocol proposed in [45] which solves υ(t + 1) = n ωi xi (t + 1)
the asymptotical consensus problem. In this paper, we assume i=1 ωi i=1
⎛ ⎞
that αij = αji . n 
n
 
1
Next, we will show that system (1) with protocol (2) = n ωi ⎝xi (t) + h aij xj (t) − xi (t) ⎠
achieves FTC under certain mathematical conditions. To solve i=1 ωi i=1 j=1
this problem, we introduce U = {ξ ∈ Rn : ωT ξ = 0 and 1 n
ξ = 1}. Since ω > 0, it is clear that ξ = 0 and ξ ∈ / span(1). = n ωi xi (t)
L is Laplacian matrix of undirected connected graph, then i=1 ωi i=1

ξ T Lξ > 0. 1 
n 
n
 
Theorem 1: Suppose that the interaction network G (A) is + h n ωi aij xj (t) − xi (t)
i=1 ωi i=1 j=1
strongly connected and satisfies the detailed balance condition,
and 0 < h < (1/d̄). Then, system (1) with protocol (2) reaches = υ(t).
FTC if there exist finite k and t̄∗ ∈ [tk , t̄k ) such that Therefore, υ(t) is time-invariant. Let γi (t) = xi (t) − υ(t),
4 1−α0 we have ωT γ (t) = 0 and γ (t + 1) = (I − hL)x(t) − 1υ(t) =
T0 ≥ V 2 (γ (0)) (3) (I − hL)(x(t) − 1υ(t)) = (I − hL)γ (t). Take Lyapunov
(1 − α0 )K
function V(γ (t)) = (1/2) ni=1 ωi γi2 (t) = (1/2)γ T (t)Wγ (t)
where for continuous-time subsystem (1a) and discrete-time subsys-
tem (1b).

k−1
 
T0 = t̄∗ − tk + t̄i − ti When t ∈ (t̄k−1 , tk ], discrete-time subsystem (1b) is acti-
i=1
vated. Thus, we have
⎛ ⎞2
V(γ (t + 1)) − V(γ (t))
1 ⎝ 
n n
1
V(γ (0)) = ωi xi (0) − n ωj xj (0)⎠ 1 
2 j=1 ωj
= γ T (t + 1)Wγ (t + 1) − γ T (t)Wγ (t)
i=1 j=1 2
1  
= γ T (t) (I − hL)T W(I − hL) − W γ (t)
and K = (K1 K2 )[(1+α0 )/2] with 2
1  
1   2 ≤ λmax (I − hL)T W(I − hL) − W γ T (t)γ (t)
K1 = min ωi aij α0 +1 2
n   2 i,j∈I ≤0 (4)
ωi aij α0 +1 a =0n
i,j=1 ij
! "
αij +1 where the last inequality follows from Lemma 4.
 2 α0 +1 −1 When t ∈ (tk , t̄k ], continuous-time subsystem (1a) is acti-
× x̄k (0) − xk (0) >0
vated. Similar to the proof technique of Wang and Xiao [25],
and we have
4c0 
n n
 α
K2 = , c0 = min ξ T L(B)ξ > 0 V̇(γ (t)) = ωi γi (t) aij sig γj (t) − γi (t) ij
ωmax ξ ∈U
# $ i=1 j=1
U = ξ ∈ R : ω ξ = 0 and ξ = 1
n T
1 
n
 α
  = ωi aij γi (t)sig γj (t) − γi (t) ij
  2 2
B = ωi aij α0 +1 ∈ Rn×n . i,j=1

+ ωj aji γj (t)sig(γi (t) − γj (t))αji
Proof: First, we show that there exist finite k satisfying (3). 1   α +1
n
From Assumption 1, we know that T0 ≥ (k − 1)τ ∗ . Then for =− ωi aij γj (t) − γi (t) ij
2
any finite k ≥ (4V [(1−α0 )/2] (0)/(1 − α0 )Kτ ∗ ) + 1, (3) holds. i,j=1
The interaction network G (A) is strongly connected and sat-   α0 +1
1  
n
 2   2(αij +1) 2
isfies the detailed balance condition, i.e., there exists a positive =− ωi aij α0 +1 γj (t) − γi (t) α0 +1
column vector ω = [ω1 , ω2 , . . . , ωn ]T such that ωi aij = ωj aji 2
i,j=1
for all i, j ∈ In .  ⎛
 n   2   2(αij +1)
Let υ(t) = (1/ ni=1 ωi ) ni=1 ωi xi (t). Due to αij = αji , 1⎜ ωi aij α0 +1 γj (t) − γi (t) α0 +1
≤− ⎜
i,j=1
ωi aij = ωj aji for all i, j ∈ In , we have 2⎝ 2γ T (t)L(B)γ (t)
1 
n
υ̇(t) = n ωi ẋi (t) ⎞ α0 +1
i=1 ωi i=1
2
2γ T (t)L(B)γ (t) ⎟
⎟
1 n 
n
 α × V(γ (t)) ⎠ (5)
= n ωi aij sig xj (t) − xi (t) ij V(γ (t))
i=1 ωi i=1 j=1
=0 where the last inequality follows from Lemma 2.

Authorized licensed use limited to: Indian Institute Of Technology (IIT) Mandi. Downloaded on August 01,2024 at 05:31:37 UTC from IEEE Xplore. Restrictions apply.
LIN AND ZHENG: FTC OF SWITCHED MASs 1539

We know that x̄k (t) = maxi∈V xi (t) and xk (t) = mini∈V xi (t). speed of system (1) is influenced by λ2 (L(B)) where B =
When t ∈ (t̄k−1 , tk ], by Lemma 4, matrix I − hL = [l̄ij ]n×n is [(aij )[2/(α0 +1)] ] ∈ Rn×n . Furthermore, the larger algebraic
a row stochastic matrix, i.e., nj=1 l̄ij = 1. Obviously, x̄k (t + connectivity of G (B) is, the shorter the convergence time is.
 
1) = maxi∈V nj=1 l̄ij xj (t) ≤ maxi∈V nj=1 l̄ij x̄j (t) = x̄k (t), i.e., Theorem 2: Suppose that system (1) has a leader (labeled
x̄k (t + 1) ≤ x̄k (t). We show that xk (t + 1) ≥ xk (t) in precisely as n) and n − 1 followers (labeled as 1, . . . , n − 1), and
the same way. When t ∈ (tk , t̄k ], it is obvious that x̄˙ k (t) ≤ 0 the interaction network G (Ā) among the followers is strongly
and ẋk (t) ≥ 0. Thus, x̄k (t) − xk (t) is a nonincreasing for all connected and satisfies the detailed balance condition, and
t > 0. Hence, |γj (t) − γi (t)| ≤ x̄k (t) − xk (t) ≤ x̄k (0) − xk (0). 0 < h < [1/(d̄ + bi )], i ∈ In−1 . Then, system (1) with pro-
From [25], we have tocol (2) reaches FTC if there exist finite k and t̄∗ ∈ (tk , t̄k ]
such that
n   2   2(αij +1)
ωi aij α0 +1 γ (t) − γ (t)  α0 +1 1−α0
i,j=1 j i 4
≥ K1 > 0 T0 ≥ V̄ 2 (γ (0)) (8)
2γ T (t)L(B)γ (t) (1 − α0 )K̄
where K1 is defined as in (3). where
Since U = {ξ : ωT ξ = 0 and ξ = 1} and ω > 0, we have
ξ = 0 and ξ ∈ / span{1}. By properties of Laplacian matrix 
k−1
  1
n−1
under undirected connected graph, we have ξ T L(B)ξ > 0. It T0 = t̄∗ − tk + t̄i − ti , V̄(γ (0)) = ωi (xi (0) − xn )2
2
follows from ωT γ (t) = 0 that i=1 i=1

2γ T (t)L(B)γ (t) 4 γ T (t) γ (t) and K̄ = (K̄1 K̄2 )[(1+α0 )/2] with
≥ ' L(B) ' ≥ K2
V(γ (t)) ωmax γ T (t)γ (t) γ T (t)γ (t)
1   2
where L(B) and K2 are defined as in (3). K̄1 = 2
min ωi aij α0 +1
n   2 n i,j∈I
α0 +1 a =0n
Hence, it is easy to obtain V̇(γ (t)) ≤ ω a
i,j=1 i ij
α0 +1 + 2 b
i=1 i ij
−(1/2)KV [(α0 +1)/2] (γ (t)) for t ∈ (tk , t̄k ]. By the comparison  2
α
ij +1

α0 +1 −1
theorem × x̄k (0) − xk (0) >0
1−α0 1 − α0 1−α0
V 2 (γ (t)) ≤− K(t − tk ) + V 2 (γ (tk )). (6) and
4
 
From (4) and (5), it is easy to know that 4λ1 (L(B̄) + b̃)   2
K̄2 = , B̄ = ωi aij α0 +1 ∈ R(n−1)×(n−1) .
   ωmax
V(γ (0)) ≥ V(γ (t1 )) ≥ V γ t̄1 ≥ V(γ (t2 ))
     
≥ V γ t̄2 ≥ · · · ≥ V(γ (tk )) ≥ V γ t̄k ≥ · · · (7) Proof: The interaction network among the followers is
strongly connected and satisfies the detailed balance con-
Next, we show that there exists t̄∗ ∈ (tk , t̄k ] such that dition, i.e., there exists a positive column vector ω̄ =
−[(1 − α0 )/4]K(t̄∗ − tk ) + V [(1−α0 )/2] (γ (tk )) ≤ 0. Based [ω1 , ω2 , . . . , ωn−1 ]T such that ωi aij = ωj aji for all i, j ∈ In−1 .
on (6), (7), and 1 − α0 > 0 We rewrite protocol (2) as follows:
1 − α0  ∗  1−α0 ⎧ n−1  αij
− K t̄ − tk + V 2 (γ (tk )) ⎪
⎪ a sig x (t) − x (t) + ain sig(xn − xi )αin
4 ⎨ 
j=1 ij  j i
1 − α0  ∗  1−α0    t ∈ t , t̄
≤− K t̄ − tk + V 2 γ t̄k−1 ui (t) = n−1k k  (9)
⎪
⎪ a x (t) − xi (t) + ain (xn − xi )
4 ⎩ j=1  ij j 
1 − α0  ∗  1−α0 t ∈ t̄k−1 , tk
≤− K t̄ − tk + t̄k−1 − tk−1 + V 2 (γ (tk−1 ))
4
.. where
. (
1 − α0  ∗  bi > 0 if agent i is connected to agent n
≤− K t̄ − tk + t̄k−1 − tk−1 + · · · + t̄1 − t1 ain =
4 0 otherwise
1−α0
+V 2 (γ (0)).
denotes whether the follower i is connected to the leader n
Hence, it follows from (3) that there exists t̄∗ ∈ (tk , t̄k ] and i ∈ In−1 .
such that −[(1 − α0 )/4]K(t̄∗ − tk ) + V [(1−α0 )/2] (γ (tk )) ≤ 0. First, we show that there exist finite k satisfying (8). The
By virtue of V(γ (t)) ≥ 0, V̇(γ (t)) ≤ 0 and (6), we get proof is similar to that of theorem 1 and it is omitted.
V(γ (t)) = 0 for t ≥ t̄∗ . Thus system (1) reaches consensus in Let γi (t) = xi (t) − xn , i ∈ In−1 , we have γ̄ (t + 1) =
finite time. x(t + 1) − 1xn = (I − hLFF )x(t) + (hb̄ − 1)xn = (I − hLFF )γ̄ (t),
Remark 1: If the interaction network is a undirected con- i ∈ In−1 , where L FF 1 = b̄. Take Lyapunov function
nected network, the analogous theoretical result can be estab- V̄(γ (t)) = (1/2) n−1 i=1 ωi γi (t)
2 = (1/2)γ̄ T (t)W̄ γ̄ (t)
lished for system (1) with protocol (2) to solve the finite-time for continuous-time subsystem (1a) and discrete-time
average consensus problem. It is clear that the convergence subsystem (1b).

Authorized licensed use limited to: Indian Institute Of Technology (IIT) Mandi. Downloaded on August 01,2024 at 05:31:37 UTC from IEEE Xplore. Restrictions apply.
1540 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 47, NO. 7, JULY 2017

When t ∈ (t̄k−1 , tk ], discrete-time subsystem (1b) is acti- IV. Fd TC P ROTOCOL FOR THE S WITCHED MAS
vated. Thus, we have In this section, the Fd TC protocol is proposed for system (1).
V̄(γ (t + 1)) − V̄(γ (t)) It is presented as follows:
1  ⎧
= γ̄ T (t + 1)W̄ γ̄ (t + 1) − γ̄ T (t)W̄ γ̄ (t) ⎪   m   p
2 ⎨ nj=1 aij xj (t) − xi (t) r + nj=1 aij xj (t) − xi (t) q
 
1   ui (t) = t ∈ tk , t̄k
= γ̄ T (t) (I − hLFF )T W̄(I − hLFF ) − W̄ γ̄ (t) ⎪ n
⎩    
2 j=1 aij xj (t) − xi (t) , t ∈ t̄k−1 , tk
1   (12)
≤ λmax (I − hLFF )T W̄(I − hLFF ) − W̄ γ̄ T (t)γ̄ (t)
2
≤0 (10) where m, r, p, and q are positive odd integers such that m > r
where the last inequality follows from Lemma 5. and p < q. When (m/r) = 1 and (p/q) = 1, it will become
When t ∈ (tk , t̄k ], continuous-time subsystem (1a) is acti- the consensus protocol proposed in [45] which solves the
vated. Therefore asymptotical consensus problem.
Note that f1 (x) = x(m/r) and f2 (x) = x(p/q) are odd func-
˙ (t))
V̄(γ tions. It is quite clear that function f (x) = f1 (x)+f2 (x) is a odd
⎛ ⎞

n−1 n−1 function. Moreover, it is easy to obtain that limx→0+ f (x) =
 αij
= ωi γi (t)⎝ aij sig γj (t) − γi (t) − bi sig(γi (t))αin ⎠ limx→0− f (x) = 0, i.e., f (x) is continuous function. Therefore,
i=1 j=1 the consensus protocol for continuous-time subsystem is con-
tinuous.
1 
n−1
 α +1 n−1
=− ωi aij γj (t) − γi (t) ij − ωi bi |γi (t)|αin +1 Theorem 3: Suppose that interaction network G (A) is
2 strongly connected and satisfies the detailed balance condi-
i,j=1 i=1
! " α0 +1 tion, and 0 < h < (1/d̄). Then, system (1) with protocol (12)
1 1 2 2 reaches FTC if there exist finite k and t̄∗ ∈ (tk , t̄k ] such that
≤− V̄(γ (t)) (11)
2 2 V̄(γ (t))
1 1
where T0 ≥ + (13)
K̃1 (k1 − 1) K̃2 (1 − k2 )
n   2   2(αij +1)
ω a α0 +1 γ (t) − γ (t) α0 +1 where
1 i,j=1 i ij j i
=    
2 2γ̄ (t) L B̄ + b̃ γ̄ (t)
T

k−1
 
∗
T0 = t̄ − tk + t̄i − ti
2(αij +1)
n−1 2
(ω b ) α0 +1 |γ (t)| α0 +1 i=1
i i i
+ i=1     . ! " m+r
r−m c1 2r
2γ̄ T (t) L B̄ + b̃ γ̄ (t) K̃1 = n 2r
ωmax
p! " p+q
By using the proof method in Theorem 1, it is easy to obtain
( c2 2q
1 / 2 ) ≥ K̄1 > 0. K̃2 = 2 q
From [25], we know that L(B̄) + b̃ is a positive definite ωmax
matrix and λ1 (L(B̄) + b̃) > 0. Thus m+r
k1 = >1
2r
2 2γ̄ T (t)(L(B̄) + b̃)γ̄ (t)
=
V̄(γ (t)) and k2 = [(p + q)/2q] < 1. c1 = minξ ∈U ξ T L(B̃1 )ξ > 0
2 γ̄ (t)W̄ γ̄ (t)
1 T
and c2 = minξ ∈U ξ T L(B̃2 )ξ > 0, U = {ξ ∈ Rn : ωT ξ =
λ1 (L(B̄) + b̃)
≥4 = K̄2 > 0. 0 and ξ = 1}, B̃1 = [(ωi aij )[2r(m+r)] ] ∈ Rn×n and B̃2 =
ωmax [(ωi aij )[2q/(p+q)] ] ∈ Rn×n .
Similar to Theorem 1, we get V̄(γ (t)) = 0 for t ≥ t̄∗ . Proof: First, we show that there exist finite k satisfying (13).
Therefore, system (1) reaches consensus in finite time. By Assumption 1, we get T0 ≥ (k − 1)τ ∗ . It is clear that con-
Remark 2: It is noteworthy that the time interval τ ∗ is not dition (13) is satisfied for any finite k ≥ (1/τ ∗ K̃1 (k1 − 1)) +
arbitrarily small if we want to guarantee the FTC of system (1). (1/τ ∗ K̃2 (k2 − 1)) + 1. Moreover, there must exist a tk+∗ ∈
From the proof of Theorems 1 and 2, we can find that sys- (tk , t̄k ] such that tk+∗ − tk∗ + · · · + t̄1 − t1 ≥ (1/K̃1 (k1 − 1)) and
tem (1) can reach consensus asymptotically if τ ∗ is arbitrarily t̄∗ − tk + · · · + t̄k∗ − tk+∗ ≥ (1/K̃2 (1 − k2 )).
small. The interaction network is strongly connected and satisfies
Note that T0 is dependent on the initial states of agents the detailed balance condition, i.e., there exists a positive col-
when protocol (2) is applied. The initial states of agents must umn vector ω = [ω1 , ω2 , . . . , ωn ]T such
 that ωi aij = ωj aji for
be available if we want to get T0 . However, there exists the all i, j ∈ In . Let υ(t) = (1/ ni=1 ωi ) ni=1 ωi xi (t) and γi (t) =
case that the initial states of agents are unavailable in some xi (t) − υ(t). Similar to Theorem 1, υ(t) is time-invariant and
applications. To calculate T0 without the information of initial γ (t +  1) = (I − hL)γ (t). Take Lyapunov function V(γ (t)) =
states of agents, we will propose an Fd TC protocol for the (1/2) ni=1 ωi γi2 (t) = (1/2)γ T (t)Wγ (t) for continuous-time
switched MAS to solve this problem in the following part. subsystem (1a) and discrete-time subsystem (1b).

Authorized licensed use limited to: Indian Institute Of Technology (IIT) Mandi. Downloaded on August 01,2024 at 05:31:37 UTC from IEEE Xplore. Restrictions apply.
LIN AND ZHENG: FTC OF SWITCHED MASs 1541

When t ∈ (t̄k−1 , tk ], discrete-time subsystem (1b) is acti- Based on (20) and (21), we get
vated. Similar to Theorem 1, we have  
−K̃2 (1 − k2 ) t̄∗ − tk + V 1−k2 (γ (tk ))
V(γ (t + 1)) ≤ V(γ (t)). (14)     
≤ −K̃2 (1 − k2 ) t̄∗ − tk + V 1−k2 γ t̄k−1
When t ∈ (tk , t̄k ], continuous-time subsystem (1a) is acti-  
≤ −K̃2 (1 − k2 ) t̄∗ − tk + t̄k−1 − tk−1 + V 1−k2 (γ (tk−1 ))
vated. Therefore
..
V̇(γ (t)) .
⎛  

n n ≤ −K̃2 (1 − k2 ) t̄∗ − tk + t̄k−1 − tk−1 + · · · + t̄1 − t1 + 1.
 m
= ωi γi (t)⎝ aij γj (t) − γi (t) r (22)
i=1 j=1
⎞ Hence, it follows from (13) that there exists t̄∗ ∈ (tk , t̄k ] such

n
 p
+ aij γj (t) − γi (t) q ⎠ that −K̃2 (1 − k2 )(t̄∗ − tk ) + V 1−k2 (γ (tk )) ≤ 0.
j=1
When V(γ (0)) ≥ 1, we show that there must exist tk+∗ ∈
(tk , t̄k ] such that V(γ (tk+∗ )) < 1. Based on (19) and (21), we
n !
 " m+r
1   2r  2 2r get
=− ωi aij m+r γj (t) − γi (t)
2
i,j=1 V 1−k1 (γ (t1 )) ≥ V 1−k1 (γ (0))
  p+q     
1  
n
 2q  2 2q V 1−k1 γ t̄1 ≥ −K̃1 (1 − k1 ) t̄1 − t1 + V 1−k1 (γ (t1 ))
− ωi aij p+q γj (t) − γi (t)   
2 V 1−k1 (γ (t2 )) ≥ V 1−k1 γ t̄1
i,j=1     
⎛ ⎞ m+r V 1−k1 γ t̄2 ≥ −K̃1 (1 − k1 ) t̄2 − t2 + V 1−k1 (γ (t2 ))
1 r−m 
n
  2r  2
2r ..
≤ − (2n) 2r ⎝ ωi aij m+r γj (t) − γi (t) ⎠ .
2   +   
i,j=1 V 1−k1
γ tk∗ ≥ −K̃1 (1 − k1 ) tk+∗ − tk + V 1−k1 (γ (tk )).
⎛ ⎞ p+q (23)
2q
1 
n
  2q  2
− ⎝ ωi aij p+q γ
j (t) − γi (t) )⎠ (15) Since there exists tk+∗ − tk∗ + · · · + t̄1 − t1 ≥ (1/K̃1 (k1 − 1)),
2
i,j=1
we easily obtain
where the last inequality follows from Lemma 2 and Holder     
inequality. V 1−k1 γ tk+∗ ≥ −K̃1 (1 − k1 ) tk+∗ − tk + · · · + t̄1 − t1
Similar to Theorem 1, we have + V 1−k1 (γ (0)) > 1
n   2r  
m+r γj (t) − γi (t) 2
i,j=1 ωi aij 4c1 it follows from 1 − k1 < 0 that V(γ (tk+∗ )) < 1. Hence, there
≥ (16)
V(γ (t)) ωmax must exist tk+∗ ∈ (tk , t̄k ] such that V(γ (tk+∗ )) < 1. By virtue
n   2q   of (22) and t̄∗ − tk + · · · + t̄k∗ − tk+∗ ≥ (1/K̃2 (1 − k2 )), it fol-
ω p+q γ (t) − γ (t) 2
i,j=1 i aij j i 4c2 lows that −K̃2 (1 − k2 )(t̄∗ − tk ) + V 1−k2 (γ (tk )) ≤ 0. Owing to
≥ . (17)
V(γ (t)) ωmax V(γ (t)) ≥ 0, V̇(γ (t)) ≤ 0 and (20), we get V(γ (t)) = 0
Using m > r, p < q, (16) and (17) leads to for t ≥ t̄∗ . Therefore, system (1) reaches consensus in
finite time.
p+q
V̇(γ (t)) ≤ −K̃1 V
m+r
2r (γ (t)) − K̃2 V 2q (γ (t)) (18) Theorem 4: Suppose that system (1) has a leader (labeled
as n) and n − 1 followers (labeled as 1, . . . , n − 1), and
where K̃1 = n[(r−m)/2r) ((c1 /ωmax ))[(m+r)/2r] > 0 and K̃2 = the interaction network G (Ā) among the followers is strongly
2(p/q) ((c2 /ωmax ))[(p+q)/2q] > 0. connected and satisfies the detailed balance condition, and
Obviously, for t ∈ (tk , t̄k ], we have V̇(γ (t)) ≤ −K̃1 V k1 (γ (t)) 0 < h < [1/(d̄ + bi )], i ∈ In−1 . Then, system (1) with pro-
and V̇(γ (t)) ≤ −K̃2 V k2 (γ (t)). By the Comparison theorem, tocol (12) reaches FTC if there exist finite k and t̄∗ ∈ (tk , t̄k ]
we have such that
V 1−k1 (γ (t)) ≥ −K̃1 (1 − k1 )(t − tk ) + V 1−k1 (γ (tk )) (19) 1 1
T0 ≥ + (24)
 1 (k1 − 1)
K  2 (1 − k2 )
K
and
V 1−k2 (γ (t)) ≤ −K̃2 (1 − k2 )(t − tk ) + V 1−k2 (γ (tk )). (20) where

We know from (14) and (18) that 

k−1
 
   T0 = t̄∗ − tk + t̄i − ti
V(γ (0)) ≥ V(γ (t1 )) ≥ V γ t̄1 ≥ V(γ (t2 )) i=1
     
≥ V γ t̄2 ≥ · · · ≥ V(γ (tk )) ≥ V γ t̄k ≥ · · · (21) ⎛     ⎞ m+r
r−m λ1  1 + b1
L B
2r
When V(γ (0)) < 1, we show that there must exist t̄∗ ∈ 1 =
K 2(n − 1) 2r ⎝ ⎠ >0
(tk , t̄k ] such that −K̃2 (1 − k2 )(t̄∗ − tk ) + V 1−k2 (γ (tk )) ≤ 0. ωmax

Authorized licensed use limited to: Indian Institute Of Technology (IIT) Mandi. Downloaded on August 01,2024 at 05:31:37 UTC from IEEE Xplore. Restrictions apply.
1542 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 47, NO. 7, JULY 2017

⎛ ⎞ p+q
and 2q
1⎝ 
n−1
  2q  2
  
⎛  ⎞ p+q − ωi aij p+q γ
j (t) − γ i (t) ⎠
 
p λ1 L B2 + b2
2q 2
m+r i,j=1
 2 = −2 q ⎝
K ⎠ > 0, k1 = >1
ωmax 2r  n−1  p+q
 2q 2q

and k2 = [(p + q)/2q] < 1 − (ωi bi ) p+q (γi (t))2 . (27)

  i=1
  2r
 1 = ωi aij m+r ∈ R(n−1)×(n−1)
B By Lemma 2, Holder inequality and [(r − m)/r] < 0, we have
) * ˙ (t))
  2q V̄(γ
 2 = ωi aij p+q ∈ R(n−1)×(n−1) ⎛
B
r−m  
n−1
−m  2r  2
≤ −2 r (n − 1) 2r ⎝ ωi aij m+r γj (t) − γi (t)
  2r   2q i,j=1
m+r p+q
ωi b̃ ωi b̃
b1 = and b2 = .  m+r
2 2 
n−1 2r 2r
+ (ωi bi ) m+r (γi (t))2
Proof: The interaction network among the followers is
i=1
strongly connected and satisfies the detailed balance con- ⎛
dition, i.e., there exists a positive column vector ω̄ = 1⎝ 
n−1
  2q  2
− ωi aij p+q γ
j (t) − γi (t)
[ω1 , ω2 , . . . , ωn−1 ]T such that ωi aij = ωj aji for all i, j ∈ In−1 . 2
i,j=1
We rewrite protocol (12) as follows:
⎧  m  p+q
⎪ n−1
(t) − (t)
m
r + ain (xn − xi ) r 
n−1 2q 2q
⎪
⎪ a ij xj xi + (ωi bi ) p+q (γi (t))2
⎪
⎪
j=1
⎨ + n−1 aij xj (t) − xi (t) q + ain (xn − xi ) q
p p
⎪ i=1
ui (t) = t ∈ t , t̄ 
j=1
(25) m+r p+q
⎪
⎪  k k  ≤  1 V 2r (γ (t)) − K
−K  2 V 2q (γ (t)) (28)
⎪
⎪ n−1
⎪
⎪ a x (t) − xi (t) + ain (xn − xi )
⎩ j=1  ij j  where
t ∈ t̄k−1 , tk
 ⎛  ⎞ m+r
where ain is defined as in Theorem 2 and i ∈ In−1 . λ L( 
B ) + 
b
2r
r−m 1
First, we show that there exist finite k satisfying (24).  1 = 2(n − 1) 2r
K ⎝ ⎠ >0
The proof is similar to that of Theorem 3 and it is omitted. ωmax
Moreover, there must exist tk+∗ ∈ [tk , t̄k ) such that tk+∗ − tk∗ +
 1 (k1 − 1)) and t̄∗ − tk + · · · + t̄k∗ − t+∗ ≥ and
· · · + t̄1 − t1 ≥ (1/K k
(1/K  2 (1 − k2 )). ⎛   ⎞ p+q
p λ1  ) + b
L(B
2q
Let γi (t) = xi (t) − x n , i ∈ In−1 . Take Lyapunov func- 2 =
K −2 q ⎝ ⎠ > 0.
tion V̄(γ (t)) = (1/2) n−1 i=1 ωi γi (t) = (1/2)γ̄ (t)W̄ γ̄ (t)
2 T ωmax
for continuous-time subsystem (1a) and discrete-time subsys-
tem (1b). Similar to Theorem 3, we obtain V̄(γ (t)) = 0 for t ≥ t̄∗ .
When t ∈ (t̄k−1 , tk ], discrete-time subsystem (1b) is acti- Therefore, system (1) reaches consensus in finite time.
vated. Similar to Theorem 2, we have Remark 3: It is clear that T0 in Theorems 3 or 4 is irrel-
evant to the initial states of agents. It is closely related to
V̄(γ (t + 1)) ≤ V̄(γ (t)). (26)
the algebraic connectivity, order n of system (1) and design
When t ∈ (tk , t̄k ], continuous-time subsystem (1a) is acti- parameters m, r, p, and q of protocol (12).
vated. Therefore When 0 < h < (1/λmax (L)) and the interaction network is
˙ (t))
V̄(γ
undirected connected, it can be found that system (1) which
⎛ ⎞ is composed of only discrete-time subsystem (1b) can not
n−1 n−1
 m m reach consensus in finite time. Assume that system x(k + 1) =
= ωi γi (t)⎝ aij γj (t) − γi (t) r − bi sig(γi (t)) r ⎠ (I−hL)x(k) reaches consensus in finite time, i.e., x(k+1) = c1,
i=1 j=1
⎛ ⎞ therefore we have x(k) = (I − hL)−1 x(k + 1). Due to

n−1 
n−1
 p p (I − hL)1 = 1, it is easy to obtain (I − hL)−1 1 = 1, i.e., row
+ ωi γi (t)⎝ aij γj (t) − γi (t) q − bi sig(γi (t)) q ⎠ sums of (I −hL)−1 are 1. Thus, we have x(k) = c1 = x(k +1).
i=1 j=1 By using the similar method, we get x(0) = c1 which leads to
⎛ ⎞ m+r a contradiction. In this paper, by using the FTC protocol, we
2r
r−m  
n−1
1  2r  2 can find that continuous-time subsystem (1a) and discrete-time
≤ − (2(n − 1)) 2r ⎝ ωi aij m+r γj (t) − γi (t) ⎠ subsystem (1b) can solve FTC under proper switching law
2
i,j=1
even if discrete-time subsystem (1b) cannot reach consensus
 n−1  m+r in finite time.
r−m  2r 2r
− (n − 1) 2r (ωi bi ) m+r (γi (t))2 Moreover, in this paper, we can point out that discrete-time
i=1 MASs can achieve consensus in finite time if there exists

Authorized licensed use limited to: Indian Institute Of Technology (IIT) Mandi. Downloaded on August 01,2024 at 05:31:37 UTC from IEEE Xplore. Restrictions apply.
LIN AND ZHENG: FTC OF SWITCHED MASs 1543

Fig. 4. Switching law of system (1) with Fd TC protocol.

Fig. 1. (a) Strongly connected network. (b) Leader-following network.

(a)

Fig. 2. Switching law of system (1) with FTC protocol.

(b)

Fig. 5. (a) System (1) with protocol (12). (b) System (1) with protocol (25).

classes of graphs in Fig. 1 to describe the communication

(a)
relationship of agents. Fig. 1(a) describes a strongly con-
nected network which satisfies the detail-balanced condition.
Fig. 1(b) describes a leader-following network. Suppose the
vertices 1−4 denote 4 followers and the vertice 5 denotes the
leader.
Example 1: The switching law of system (1) under FTC
protocol is shown in Fig. 2. We assume ω = [1, 2, 2, 1]T
and αij = αji = 0.2 for all i, j ∈ In . The initial states of
agents are x0 = [−3, −1, 1, 2]T and x0 = [−3, −1, 1, 2, 0]T
for strongly connected network Fig. 1(a) and leader-following
(b) network Fig. 1(b), respectively. By calculation, we can get
0 < h ≤ (5/3) and 0 < h ≤ (5/8) for different networks,
Fig. 3. (a) System (1) with protocol (2). (b) System (1) with protocol (9). respectively. For convenience, we choose h = 0.01. The state
trajectories of system (1) under FTC protocol (2) are shown
the dynamical switching behavior for this system. It can be in Fig. 3(a). The state trajectories of system (1) under FTC
found that the network only needs to be strongly connected protocol (9) are shown in Fig. 3(b).
and satisfy the detailed balance condition. Example 2: The switching law of system (1) under Fd TC
protocol is shown in Fig. 4. The directed communica-
V. S IMULATIONS tion topology, ω and h are the same as Example 1.
In this section, simulations are provided to illustrate the Parameters m = 9, r = 7, p = 5, and q = 7. The ini-
effectiveness of our theoretical results. We consider two tial values of all the agents are x0 = [2.5, 1.5, −0.5, −1.5]T

Authorized licensed use limited to: Indian Institute Of Technology (IIT) Mandi. Downloaded on August 01,2024 at 05:31:37 UTC from IEEE Xplore. Restrictions apply.
1544 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 47, NO. 7, JULY 2017

and x0 = [2.5, 1.5, −0.5, −1.5, 1]T for strongly connected [14] Z. Ji and H. Yu, “A new perspective to graphical characterization
network Fig. 1(a) and leader-following network Fig. 1(b), of multiagent controllability,” IEEE Trans. Cybern., to be published,
doi: 10.1109/TCYB.2016.2549034.
respectively. The state trajectories of system (1) under Fd TC [15] J. Ma, Y. Zheng, and L. Wang, “LQR-based optimal topology of leader-
protocol (12) are shown in Fig. 5(a). The state trajectories of following consensus,” Int. J. Robust Nonlin. Control, vol. 25, no. 17,
system (1) under Fd TC protocol (25) are shown in Fig. 5(b). pp. 3404–3421, 2015.
[16] F. Xiao and T. Chen, “Sampled-data consensus in multi-agent systems
with asynchronous hybrid event-time driven interactions,” Syst. Control
Lett., vol. 89, pp. 24–34, Mar. 2016.
VI. C ONCLUSION [17] R. Olfati-Saber and R. M. Murray, “Consensus problems in networks of
This paper discussed the FTC of the switched MAS. Two agents with switching topology and time-delays,” IEEE Trans. Autom.
Control, vol. 49, no. 9, pp. 1520–1533, Sep. 2004.
effective consensus protocols (FTC protocol and Fd TC proto-
[18] C. L. P. Chen, C. E. Ren, and T. Du, “Fuzzy observed-based adaptive
col) were proposed to solve the FTC problem and the Fd TC consensus tracking control for second-order multiagent systems with
problem, respectively. First, by using the FTC protocol for the heterogeneous nonlinear dynamics,” IEEE Trans. Fuzzy Syst., vol. 24,
switched MAS, we proved that the FTC problem under two no. 4, pp. 906–915, Aug. 2016.
[19] H. Li, X. Liao, and T. Huang, “Second-order locally dynamical con-
classes of special directed networks can be solved, respec- sensus of multiagent systems with arbitrarily fast switching directed
tively. Then, we considered the Fd TC of the switched MAS. topologies,” IEEE Trans. Syst., Man, Cybern., Syst., vol. 43, no. 6,
By applying the Fd TC protocol, we showed that the FTC pp. 1343–1353, Nov. 2013.
[20] G. Xie and L. Wang, “Consensus control for a class of networks of
problem can be solved and T0 is independent of the initial dynamic agents,” Int. J. Robust Nonlin. Control, vol. 17, nos. 10–11,
states of agents. Future work will focus on FTC problem of pp. 941–959, 2007.
the switched MAS under general directed network, FTC of [21] G.-X. Wen, C. L. P. Chen, Y.-J. Liu, and Z. Liu, “Neural-network-
the switched MAS with switching topology and finite-time based adaptive leader-following consensus control for second-order
non-linear multi-agent systems,” IET Control Theory Appl., vol. 9,
containment control of the switched MAS, etc. no. 13, pp. 1927–1934, Aug. 2015.
[22] S. Tong, L. Zhang, and Y. Li, “Observed-based adaptive fuzzy decen-
tralized tracking control for switched uncertain nonlinear large-scale
R EFERENCES systems with dead zones,” IEEE Trans. Syst., Man, Cybern., Syst.,
vol. 46, no. 1, pp. 37–47, Jan. 2016.
[1] X. Lin, X. Zhang, and Y. Wang, “Robust passive filtering for neutral-type [23] C. L. P. Chen, G.-X. Wen, Y.-J. Liu, and Z. Liu, “Observer-based adap-
neural networks with time-varying discrete and unbounded distributed tive backstepping consensus tracking control for high-order nonlinear
delays,” J. Franklin Inst., vol. 350, no. 5, pp. 966–989, 2013. semi-strict-feedback multiagent systems,” IEEE Trans. Cybern., vol. 46,
[2] X. Zhang, X. Lin, and Y. Wang, “Robust fault detection filter design for no. 7, pp. 1591–1601, Jul. 2016.
a class of neutral-type neural networks with time-varying discrete and [24] L. Xiao and S. Boyd, “Fast linear iterations for distributed averaging,”
unbounded distributed delays,” Optim. Control Appl. Methods, vol. 34, Syst. Control Lett., vol. 53, no. 1, pp. 65–78, 2004.
no. 5, pp. 590–607, 2013.
[25] L. Wang and F. Xiao, “Finite-time consensus problems for networks
[3] C. L. P. Chen, Y.-J. Liu, and G.-X. Wen, “Fuzzy neural network-based of dynamic agents,” IEEE Trans. Autom. Control, vol. 55, no. 4,
adaptive control for a class of uncertain nonlinear stochastic systems,” pp. 950–955, Apr. 2010.
IEEE Trans. Cybern., vol. 44, no. 5, pp. 583–593, May 2014.
[26] F. Xiao, L. Wang, and T. Chen, “Finite-time consensus in networks of
[4] Y.-J. Liu, J. Li, S. C. Tong, and C. L. P. Chen, “Neural network control-
integrator-like dynamic agents with directional link failure,” IEEE Trans.
based adaptive learning design for nonlinear systems with full-state
Autom. Control, vol. 59, no. 3, pp. 756–762, Mar. 2014.
constraints,” IEEE Trans. Neural Netw. Learn. Syst., vol. 27, no. 7,
[27] F. Xiao, L. Wang, J. Chen, and Y. Gao, “Finite-time formation control
pp. 1562–1571, Jul. 2016.
for multi-agent systems,” Automatica, vol. 45, no. 11, pp. 2605–2611,
[5] Y.-J. Liu, S. C. Tong, C. L. P. Chen, and D.-J. Li, “Neural con-
2009.
troller design-based adaptive control for nonlinear MIMO systems with
unknown hysteresis inputs,” IEEE Trans. Cybern., vol. 46, no. 1, [28] Z. Meng, W. Ren, and Z. You, “Distributed finite-time attitude contain-
pp. 9–19, Jan. 2016. ment control for multiple rigid bodies,” Automatica, vol. 46, no. 12,
[6] Y.-J. Liu and S. C. Tong, “Optimal control-based adaptive NN design pp. 2092–2099, 2010.
for a class of nonlinear discrete-time block-triangular systems,” IEEE [29] J. Cortés, “Finite-time convergent gradient flows with applications to
Trans. Cybern., vol. 46, no. 11, pp. 2670–2680, Nov. 2016. network consensus,” Automatica, vol. 42, no. 11, pp. 1993–2000, 2006.
[7] F. Li, P. Shi, L. Wu, M. V. Basin, and C.-C. Lim, “Quantized [30] Q. Hui, “Finite-time rendezvous algorithms for mobile autonomous
control design for cognitive radio networks modeled as nonlinear semi- agents,” IEEE Trans. Autom. Control, vol. 56, no. 1, pp. 207–211,
Markovian jump systems,” IEEE Trans. Ind. Electron., vol. 62, no. 4, Jan. 2011.
pp. 2330–2340, Apr. 2015. [31] F. Jiang and L. Wang, “Finite-time weighted average consensus with
[8] Y. Zheng, Y. Zhu, and L. Wang, “Consensus of heterogeneous respect to a monotonic function and its application,” Syst. Control Lett.,
multi-agent systems,” IET Control Theory Appl., vol. 5, no. 16, vol. 60, no. 9, pp. 718–725, 2011.
pp. 1881–1888, Nov. 2011. [32] Y. Zheng and L. Wang, “Finite-time consensus of heterogeneous
[9] B. Wang, J. Wang, B. Zhang, and X. Li, “Global cooperative con- multi-agent systems with and without velocity measurements,” Syst.
trol framework for multiagent systems subject to actuator saturation Control Lett., vol. 61, no. 8, pp. 871–878, 2012.
with industrial applications,” IEEE Trans. Syst., Man, Cybern., Syst., [33] X. Liu, M. Z. Q. Chen, H. Du, and S. Yang, “Further results on
to be published, doi: 10.1109/TSMC.2016.2573584. finite-time consensus of second-order multi-agent systems without veloc-
[10] J. Ma, Y. Zheng, and L. Wang, “Topology selection for multi-agent ity measurements,” Int. J. Robust Nonlin. Control, vol. 26, no. 14,
systems with opposite leaders,” Syst. Control Lett., vol. 93, pp. 43–49, pp. 3170–3185, 2015.
Jul. 2016. [34] X. Liu, J. Lam, W. Yu, and G. Chen, “Finite-time consensus of
[11] Z. Ji, H. Lin, and H. Yu, “Protocols design and uncontrollable topologies multiagent systems with a switching protocol,” IEEE Trans. Neural
construction for multi-agent networks,” IEEE Trans. Autom. Control, Netw. Learn. Syst., vol. 27, no. 4, pp. 853–862, Apr. 2016.
vol. 60, no. 3, pp. 781–786, Mar. 2015. [35] S. E. Parsegov, A. E. Polyakov, and P. S. Shcherbakov, “Fixed-time
[12] Y. Guan, Z. Ji, L. Zhang, and L. Wang, “Decentralized stabilizability of consensus algorithm for multi-agent systems with integrator dynamics,”
multi-agent systems under fixed and switching topologies,” Syst. Control in Proc. 4th IFAC Workshop Distrib. Estimation Control Netw. Syst.,
Lett., vol. 62, no. 5, pp. 438–446, 2013. Koblenz, Germany, 2013, pp. 110–115.
[13] Y. Chao and Z. Ji, “Necessary and sufficient conditions for multi-agent [36] M. Defoort, A. Polyakov, G. Demesure, M. Djemai, and K. Veluvolu,
controllability of path and star topologies by exploring the information “Leader-follower fixed-time consensus for multi-agent systems with
of second-order neighbours,” IMA J. Math. Control Inf., to be published, unknown non-linear inherent dynamics,” IET Control Theory Appl.,
doi: 10.1093/imamci/dnw013. vol. 9, no. 14, pp. 2165–2170, Sep. 2015.

Authorized licensed use limited to: Indian Institute Of Technology (IIT) Mandi. Downloaded on August 01,2024 at 05:31:37 UTC from IEEE Xplore. Restrictions apply.
LIN AND ZHENG: FTC OF SWITCHED MASs 1545

[37] J. Fu and J. Wang, “Fixed-time coordinated tracking for second-order [48] R. A. Horn and C. R. Johnson, Matrix Analysis. Cambridge, U.K.:
multi-agent systems with bounded input uncertainties,” Syst. Control Cambridge Univ. Press, 2012.
Lett., vol. 93, pp. 1–12, Jul. 2016. [49] P. Hartman, Ordinary Differential Equations. Basle, Switzerland:
[38] D. B. Kingston and R. W. Beard, “Discrete-time average-consensus Birkhäuser, 1982.
under switching network topologies,” in Proc. Amer. Control Conf.,
Minneapolis, MN, USA, 2006, pp. 3551–3556.
[39] A. Y. Kibangou, “Step-size sequence design for finite-time average con-
sensus in secure wireless sensor networks,” Syst. Control Lett., vol. 67,
pp. 19–23, May 2014. Xue Lin received the B.Sc. degree in applied
[40] F. Li, L. Wu, P. Shi, and C.-C. Lim, “State estimation and slid- mathematics from Suihua University, Suihua,
ing mode control for semi-Markovian jump systems with mismatched China, in 2010, and the M.Sc. degree in opera-
uncertainties,” Automatica, vol. 51, pp. 385–393, Jan. 2015. tional research and cybernetics from Heilongjiang
[41] Y. Li, S. Sui, and S. Tong, “Adaptive fuzzy control design for stochastic University, Harbin, China, in 2013. She is currently
nonlinear switched systems with arbitrary switchings and unmodeled pursuing the Ph.D. degree in control science and
dynamics,” IEEE Trans. Cybern., vol. 47, no. 2, pp. 403–414, Feb. 2017. engineering from Xidian University, Xi’an, China.
[42] Y. Li and S. Tong, “Adaptive fuzzy output-feedback stabilization con- Her research interests include coordination
trol for a class of switched nonstrict-feedback nonlinear systems,” IEEE of multiagent systems, consensus problems, and
Trans. Cybern., vol. 47, no. 4, pp. 1007–1016, Apr. 2017. neutral-type neural networks.
[43] F. Li, L. Wu, and P. Shi, “Stochastic stability of semi-Markovian jump
systems with mode-dependent delays,” Int. J. Robust Nonlin. Control,
vol. 24, no. 18, pp. 3317–3330, 2014.
[44] Y. Zheng and L. Wang, “Distributed consensus of heterogeneous
multi-agent systems with fixed and switching topologies,” Int. J. Control,
vol. 85, no. 12, pp. 1967–1976, 2012. Yuanshi Zheng was born in Jiangshan, China. He
[45] Y. Zheng and L. Wang, “Consensus of switched multi-agent systems,” received the bachelor’s and master’s degrees from
IEEE Trans. Circuits Syst. II, Exp. Briefs, vol. 63, no. 3, pp. 314–318, Ningxia University, Yinchuan, China, in 2006 and
Mar. 2016. 2009, respectively, and the Doctorate degree from
[46] X. Lin, Y. Zheng, and L. Wang, “Consensus of switched multi- Xidian University, Xi’an, China, in 2012.
agent systems with random networks,” Int. J. Control, vol. 90, no. 5, He is currently an Associate Professor with Xidian
pp. 1113–1122, 2017. University. His research interests include coordi-
[47] Y. Zhu, Y. Zheng, and L. Wang, “Containment control of switched nation of multiagent systems, game theory, and
multi-agent systems,” Int. J. Control, vol. 88, no. 12, pp. 2570–2577, collective intelligence.
2015.

Authorized licensed use limited to: Indian Institute Of Technology (IIT) Mandi. Downloaded on August 01,2024 at 05:31:37 UTC from IEEE Xplore. Restrictions apply.