ISA Transactions 71 (2017) 3–9

Contents lists available at ScienceDirect

ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans

Research article

Coordinated tracking of linear multiagent systems with input

saturation and stochastic disturbances$
Qingling Wang a,b, Changyin Sun a,b,n
a
School of Automation, Southeast University, Nanjing 210096, China
b
Key Laboratory of Measurement and Control of Complex Systems of Engineering, Ministry of Education, Nanjing 210096, China

art ic l e i nf o a b s t r a c t

Article history: This paper addresses the coordinated tracking problem for linear multiagent systems with input sa-
Received 21 November 2016 turation and stochastic disturbances. The objective is to construct a class of tracking control laws that
Received in revised form achieve consensus tracking in the absence of disturbances, while guaranteeing a bounded variance of the
27 June 2017
state difference between the follower agent and the leader in the present of disturbances, under the
Accepted 12 July 2017
assumptions that each agent is asymptotically null controllable with bounded controls (ANCBC) and the
Available online 22 July 2017
network is connected. By using the low gain feedback technique, a class of tracking control algorithms
Keywords: are proposed, and the coordinated tracking problem is solved through some routine manipulations. Fi-
Coordinated tracking nally, numerical examples are provided to demonstrate the theoretical results.
Stochastic disturbances
& 2017 ISA. Published by Elsevier Ltd. All rights reserved.
Input saturation
Low gain feedback

1. Introduction
During the last several decades, coordinated control of mul-
tiagent systems have received considerable attention due to its
wide applications, see, for instance, [6,3,18,15,2,43,46,9,13,10,47],
and the survey papers [27,23] and references therein for more
details and developments. Many researchers have done much
effort on coordinated tracking of multiagent systems, and various
coordinated tracking algorithms have been developed, including
the consensus tracking algorithm and the swarm tracking algo-
rithm. The consensus tracking means a group of agents, only
relying on their neighbors' information, precisely reach the mo-
tion of the leader, while the swarm tracking means a group of
agents, only relying on their neighbors' information, reach in a
predesigned set of the motion of the leader. Recent efforts of
coordinated tracking are mainly focused on multiagent systems
with linear and nonlinear systems [21,19,44,12,42,25,40] while
previous efforts were mainly on first- and second-order dynamics
[7,41,34,5].
☆
This work was partially supported by the National Natural Science Foundation
of China (61503079, 61520106009 and 61533008), the Jiangsu Natural Science
Foundation (BK20150625), the Fundamental Research Funds for the Central Uni-
versities, and a project funded by the Priority Academic Program Development of
Jiangsu Higher Education Institutions. This paper was not presented at any IFAC
meeting.
n turation is considered in [4]. It is worth to mention that dis-
Corresponding author at: School of Automation, Southeast University, Nanjing
210096, China
E-mail addresses: csuwql@gmail.com (Q. Wang), cysun@seu.edu.cn (C. Sun).
Input saturation has been widely considered owing to its uni-
versal existence in control systems. There have been plenty
methods proposed for the stabilization problem of a single linear
system with input saturation [35,36,16,48]. Among them, the low
gain feedback [17,14] is an effective method to design a class of
linear control laws in the semi-global stabilization framework. The
feature of the low gain feedback is that given any bounded set of
initial conditions, the closed-loop system can always stay in the
linear domain by scheduling the feedback gain parameter small
enough. Recently, the low gain feedback has been extended to
treat the coordinated tracking problem of multiagent systems. By
using low gain feedback, the semi-global consensus tracking for
linear multiagent systems with actuator saturation has been in-
vestigated in [32,38,39]. Meanwhile, there are other works focused
on coordinated tracking problem with saturation constraints from
different perspectives. For example, in multi-agent systems, the
coordinated tracking problem with input saturation has been
studied for single-integrators [11] and double-integrators
[26,45,1], respectively. In [20], the global consensus tracking pro-
blem was solved for high-order multiagent systems with input
saturation. By using low and high gain feedback method, the semi-
global coordinated tracking of linear multi-agent systems with
input saturation was studied in [31]. With a modified algebraic
Riccati equation, the problem of semi-global leader-following
synchronization of discrete-time linear agents with actuator sa-
turbance uncertainties also universal exist in control systems.
Generally, disturbance uncertainties can be divided into two types,

http://dx.doi.org/10.1016/j.isatra.2017.07.016
0019-0578/& 2017 ISA. Published by Elsevier Ltd. All rights reserved.
4 Q. Wang, C. Sun / ISA Transactions 71 (2017) 3–9
that is, input-additive disturbances and non-input-additive dis-
turbances. For the input-additive disturbances, the disturbances
can be shown in form of δ ( u + ω), while the form of δ ( u) + ω
represents the non-input-additive disturbances, where δ( ·) is the
saturation function, u is the control input, and ω is the dis-
turbances. It is known [28] that the general ANCBC systems with
input saturation can achieve stability with a linear control law
when the disturbance is input-additive, while in general [30], it is
impossible to achieve stability via a linear control law in the
presence of the non-input-additive disturbance. This implies that
the picture is complicated when considering the non-input-ad-
ditive disturbance. As for the multiagent systems, some previous
works have considered both input saturation and input-additive
disturbances. For example, the global coordinated tracking for
linear multiagent systems with input saturation and input-ad-
ditive disturbances was shown to be achievable in [33,37]. Al-
though much effort has been made to solve the coordinated
tracking problem for linear multiagent systems with both input
saturation and input-additive disturbances, to the best of our
knowledge, stochastic disturbances, the non-input-additive dis-
turbances, has not been considered in the existent results.
With the above observations, in this paper, we deal with the
coordinated tracking for a group of agents with input saturation
and stochastic disturbances. By using the scheduled low gain de-
sign technique, we propose a class of tracking control algorithms
that can always achieve consensus tracking in the absence of
disturbances and guarantee a bounded variance of the state dif-
ference between the follower agent and the leader in the present
of disturbances, under the assumptions that each agent is ANCBC
and the network is connected. The major contributions of this
paper are twofold. Firstly, compared with the results in [33,37]
with input-additive disturbances, the disturbances discussed in
this paper are non-input-additive, which implies that the scenario
of the stability analysis of multiagent systems is more complex.
Secondly, to the best of our knowledge, this is the first attempt to
take the stochastic disturbances into account for coordinated
tracking control of multiagent systems. Consequently, one of the
main challenges is how to properly deal with stochastic and non-
input-additive disturbances for achieving bounded variance in the
controlled multiagent system. It is appropriate to emphasize that it
is not clear that such coordinated tracking algorithms will exist,
and we provide an explicit coordinated tracking design method
based on the low gain feedback and a scheduling parameter.
The rest of the paper is organized as follows. Section 2 for-
mulates the coordinated tracking problem. Section 3 presents the
main results on the coordinated tracking. Section 4 provides si-
mulation examples for verifying the theoretical results. Finally,
some conclusion remarks are given in Section 5.
Before closing this section, some notations will be stated here. The
notation diag{ω1, ω2, … , ωn} denotes a diagonal matrix whose diag-
onal entries are ωi , i = 1, 2, … , n. col { x1, x2, … , x n} and 1N denote a
column vector and an N-dimensional column vector with all elements
of 1, respectively. Matrices, if their dimensions are not explicitly sta-
ted, are assumed to be compatible for algebraic operations. The no-
tation P > 0 ( ≥ 0) means that P is a real symmetric positive (semi-
positive) definite matrix. In ∈ n × n and 0 represent, respectively, the
identity matrix and zero matrix. The set of real numbers is denoted by
 . The set of real-valued vectors of length m is given by m . The set of
real-valued m
n matrices is given by m × n .
2. Problem formulation
In the beginning, the graph theory will be introduced. Let
 = (  ,  ,  ) be a weighted directed graph, where  = { v1, v2, …, vN }
is a finite nonempty set of nodes and edges  ⊆  × .
 = ⎡⎣ aij ⎤⎦ ∈  N × N is the weighted adjacency matrix, where aij is the
coupling strength of the directed edge ( j, i) satisfying aij > 0 if ( j, i)
is an edge of  and aij = 0 otherwise. Let Ni = { j ∈  : ( j, i) ∈ } be
N
the set of neighbors of node i in  . Let di = ∑ j = 1 aij the in-degree of
vertex i, and D = diag{ d1, … , dN } the in-degree matrix of  . The
Laplacian matrix L = ⎡⎣ lij ⎤⎦ of weighted digraph  is defined by
L = D −  . For any pair of vertices ( i, j ), if aij = aji , the graph is called
an undirected graph. In this case,  is connected if there is a directed
path between any pair of distinct nodes; otherwise, it is termed a
directed graph. A directed tree is a directed graph, where every
node, except the root, has exactly one parent. A spanning tree of a
directed graph is a directed tree formed by graph edges that connect
all the nodes of the graph. We say that a graph has (or contains) a
spanning tree if a subset of the edges forms a spanning tree.
Given a graph  , endow each of its N agents with a state vector
x i ∈ n and a control input ui ∈ m , and we consider each agent,
labeled from 1 to N, is described by a stochastic differential equation
dxi(t ) = Axi (t )dt + Bδ( ui (t ))dt + Edωi(t ), i = 1, 2, …, N , (1)
where A ∈ n × n , B ∈ n × m , E ∈ n × l , the disturbance ωi ∈ l is a
Wiener process (a Brownian motion) with mean 0 and rate Qi, the
initial condition x i(0) is a Gaussian random vector which is in-
dependent of ωi, and δ( ·) is a standard saturation function given as
⎧ u, if u ≤ 1
δ( u) = ⎨ .
⎩ sign(u), if u > 1
The dynamics of the leader agent, indexed by 0, is described by
dx 0(t ) = Ax 0 (t )dt. (2)
Let  be a graph generated by graph  and the leader agent. If agent
i is a neighbor of the leader at time t, then denote it by bi0 > 0,
otherwise bi0 = 0. We also assume that bi0 > 0 for at least one i, for
all i = 1, 2, … , N . Let L be the Laplacian of an undirected graph 
consisting of N agents, and denote B0 = diag{ b10 , b20 , … , bN 0},
b0 = col{ b10 , b20 , … , bN 0} and 0 < λ1 ≤ λ2 ≤ ⋯ ≤ λN the N eigenvalues
of matrix L + B0 , which can be easily obtained according to [7].
It is known from [35] that a linear system with input saturation
can be stabilized if and only if ( A , B) is stabilizable, and all eigen-
values of A are in the closed left-half plane. Without loss of gen-
erality, we assume that the pair ( A , B) is given in the following form:
⎡ 0 ⎤ ⎡ 0⎤
A = ⎢ A 0− ⎥, B = ⎢ B− ⎥
⎣0 A ⎦ ⎣B ⎦
where A− contains all eigenvalues of A that have negative real parts,
and A0 contains all eigenvalues of A that are on the imaginary axis.
( )
The stabilizability of ( A , B) then implies that A0 , B 0 is controllable.
Clearly, ( A− , B−) does not affect the stabilizability property of the
system. In what follows, without loss of generality, we will further
give an assumption:
Assumption 1. The pair ( A , B) is controllable and all the eigen-
values of A are on the imaginary axis.
Remark 1. According to Assumption 1, no exponentially unstable
mode exists in A . It is known that Assumption 1 is actually a ne-
cessary condition for stabilizing linear systems with input sa-
turation by low gain feedback [16].
Assumption 2. The generated graph  consisting of the follower
agents and the leader contains a directed spanning tree rooted at
the leader.
Remark 2. Assumption 2 is quite general for multiagent systems
[37,38], and it is reasonable that for the leader-follower
 Q. Wang, C. Sun / ISA Transactions 71 (2017) 3–9 5

framework, the follower agents can exchange information in bi- Step 1: Let
directional ways.

The coordinated tracking problem can be defined as follows.

{ ( ) (
ε( x ) = max ε( x ) ∈ ( 0, 1)∣tr IN ⊗ BT Pε( x )B x¯ T IN ⊗ Pε( x ) x¯ ≤ λN−2 , ) } (7)

where x¯ = col{ x¯1, x¯2, … , x¯N }, x¯ i = x i − x0, i = 1, 2, … , N is the dif-

Definition 1. Consider the follower agents (1) and a leader (2), the
coordinated tracking problem is to find a control law ui, which
only uses local relative information from neighbor agents, such
that, for all possible values for the rate Qi and the stochastic pro-
cess ωi, the following properties hold:
(i) In the absence of the disturbance ωi, the consensus tracking is
achieved, that is, limt →∞ x i(t ) − x0(t ) = 0 , i = 1, 2, … , N .
(ii) In the presence of the disturbance ωi, the variance of the state
difference between the follower agent and the leader is
bounded, that is, limt →∞ x i(t ) − x0(t ) < ∞, i = 1, 2, … , N .
It is pointing out that Definition 1 is motivated from the ex-
ternal stochastic stabilization problem of [29], and it imposes a
stronger objective than that of [29] where the coordinated track-
ing of the multi-agent system (1) and (2) with input saturation and
stochastic disturbances should be achieved.
ference between the states of agent i and the leader agent.
Step 2: Solve the following parametric ARE
AT Pε( x ) + Pε( x )A − 2γPε( x )BBT Pε( x ) = − ε( x )Pε( x ), (8)
where γ ≤ λ1, and the existence of the positive definite Pε( x ) is
guaranteed by Lemma 1.
Step 3: Construct a nonlinear feedback control law for agent i as
⎛ N ⎞
ui = − BT Pε( x )⎜⎜ ∑ aij xi − xj + bi0( xi − x 0)⎟⎟,
( )
⎝ j=1 ⎠ (9)
where −BT Pε( x ) is the coupling gain.
Based on the above design procedure, we have the following
lemma.
Lemma 2. Assume that the pair (A,B) satisfies Assumption 1, and
define Fε( x ) = − ( L + B0) ⊗ BT Pε( x ) . We have that

3. Main results Fε( x ) x ≤ 1.

Proof. It is clear from (7) that,

In this section, the defined coordinated tracking problem will
2
be investigated, and we firstly introduce the following parametric Fε( x ) x
2
= −( L + B0) ⊗ BT Pε( x ) x
algebraic Riccati equation (ARE),
⎡ 2 ⎤
= x T ⎣ ( L + B0) ⊗ Pε( x )BBT Pε( x )⎦x
AT P (ε) + P (ε)A − γP (ε)BBT P (ε) = − εP (ε), (3)

where ε > 0 and γ > 0 are scalars. This parametric ARE was first (
≤ λN2x T IN ⊗ Pε( x )BBT Pε( x ) x )
introduced in [48] for a linear system, and then was extended in
[38] for a multiagent system. The properties of (3) can be sum-
≤ λN2tr (I N ⊗ B P ( )B)x¯ ( I
T
ε x
T
N )
⊗ Pε( x ) x¯ ,
marized as follows.
which is less than or equal to 1. □
Lemma 1. Assume that (A,B) satisfies Assumption 1. Then, for an
arbitrary ε > 0 and γ > 0, the ARE (3) has a unique positive solution
Lemma 3 ([24]). Let Vn ≥ 0 be random variables and { n} be a fil-
P (ε ) = W −1(ε ), where W (ε ) satisfies the parametric Lyapunov equation tration to which { Vn} is adapted and suppose that there exist con-
stants a > 0, J and M < ∞ and p > 2 such that V0s is bounded for
⎛ ε ⎞T ⎛ ε ⎞
W ⎜ A + In⎟ + ⎜ A + In⎟W = γBBT ,
⎝ s ∈ ⎣⎡ 1, p⎤⎦ and for all n
2 ⎠ ⎝ 2 ⎠ (4)
 ( Vn + 1 − Vn∣n) < − a, on the event that { Vn > J}, (10)
and
dP (ε) and
> 0, ∀ ε > 0, limP (ε) = 0.
dε (5) p
ε→ 0
(
 Vn + 1 − Vn ∣n ≤ M. ) (11)
Furthermore, P (ε ) is a polynomial function in ε for a scalar input. In
general, P (ε ) is a rational function in ε. Then for any r ∈ ( 0, p − 1), there is a c = c( a, J , M , p, r ) such
that  Vnr < c for all n.
Remark 3. It is observed that P (ε ) is positive definite, and note
from (3) that The following is the main result on the coordinated tracking
problem.
AT + P (ε)AP −1(ε) − γP (ε)BBT = − εIn.
Theorem 1. Consider N follower agents described by the stochastic
By taking trace on both sides, we have
differential Eq. (1), with a leader described by (2). Let Assumptions
( ) ( ) ( )
tr AT + tr P (ε)AP −1(ε) − tr γP (ε)BBT = − tr( εIn), 1 and 2 hold. Then, under the feedback control law (9), all the agents
(1) can coordinated track to the leader (2).
which indicates
Proof. Regarding x¯ i = x i − x0, i = 1, 2, … , N , the state difference
nε
(
tr BT P (ε)B =) γ
.
(6)
between agent i and the leader. By using (1), (2) and (9), we obtain
dx i = Ax¯ i dt + Bδ ( ui )dt + Edωi
T
where we have used tr(Y X ) = tr(Y X ) and tr(A) = tr(A ) = 0. ⎛ ⎛ N ⎞⎞
= Ax¯ i dt + Bδ ⎜ −BT Pε( x )⎜⎜ ∑ a ij x i − x j + bi0 ( x i − x 0 )⎟⎟⎟
( )
In what follows, we propose a class of scheduled control laws to ⎜ ⎟
⎝ ⎝ j =1 ⎠⎠
solve the coordinated tracking problem, which can be carried out
dt + Edωi , (12)
in three steps.
6 Q. Wang, C. Sun / ISA Transactions 71 (2017) 3–9

which can be written in a compact form ∂V

dV =
∂ε
dt + 2x T (IN ⊗ Pε( x ))dx + tr IN ⊗ ET Pε( x )E dt , ( ) (20)
dx = ( IN ⊗ A)xdt
¯ + ( IN ⊗ B)δ − ( ( ( L + B ) ⊗ B P ( ))x¯)dt
0
T
ε x
and substituting (19), we obtain
+ ( IN ⊗ E )dω, (13)
1 T
where x¯ = col{ x¯1, x¯2, … , x¯N } and ω = col{ ω1, ω2, … , ωN }. Recalling
dV = −
ε( x )
( )
x¯ IN ⊗ Pε( x ) x¯ + 2x T (IN ⊗ Pε( x ))dx

from Lemma 2 that Fε( x )x ≤ 1, we can continue (13) as follows

(
+ tr IN ⊗ ET Pε( x )E dt )
dx = ⎡⎣ IN ⊗ A − ( L + B0) ⊗ BBT Pε( x )⎤⎦xdt
¯ + ( IN ⊗ E )dω. (14)
⎛ 1 ⎞
≤ −⎜⎜
⎝ ε( x )
+ ε( x )⎟⎟x¯ T IN ⊗ Pε( x ) xdt
⎠
¯ ( )
It is observed that the coordinated tracking problem is tantamount
to the external stochastic stabilization problem of system (14). (
+ tr IN ⊗ ET Pε( x )E dt + 2x¯ T IN ⊗ Pε( x )E dω. ) ( ) (21)
That is, the equilibrium point x¯ = 0 of system (14) is globally stable
in the absence of the disturbance ω , while the variance of the state Define
difference x̄ of system (14) is bounded in the presence of the ⎛ 1 ⎞
disturbance ω . f1(x ) = − ⎜⎜
⎝ ε( x )
(
+ ε( x )⎟⎟x¯ T IN ⊗ Pε( x ) x¯ + tr IN ⊗ ET Pε( x )E ,
⎠
) ( ) (22)
Consider a Lyapunov function as
and based on (7), we have
V = x T (IN ⊗ Pε( x ))x.
⎛ 1 ⎞
According to Lemma 1, it is easily verified that V is a global Lya- f1(x ) = −⎜⎜
⎝ ( )
ε x ⎠
(
+ ε( x )⎟⎟x¯ T IN ⊗ Pε( x ) x¯ + tr IN ⊗ ET Pε( x )E ) ( )
punov function for system (14) in the absence of the disturbance
ω , which establishes global asymptotic stability. It remains to show ⎛ 1 ⎞
that the variance of the state difference x̄ of system (14) is = ⎜⎜
⎝ ε( x )
(
+ ε( x )⎟⎟x¯ T IN ⊗ Pε( x ) x¯ − tr IN ⊗ ET Pε( x )E
⎠
) ( )
bounded in the presence of the disturbance ω . In view of Lemma 3,
2
we will use Vt = x T (IN ⊗ Pε( x ))x for t ¼n and { t} is the natural fil- γε( x ) + γ
tration generated by the stochastic process x .
≤
λN2Nnε( x )
2 (
+ tr IN ⊗ ET Pε( x )E ),
Firstly, we will prove (11) for arbitrary integers p. Note from (7)
which means that f1(x ) is bounded. Note that
( ) ( )
that tr IN ⊗ BT Pε( x )B x¯ T IN ⊗ Pε( x ) x¯ ≤ λN−2, we have
⎡ ⎤
d⎣ tr( I ⊗ B P ( )B)x¯ ( I ⊗ P ( ))x¯ ⎦ = 0.
N
T
ε x
T
N ε x
(
f2 (x ) = 2x¯ T IN ⊗ Pε( x )E , ) (23)

and with (7), we have

By using the It o stochastic differential rule, it is obtained that
1/2 1/2
⎡
⎢ d IN ⊗ Pε( x ) ( ) ⎡ ⎤⎤
d⎣⎢ tr IN ⊗ BT Pε( x )B ⎦⎥ ⎥ ( )
f2 (x ) ( ) ( I ⊗ P ( )) I ⊗ E
≤ 2 x¯ T IN ⊗ Pε( x ) x¯ N ε x N

⎢
( T
⎢ tr IN ⊗ B Pε( x )B x¯ T
dε
) T
x¯ + x¯ IN ⊗ Pε( x ) x¯ (dε
) ⎥dε
⎥ = 2 tr( I ⊗ B P ( )B)x¯ ( I ⊗ P ( ))x¯
T
tr( I ⊗ B P ( )B)T
1/2
T
−1/2

⎣ ⎦ N ε x N ε x N ε x

⎡ ⎤
( ) ( )
= − tr IN ⊗ BT Pε( x )B ⎢⎣ 2x¯ T IN ⊗ Pε( x ) dx¯ + tr IN ⊗ ET Pε( x )E dt ⎥⎦ . ( ) (15) × ( I ⊗ P ( )) N I ⊗E ε x
1/2
N

−1/2 1/2
It is noted that
≤ 2λ tr( I ⊗ B P ( )B)
−1
N N ( I ⊗ P ( )) I ⊗ E ,
T
ε x N ε x N

(
d IN ⊗ Pε( x ) )
dV = x T
dε
x dε + 2x T (IN ⊗ Pε( x ))dx + tr IN ⊗ ET Pε( x )E dt , ( ) (16)
which means that f2 (x ) is bounded. Thus we have,
t t
and substituting (15), we obtain Vt − Vn ≤ ∫n f1(x )dt + ∫n f2 (x )dω .
(24)
⎡ ⎡ ⎤⎤
⎢ −1 d⎣ tr IN ⊗ BT Pε( x )B ⎦ ⎥ ( ) It follows from [8] that all higher order moments of
t
∫n f2 (x )dω are
T
( T
dV = − ⎢ tr IN ⊗ B Pε( x )B x¯ IN ⊗ Pε( x ) x¯
⎢
) (
dε
⎥dε .
⎥
) bounded, and given that f1(x ) is bounded, we trivially find that
⎣ ⎦ (17)
p
 Vt − Vn ∣n is bounded for any p and for any t ∈ ⎡⎣ n, n + 1⎤⎦
( )
According to (6), we have
which implies (11).
nε( x )
(
tr BT Pε( x )B = ) . Next, we will prove (10). Since the condition is tr IN ⊗ BT (
2γ

It follows that
Pε( x )B x¯) (I
T
N ⊗ Pε( x ) x¯ ≤ ) λN−2 , there exists a κ > 0 so that x¯ T ( IN ⊗

Nnε( x )
)
Pε( x ) x¯ ≥ κ . Consider a J1 such that V > J1 implies that
(
tr IN ⊗ B Pε( x )B =T
) 2γ
.
(18) κ⎛ 1 ⎞
(
tr IN ⊗ ET Pε( x )E ≤ ) ⎜⎜
2 ⎝ ε( x )
+ ε( x )⎟⎟.
⎠
Therefore, we can continue (17) as
According to (21), we have
dV 1 T
dε
=−
ε( x )
x¯ IN ⊗ Pε( x ) x¯ .( ) (19) n+ 1 ⎡ ⎛ ⎞ ⎤
(
 Vn + 1 − Vn ∣n ≤
p
) ∫ ⎢ −κ ⎜⎜ 1 + ε( x )⎟⎟ + tr IN ⊗ ET Pε x E ⎥dt . ( )
⎢⎣ ⎝ ε( x ) ( ) ⎥⎦
n ⎠
On the other hand, it is noted that
 Q. Wang, C. Sun / ISA Transactions 71 (2017) 3–9 7

Assuming Vn > J with J > J1, and using the Doob's martingale in- traces in digital signal process (DSP) systems. Transmission lines(TLs)
equality [22], we have are becoming increasingly important in industrial applications, such
as power lines and telephone lines. In high-speed DSP systems, traces
⎤  ⎡⎣ Vt − Vn ⎤⎦
2
⎡ become TLs when the signals on those traces have a rise-time less
 ⎢ sup Vt − Vn > J − J1 ⎥ ≤ .
⎢⎣ t ∈ ⎡⎣ n, n + 1⎤⎦ ⎦⎥
2 than twice the propagation delay. There are two types of TLs, lossless
J − J1 ( ) and lossy. Although all TLs are lossy, in practice, the losses on lines or
That is, for any ρ, we can choose J such that traces are negligible compared to losses in the entire signal chain.
Therefore, in this example, we use two-section lossless TLs with four
⎡ ⎤
 ⎢ inf Vt < J1 ∣Vn > J ⎥ < ρ . elements as shown in Fig. 1. Here the inductance L, capacitance C,
⎣ t ∈ ⎡⎣ n, n + 1⎤⎦ ⎦ current I and S are all normalized unit. u is the input of the lossless TL.
r and y are the outputs of the first and second sections, respectively.
However, for Vn > J , we have
Meanwhile, the output of second section usually subject to stochastic
⎡ ⎤ disturbances. Therefore, define the state as x=
 ( Vn + 1 − Vn∣n) ≤ −  ⎢ inf Vt > J1 ∣Vn > J ⎥T1 col{ x1, x2, x3, x4} = col{ I , r , S, y} . Then, the model can be given by
⎣ ⎣
t ∈ ⎡ n, n + 1⎤
⎦ ⎦
⎡ ⎤ dx(t ) = Ax(t )dt + Bδ( u(t ))dt + Edω(t )
+  ⎢ inf Vt < J1 ∣Vn > J ⎥T2,
⎣ t ∈ ⎡⎣ n, n + 1⎤⎦ ⎦
where
⎛ 1 ⎞ ⎡ ⎤
where −T1 = − κ
2 ( 1
ε( x )
+ ε( x ) ) is an upper bound for −κ ⎜
⎝ ε( x )
+ ε( x )⎟+
⎠
1
⎢0 −L 0 0 ⎥ ⎡ 1⎤ ⎡ 0⎤
⎢1 ⎥ ⎢ L⎥ ⎢ ⎥
(
tr IN ⊗ ET Pε( x )E ) and T = tr( I 2 N )
⊗ ET Pε( x )E . Since we have given that 1
⎢ C 0 −C 0 ⎥
A=⎢
⎢ ⎥
B = ⎢ 0 ⎥,
⎢ 0⎥
V > J1, we obtain 1 1
⎥, E = ⎢ 0 ⎥.
⎢0 L 0 −L⎥ ⎢ 0⎥ ⎢ 1⎥
 ( Vn + 1 − Vn∣n) ≤ − T1( 1 − ρ) + T2ρ , ⎢ ⎥ ⎢⎣ 0⎥⎦ ⎢⎣ L ⎥⎦
⎢⎣ 0 0 1 0 ⎥⎦
C
which is less than −a for a = T1/2 provided ρ is small enough. This
implies that (10) is satisfied. Therefore, we conclude from Lemma It is easy to obtain that Assumption 1 holds.
3 that Vtr is bounded for all positive integers r. The coordinated tracking problem is formulated as follows. Let
In what follows, we will prove that the variance of x is bounded one two-section lossless TL be the leader agent with u ¼0 and
under the condition that Vtr is bounded for all positive integers r. trying to produce a desired signal. Four two-section lossless TLs
are the follower agents, and these agents can get information from
Since Pε−(1x ) is a rational function in ε( x ), there exists η > 0 and an
their neighbors with an graph. That is, the Laplacian matrix L and
integer q such that the matrix B0 are given as follows
−q
Pε−(1x ) ≤ ηε( x ) , ⎡ 2 −1 0 −1⎤ ⎡1 0 0 0⎤
⎢ ⎥ ⎢ ⎥
−1 2 −1 0⎥ 0 0 0 0⎥
which implies L=⎢ , B0 = ⎢ ,
⎢ 0 −1 2 −1⎥ ⎢0 0 0 0⎥
q ⎢⎣ −1 0 −1 2 ⎥⎦ ⎢⎣ 0 0 0 0⎥⎦
Pε( x ) ≥ η−1ε( x ) .
which means that Assumption 2 is satisfied, and the eigenvalues of
This yields
L + B0 are { 0.1864, 2, 2.4707, 4.3429}. The objective of coordinated
q
( )
x¯ T IN ⊗ Pε( x ) x¯ ≥ η−1ε( x ) x¯ T x.
¯ tracking problem is to design a class of control laws ui for the
follower agents, such that the output signals of two-section loss-
It is from tr( I N ) ( )
⊗ BT Pε( x )B x¯ T IN ⊗ Pε( x ) x¯ ≤ λN−2 that less TLs tracking to that of the leader agent.
It is known for control engineers that make the output signals of
−q T 2ηγ
T
x¯ x¯ ≤ ηε( x ) x¯ ( I ⊗ P ( ))x¯ ≤
N ε x , lossless TLs synchronized tracking is not easy. Under the class of
Nλ nε x
( 2
N ( ) q + 1) control laws (9) with random Gaussian initial conditions and sto-
chastic disturbances, the trajectories of state differences between the
which implies the expectation of the term on the right is bounded.
followers and the leader are shown in Fig. 2. Fig. 3 shows the tra-
Therefore, we have  x¯ T x¯ < ∞, which completes the proof. □
jectories of 2-norm of control inputs of the followers. It is seen that
Remark 4. In view of the proof of Theorem 1, it is clear from (19) the output signals achieve coordinated tracking with a small
that bounded variance of the state difference. We set L = C = 1 in this
example.

xT
(
d IN ⊗ Pε( x )
x=−
) 1 T
x IN ⊗ Pε( x ) x , ( )
dε ε( x )

which implies

(
d IN ⊗ Pε( x ) ) =− 1
dε ε( x )
(
IN ⊗ Pε( x ) . )
In fact, the key issue of the proof is that the relationship between
( )
d IN ⊗ Pε( x ) /dε and (I
N )
⊗ Pε( x ) is analytically discovered.

4. Simulations

In this section, to verify the efficiency of findings, the class of

proposed control laws are implemented on an industrial setting, i.e, Fig. 1. Two-section lossless transmission lines.
8 Q. Wang, C. Sun / ISA Transactions 71 (2017) 3–9

Fig. 2. The trajectories of state differences between the followers and the leader.

Fig. 3. The trajectories of 2-norm of control inputs of the followers.

5. Conclusions the network is connected, we proposed a class of tracking control

algorithms that can always achieve consensus tracking in the ab-
In this paper, we investigated the coordinated tracking control of sence of disturbances and guarantee a bounded variance of the
linear multiagent systems with input saturation and stochastic state difference between the follower agent and the leader in the
disturbances. Under the assumption that the system is ANCBC and present of disturbances. Finally, an industrial setting was introduced
 Q. Wang, C. Sun / ISA Transactions 71 (2017) 3–9
to verify the effectiveness of the proposed tracking control algo-
rithms. Future effort will focus on the coordinated tracking problem
with directed or switching interaction topologies.
References
[1] Abdessameud A, Tayebi A. On consensus algorithms design for double in-
tegrator dynamics. Automatica 2013;49:253–60.
[2] Cao M, Morse A, Anderson B. Reaching a consensus in a dynamically changing
environment: a graphical approach. SIAM J Control Optim 2008;47:575–600.
[3] Chen G, Li C, Dong Z. Parallel and distributed computation for dynamical
economic dispatch. IEEE Trans Smart Grid 2017;8:1026–7.
[4] Chen M, Zhang L, Su H, Chen G. Stabilizing solution and parameter depen-
dence of modified algebraic riccati equation with application to discrete-time
network synchronization. IEEE Trans Autom Control 2016;61:228–33.
[5] Fan M, Chen Z, Zhang H. Semi-global consensus of nonlinear second-order
multi-agent systems with measurement output feedback. IEEE Trans Autom
Control 2014;59:2222–7.
[6] Fax J, Murray R. Information flow and cooperative control of vehicle forma-
tions. IEEE Trans Autom Control 2004;49:1465–76.
[7] Hong Y, Hu J, Gao L. Tracking control for multi-agent consensus with an active
leader and variable topology. Automatica 2006;42:1177–82.
[8] Langovoy M. Algebraic polynomials and moments of stochastic integrals. Stat
Probab Lett 2011;81:627–31.
[9] Li H, Chen G, Huang T, Dong Z. High-performance consensus control in net-
worked systems with limited bandwidth communication and time-varying
directed topologies. IEEE Trans Neural Netw Learn Syst 2017;28:1043–54.
[10] Li H, Chen G, Huang T, Dong Z, Zhu W, Gao L. Event-triggered distributed
average consensus over directed digital networks with limited communication
bandwidth. IEEE Trans Cybern 2016;46:3098–110.
[11] Li Y, Xiang J, Wei W. Consensus problems for linear time-invariant multi-agent
systems with saturation constraints. IET Control Theory Appl 2011;5:823–9.
[12] Li Z, Duan Z, Chen G, Huang L. Consensus of multiagent systems and syn-
chronization of complex networks: a unified viewpoint. IEEE Trans Circuits
Syst I 2010;57:213–24.
[13] Lin X, Zheng Y. Finite-time consensus of switched multiagent systems. IEEE
Trans Syst Man Cybern: Syst 2016.http://dx.doi.org/10.1109/TSMC.2016.
2631659.
[14] Lin Z. Low gain feedback.London: Springer-Verlag; 1998.
[15] Lin Z, Francis B, Maggiore M. Necessary and sufficient graphical conditions for
formation control of unicycles. IEEE Trans Autom Control 2005;50:121–7.
[16] Lin Z, Saberi A. Semi-global exponential stabilization of linear systems subject
to input saturation via linear feedbacks. Syst Control Lett 1993;21:225–39.
[17] Lin Z, Stoorvogel A, Saberi A. Output regulation for linear systems subject to
input saturation. Automatica 1996;32:29–47.
[18] Liu J, Luo W, Yang X, Wu L. Robust model-based fault diagnosis for PEM fuel
cell air-feed system. IEEE Trans Ind Electron 2016;63:3261–70.
[19] Liu S, Xie L, Zhang H. Distributed consensus for multi-agent systems with
delays and noises in transmission channels. Automatica 2011;47:920–34.
[20] Meng Z, Zhao Z, Lin Z. On global leader-following consensus of identical linear
dynamic systems subject to actuator saturation. Syst Control Lett
2013;62:132–42.
[21] Ni W, Cheng D. Leader-following consensus of multi-agent systems under
fixed and switching topologies. Syst Control Lett 2010;59:209–17.
[22] Oksendal B. Stochastic differential equations: an introduction with applica-
tions. Springer Science & Business Media; 2013.
[23] Olfati-Saber R, Fax J, Murray R. Consensus and cooperation in networked
multi-agent systems. Proc IEEE 2007;95:215–33.
9
[24] Pemantle R, Rosenthal J. Moment conditions for a sequence with negative drift
to be uniformly bounded in Lr . Stoch Process Appl 1999;82:143–55.
[25] Qin J, Gao H, Zheng W. Second-order consensus for multi-agent systems with
switching topology and communication delay. Syst Control Lett 2011;60:390–7.
[26] Ren W. On consensus algorithms for double-integrator dynamics. IEEE Trans
Autom Control 2008;53:1503–9.
[27] Ren W, Beard R, Atkins E. A survey of consensus problems in multi-agent
coordination. In: Proceedings the 2005 American control conference; 2005. p.
185–64.
[28] Saberi A, Hou P, Stoorvogel A. On simultaneous global external and global
internal stabilization of critically unstable linear systems with saturating ac-
tuators. IEEE Trans Autom Control 2000;45:1042–52.
[29] Stoorvogel A, Saberi A. On global external stochastic stabilization of linear
systems with input saturation. Automatica 2015;51:9–13.
[30] Stoorvogel A, Saberi A, Shi G. On achieving Lp performance with global internal
stability for linear plants with saturating actuators. In: Proceedings of the 38th
IEEE conference on decision and control, Phoenix, AZ; 1999. p. 2762–67.
[31] Su H, Chen M, Chen G. Robust semi-global coordinated tracking of linear
multi-agent systems with input saturation. Int J Robust Nonlinear Control
2015;25:2375–90.
[32] Su H, Chen M, Lam J, Lin Z. Semi-global leader-following consensus of linear
multi-agent systems with input saturation via low gain feedback. IEEE Trans
Circuits Syst I 2013;60:1881–9.
[33] Su H, Chen M, Wang X. Global coordinated tracking of multi-agent systems
with disturbance uncertainties via bounded control inputs. Nonlinear Dyn
2015;82:2059–68.
[34] Su H, Wang X, Lin Z. Flocking of multi-agents with a virtual leader. IEEE Trans
Autom Control 2009;54:293–307.
[35] Sussmann H, Sontag E, Yang Y. A general result on the stabilization of linear
systems using bounded controls. IEEE Trans Autom Control 1994;39:2411–25.
[36] Teel A. Global stabilization and restricted tracking for multiple integrators
with bounded controls. Syst Control Lett 1992;18:165–71.
[37] Wang Q, Sun C, Xin X. Robust consensus tracking of linear multiagent systems
with input saturation and input-additive uncertainties. Int J Robust Nonlinear
Control 2016.http://dx.doi.org/10.1002/rnc.3690.
[38] Wang Q, Yu C, Gao H. Semiglobal synchronization of multiple generic linear
agents with input saturation. Int J Robust Nonlinear Control 2014;24:3239–54.
[39] Wang Q, Yu C, Gao H. Synchronization of identical linear dynamic systems
subject to input saturation. Syst Control Lett 2014;64:107–13.
[40] Wen G, Duan Z, Ren W, Chen G. Distributed consensus of multi-agent systems
with general linear node dynamics and intermittent communications. Int J
Robust Nonlinear Control 2014;24:2438–57.
[41] Xie G, Wang L. Consensus control for a class of networks of dynamic agents.
Int J Robust Nonlinear Control 2007;17:941–59.
[42] You K, Xie L. Network topology and communication data rate for con-
sensusability of discrete-time multi-agent systems. IEEE Trans Autom Control
2011;56:2262–75.
[43] Yu C, Anderson B, Dasgupta S, Fidan B. Control of minimally persistent for-
mations in the plane. SIAM J Control Optim 2009;48:206–33.
[44] Yu W, Chen G, Cao M, Kurths J. Second-order consensus for multiagent sys-
tems with directed topologies and nonlinear dynamics. IEEE Trans Syst, Man
Cybern - Part B 2010;40:881–91.
[45] Zheng Y, Wang L. Consensus of heterogeneous multi-agent systems without
velocity measurements. Int J Control 2012;85:906–14.
[46] Zheng Y, Wang L. Finite-time consensus of heterogeneous multi-agent sys-
tems with and without velocity measurements. Syst Control Lett
2012;61:871–8.
[47] Zheng Y, Wang L. Consensus of switched multi-agent systems. IEEE Trans
Circuits Syst II 2016;63:314–8.
[48] Zhou B, Duan G, Lin Z. A parametric Lyapunov equation approach to the design
of low gain feedback. IEEE Trans Autom Control 2008;53:1548–54.