没有合适的资源?快使用搜索试试~ 我知道了~
02014IEEE IE0--Coordination for Linear Multiagent Systems With D...
需积分: 0 0 下载量 18 浏览量
2024-04-16
14:30:09
上传
评论
收藏 649KB PDF 举报
温馨提示
试读
11页
02014IEEE IE0--Coordination for Linear Multiagent Systems With Dynamic Interaction Topology in the Leader-Following Framework.pdf
资源推荐
资源详情
资源评论
2412 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 61, NO. 5, MAY 2014
Coordination for Linear Multiagent Systems With
Dynamic Interaction Topology in the
Leader-Following Framework
Jiahu Qin, Member, IEEE, Changbin Yu, Senior Member, IEEE, and Huijun Gao, Senior Member, IEEE
Abstract—This paper considers mainly the leader-following
consensus for multiple agents with general linear system dynamics
under switching topologies. Three different settings are systemat-
ically considered. We first consider the setting that the underly-
ing interaction topologies switch arbitrarily among the possible
weakly connected digraphs and then extend it to a more general
setting that the weak connectivity of the interaction topologies is
kept for some disconnected time intervals with short length due
to the communication constraints among agents. Exponentially,
consensus control is proved to be achieved, and the convergence
rate can be specified as well for both settings in spite of the relaxed
conditions on the system dynamics of each individual agent which
even allow that each agent has exponentially unstable mode, while
for the last case where the weak connectivity is only maintained on
the joint of the interaction topologies, consensus control is proved
to be achieved when the system matrix of each individual agent
satisfies certain stability conditions.
Index Terms—Consensus control, general linear agents, leader-
following consensus, switching topologies.
I. INTRODUCTION
D
ESIGNING a feasible and simple distributed control law
for each agent such that a team of agents as a whole
can perform complex tasks has been the focus of significant
research subjects in systems and control community. This is
partly motivated by its wide applications in such areas as for-
mation control [1]–[3], swarming/flocking [4], [5], consensus/
Manuscript received January 22, 2013; revised April 4, 2013; accepted
May 7, 2013. Date of publication July 16, 2013; date of current version
October 18, 2013. This work was supported in part by the Australian Research
Council through Discovery Project DP130103610, by a Queen Elizabeth II
Fellowship under Grant DP110100538, by the Overseas Expert Program of
Shandong Province, by the National Natural Science Foundation of China
under Grant 61333012, by a Grant from the Shandong Academy of Science
Development Fund for Science and Technology, and by the Pilot Project for
Science and Technology in Shandong Academy of Science.
J. Qin was with the Shandong Provincial Key Laboratory of Computer
Network, Shandong Computer Science Center, Jinan 250014, China, and
also with the Australian National University, Canberra, ACT 0200, Australia
(e-mail: jiahu.qin@anu.edu.au).
C. Yu is with the Australian National University, Canberra, ACT 0200,
Australia. He is also with the National ICT (Information & Communica-
tions Technology) Australia Ltd., Canberra, ACT 2601, Australia, and also
with the Shandong Computer Science Center, Jinan 250014, China (e-mail:
brad.yu@anu.edu.au).
H. Gao is with the Research Institute of Intelligent Control and Systems,
Harbin Institute of Technology, Harbin 150001, China, and also with King
Abdulaziz University (KAU), Jeddah, Saudi Arabia (e-mail: huijungao@
gmail.com; hjgao@hit.edu.cn).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TIE.2013.2273480
synchronization [6]–[13], congestion control in communication
networks, networked control systems [14]–[18], and sensor
networks [19]–[23].
Information consensus, as one of the central problems aris-
ing in multiagent coordination with the collective objective
of reaching agreement about some variables of interests such
as attitude, position, velocity, voltage, etc., has been studied
over the past two decades from various different perspectives;
see, e.g., [24]–[33] and the references therein for an overview
and recent developments on the topic. In these works, the
agent usually has no dynamics in the absence of information
exchange between agents, and it is the exchange of information
only that determines the evolutions of the states of agents. In
addition to these works, research has recently begun to address
the consensus (synchronization) for multiagent systems of a
more general model, i.e., each agent is assumed to take general
linear system dynamics. Different from the integrator agents,
here, the individual dynamics of the agent and the exchange of
information between agents combined will affect the consensus
behavior. Some efforts have been made toward this line; see,
e.g., [34]–[40].
In the discrete-time setting, Tuna [38] studies the leaderless
consensus problem via the distributed state feedback controller,
and You and Xie [40] provide the necessary and sufficient
conditions for the consensusability of multiagent system via
the static state feedback controller, while in the continuous-time
setting, Ma and Zhang [35] consider the consensusability using
the static output feedback controller, and Li et al. [34] consider
the consensus via the use of the distributed observer-type
consensus protocol based on relative output measurements. All
of these works are conducted under a fixed interaction topology.
The consensus of general linear multiagent systems under the
switching interaction topology is investigated in [37] and [39]
via the dynamic feedback controller rather than the static one.
However, as also indicated in [37], conditions on the individual
agent dynamics as well as the interaction topology become very
restrictive when using the static feedback controller. Using the
static feedback controller, Qin et al. [41] study the leaderless
consensus under arbitrarily switching topologies provided that
such topologies are being kept strongly connected and bal-
anced, and Ni and Cheng [36] consider the leader-following
consensus under the undirected interaction topology where the
system matrix of each individual agent needs to satisfy a certain
restrictive assumption and the weighting factors associated
with the interactions among agents are assumed to vary in a
piecewise constant mode.
0278-0046 © 2013 IEEE
QIN et al.: COORDINATION FOR LINEAR MULTIAGENT SYSTEMS WITH DYNAMIC INTERACTION TOPOLOGY 2413
In this paper, we consider the convergence of leader-
following consensus algorithms for general linear agents un-
der the switching topology by expanding on our preliminary
work reported in [41]. More specifically, with the balanced
underlying i nteraction topology condition among the follower
agents, the leader-following consensus is studied regarding
the following three aspects: 1) the case that the underlying
interaction topology among the leader and the follower agents,
for example,
¯
G
σ(t)
, i s being kept weakly connected; 2) the case
that the weak connectedness of
¯
G
σ(t)
can only be guaranteed at
some disconnected time intervals, where the length of such time
intervals may be a very small number due to the communication
constraints among agents; and 3) the case that
¯
G
σ(t)
is repeat-
edly jointly weakly connected. For cases 1) and 2), we will
show that consensus can be achieved exponentially fast even if
each individual agent may have exponentially unstable mode,
while consensus control for case 3) requires somewhat restric-
tive stability conditions on each individual agents. It is worth
mentioning that, for all such cases, this paper takes account
of the general case that the weighting factors modeling the
communications among agents are assumed to be dynamically
changing in an infinite set as opposed to the finite case in most
of the existing literature such as [36], [39], and [28]. Indeed,
in most practical settings, the weight factors may be time
varying in a piecewise continuous mode rather than piecewise
constant [8].
The rest of this paper is structured as follows. Section II
describes the existing graph theory notions applied as well as
the problem statement in this paper while Section III aims
to analyze the leader-following consensus in various different
settings, including the setting that the weak connectivity can be
guaranteed for
¯
G
σ(t)
, the setting that such a weak connectivity
can be kept for a short time across each time interval, and the
setting that
¯
G
σ(t)
is repeatedly jointly weakly connected. Some
concluding remarks are made in Section IV.
Basic Notation and Notions: The f ollowing notations will be
used throughout this paper. Denote by M>0(M<0) that M
is symmetric positive (negative) definite and by M ≥ 0(M ≤
0) that M is symmetric and positive (negative) semidefinite. If
all the eigenvalues of M are real, then denote by λ
max
(M)
and λ
min
(M) the maximum and minimum eigenvalues of
M, respectively. Denote by diag{A
1
,A
2
,...,A
n
} the block
diagonal matrix with its ith main diagonal matrix being a
square matrix A
i
,i=1,...,n. A matrix is called nonnegative
(positive) whenever all its elements are nonnegative (positive).
II. P
RELIMINARIES AND PROBLEM STATEMENT
Let G =(V,ε,A) be a weighted digraph of order N with
a finite nonempty set of nodes V = {1, 2,...,N}, a set of
edges ε ⊂V×V, and a weighted adjacency matrix A =
[a
ij
] ∈ R
N×N
in which a
ij
> 0 ⇔ (j, i) ∈ ε. Moreover, we as-
sume a
ii
=0,i=1,...,N. Denote by L(G)=L =[
ij
] the
Laplacian matrix associated with G, where
ij
= −a
ij
,i= j,
and l
ii
=
N
k=1,k=i
a
ik
. Con versely , any given N × N matrix
L =[
ij
] with nonpositive off-diagonal elements and satisfying
L1
n
=0, which shall be called graph Laplacian in the sequel
for convenience, can be deemed as a Laplacian matrix of a
weighted digraph, for example, G(L), which is defined as
follows: G(L) is a digraph with node set {1, 2,...,N}, and
there is an edge in G(L) from j to i (j = i) if and only if
ij
>
0. Digraph G is called balanced if and only if 1
T
L(G)=0[28].
Given a digraph G, denote by G the mirror graph [28] of G.IfG
is a balanced digraph, then L(G)+L
T
(G)/2 is the Laplacian
matrix of G, i.e., L(G )=L(G)+L
T
(G)/2 (see [28, Th. 7]).
A digraph is called strongly connected if any two distinct
nodes of the graph can be connected by a directed path, while
it is called weakly connected if replacing all of its directed
edges with undirected edges produces a connected graph. For
an undirected graph G with the Laplacian matrix being L,
denote by λ
2
(L(G)) the algebraic connectivity of G [42] which
is defined as
λ
2
(L(G)) = min
x=0,1
T
x=0
x
T
L(G)x
x
T
x
. (1)
Note that λ
2
(L(G)) > 0 if and only if undirected graph G is
connected [42].
Consider a multiagent system consisting of N agents, in-
dexed as agents 1, 2,...,N, and a leader with index 0 which
evolves according to the following dynamics:
˙x
0
= Ax
0
. (2)
Each agent is regarded as a node in a digraph G, the dynamics
of which is described by
˙x
i
= Ax
i
+ Bu
i
,i=1,...,N (3)
where x
i
∈ R
n
is the ith agent’s state and u
i
∈ R
m
is the
control input for agent i which uses only the state information
from its neighboring agents.
In this paper, we will consider the following control law for
agent i, i =1, 2,...,N:
u
i
(t)=K
j∈N
i
(t)
a
ij
(t)(x
j
(t) − x
i
(t))
+ Kd
i
(t)(x
0
(t) − x
i
(t)) (4)
where N
i
(t) is the set of neighbors of agent i at time t,
K ∈ R
m×n
is a feedback matrix to be designed, and d
i
(t) ∈
{0}∪[d
,
¯
d],
¯
d>d> 0. Agent i can receive information from
the leader at time t if and only if d
i
(t) ≥ d. It is further assumed
that all the nonzero and hence positive weighting factors are
chosen from [α
, ¯α], i.e., a
ij
(t) ∈ [α, ¯α], where 0 <α< ¯α,if
and only if j ∈ N
i
(t).
Let G be the set of all possible balanced interaction topolo-
gies among the N agents, i.e.,
G = {G|G (V,,A =[a
ij
])
is balanced and a
ij
∈{0}∪[α, ¯α]}
and σ(t):[0, ∞) →I
G
be a switching signal whose image at
time t is the index associated with a graph in G.
Remark 1: Different from most of the existing literature (see,
e.g., [28], [36], and [39]) in which the index set I
G
is finite,
2414 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 61, NO. 5, MAY 2014
G here is an infinite set as the weighting factors are allowed
to be dynamically changing in an infinite set {0}∪[α
, ¯α], and
thus, I
G
is also infinite.
Definition 1: The leader-following consensus of system (2)
and (3) is said to be reached by employing control law (4) if,
for any initial state x
i
(0), i =0, 1,...,N, there holds
lim
t→∞
x
i
(t) − x
0
(t) =0,i=1,...,N.
III. C
ONSENSUS ANALYSIS UNDER SWITCHING TOPOLOGY
The following is a standard assumption made on matrix pair
(A, B) when considering the consensus for linear multiagent
systems; see, e.g., [36]–[38].
Assumption 1: The pair (A, B) is stabilizable.
Remark 2: Assumption 1 allows that A has exponentially
unstable mode, i.e., A may have eigenvalues with positive real
parts.
A. Consensus Under Weakly Connected Topologies
Denote the state error between agent i and the
leader as e
i
(t)=x
i
(t) − x
0
(t), i =1, 2,...,N.Let
e(t)=[e
T
1
(t),e
T
2
(t),...,e
T
N
(t)]
T
, and then, from (3) and
(4), one obtains the following compact form of the error
dynamics:
˙e(t)=
I
N
⊗ A −
L
σ(t)
+ D
σ(t)
⊗ BK
e(t) (5)
where D
σ(t)
=diag{d
1
(t),d
2
(t),...,d
N
(t)} and L
σ(t)
is the
Laplacian matrix of G
σ(t)
.
Definition 2: Given any digraph G with node set
V = {1,...,N} and nonnegative diagonal matrix
D =diag{d
1
,d
2
,...,d
N
}, denote by
¯
G the induced digraph
from G and D. More specifically,
¯
G consists of digraph G,
node 0, and the directed edges from the node 0 to the nodes
in G, where the existence of such edges follows the following
rule: There is a directed edge from node 0 to node, for example,
i,inG if and only if d
i
> 0, i =1,...,N.
To proceed further, we also need the following results con-
cerning the induced graph
¯
G.
Lemma 1 [43, Lemma 2]: Given any digraph G,
¯
G is weakly
connected if and only if there exists at least a directed edge
from node 0 to each of the weakly connected component of G.
Furthermore, if G is balanced, then L(G)+L
T
(G)/2+D>
0 if and only if
¯
G is weakly connected.
Lemma 2: Let
Π={L + D =[
ij
]+diag{d
1
,...,d
N
}
|L is a graph Laplacian, 1
T
L =0;
−
ij
∈{0}∪[α, ¯α],d
i
∈{0}∪[d,
¯
d],
i, j =1,...,N, i = j; and
¯
G,
the digraph induced by G(L) and D,
is weakly connected} .
Then, there exists a μ>0 such that λ
min
(
¯
L +
¯
L
T
/2) ≥ μ,
∀
¯
L ∈ Π.
Proof: Denote by Ω the set of all N × N nonnegative
matrices with zero diagonal elements. It is clear that there are
only finite different types of matrices (two nonnegative matrices
P
1
and P
2
are said to be of the same type [44], P
1
∼ P
2
,ifthey
have zero elements and positive elements in the same places)
in Ω and all the digraphs associated with the matrices having
the same type in Ω are also with the same topological structure.
Note that Ω can be partitioned by the equivalence relation ∼.
Let [A]:={B ∈ Ω|B ∼ A} denote the equivalence class to
which A (A ∈ Ω) belongs. Without loss of generality, denote
by [A
1
], [A
2
],...,[A
m
] all the equivalence classes in Ω, and
Π
k
1
=
L =[
ij
]|L is a graph Laplacian, 1
T
L =0;
−
ij
∈{0}∪[α, ¯α],i,j =1,...,N,i= j;
− L +diag{
11
,...,
NN
}∼A
k
} .
Note that the set of all N × N matrices can be viewed as the
metric space R
N
2
. In what follows, we first prove that Π
k
1
,
k =1, 2,...,M, is compact in R
N
2
. We only prove that Π
1
1
is a compact set, the compactness of Π
k
1
, k =2,...,M,fol-
low exactly the same way. In fact, each L =[
ij
] in Π
1
1
can
be viewed as a vector [
1,1
,...,
1,N
,
2,1
,...,
2,N
,
N,1
,...,
N,N
] in R
N
2
. To prove that Π
1
1
is compact in the Euclidean
space R
N
2
, it is equivalent to prove that it is a closed and
bounded set. Let Υ
1
={[
ij
]|−
ij
∈{0}∪[α, ¯α],i,j=1,...,
N,i = j;
ii
∈ [0,N¯α],i=1,...,N}, Υ
2
={[
ij
]|
N
j=1
ij
=
0,i=1,...,N}, and Υ
3
={[
ij
]|
N
i=1
ij
=0,j=1,...,N}.
It can be easily derived from the definition of the Laplacian
matrix that Υ
1
∩ Υ
2
∩ Υ
3
is the set consisting of all graph
Laplacian L satisfying 1
T
L =0and the off-diagonal elements
of L are chosen from the set 0 ∪ [α
, ¯α]. We first prove that
Υ
1
∩ Υ
2
∩ Υ
3
is closed and bounded in R
N
2
. In fact, Υ
1
is
closed and bounded, and the sets Υ
2
and Υ
3
are closed but
unbounded. The former argument holds since it is the product
space of N
2
closed and bounded sets in R
1
. For the latter
argument,we only prove in the following that the set Υ
2
is
closed, and a similar proof can be derived for that of Υ
3
.Let
S
i
={[
i,1
,...,
i,N
]|[
i,1
,...,
i,N
] is the vector taken from
the i−th row of [
ij
] ∈ Υ
2
} ,i=1,...,N.
Then, Υ
2
= S
1
× S
2
×···×S
N
. It is clear that Υ
2
is a closed
set in R
N
2
if each S
i
,i=1, 2,...,N is closed in R
N
.In
order to prove that S
i
is closed, we introduce the following
continuous multivariate function:
f : R
N
→ R
1
,f(x):=
N
i=1
x
i
, ∀x =[x
1
,x
2
,...,x
N
]∈ R
N
.
Since f is continuous and {1} is a closed set in R
1
, f
−1
({1})
is closed in R
N
, i.e., each set S
i
,i=1,...,N is closed in R
N
.
Therefore, Υ
1
∩ Υ
2
∩ Υ
3
is closed and bounded.
剩余10页未读,继续阅读
资源评论
NaClick
- 粉丝: 0
- 资源: 1
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功