Professional Documents
Culture Documents
Adaptive Group Consensus of Multi-Agent in Switching Networks
Adaptive Group Consensus of Multi-Agent in Switching Networks
32 www.erpublication.org
Adaptive Group Consensus of Multi-agent in Networks with Switching Networks
zero eigenvalue with 1N (11 1)T R N as its eigenvector, where xi (t ) R is the position of agent i , ui (t ) R is the
and all the non-zero eigenvalues of L are positive. The control input, and fi ( xi t ) R is the unknown dynamics of
Laplacian L has a simple zero eigenvalue if and only if agent i which is nonlinear. To guarantee the existence of
graph G is connected. unique solution of equation (2), we suppose that fi ( xi t ) is
In this paper, we consider the case that the interconnected continuous in t and Lipschitz in xi (t ) .
graphs of the systems are switching over time. The set of all We assume that there are m virtual leaders in m groups
possible switching graphs is denoted by {G p p P} , where
respectively, and the dynamics of these leaders are described
P is an index set. For describing the time dependence of by
graphs, a piecewise constant switching signal (t ) x 0 ( i )(t ) v0 (i ) (t ) i 1 2 N (3)
[0 ) P is defined. We use G ( t ) to denote the where x0 ( i ) (t ) R is the virtual leaders position of group
underlying graphs at time t on N nodes. Since the (i ) and v0 ( i ) (t ) R the velocity which is also unknown.
Laplacian L of the graph, the neighbors set Ni of agent i ,
Remark 1. The states of all agents are assumed to be in R . It
and the (i j ) th entry aij of A are all vary with time, their is trivial to extend this case to R n by introducing Kronecker
time varying versions are denoted by L ( t ) , aij (t ) and Ni (t ) , product.
respectively. Definition 1. For the N agents in the multi-agent system
(2), we say group consensus is achieved if, there exist
Assumption 1. The switching times of the switching signal controllers ui (t ) i 1 2 N such that
(t ) is finite in any bounded time intervals. lim xi (t ) x0 (i ) (t ) 0 (4)
t
We assume that N agents in multi-agent systems are for any initial condition xi (0) i 01 N .
divided into m groups, and denote the set of the i th group of
agents by Vi (t ) i 1 2 m . Then V (t ) V1(t ) V2(t )
Vm(t ) {1 2 N} and Vk(t ) Vs(t ) (k s) . Let (i) IV. MAIN RESULTS
(i ) Similar to [21], we assume that the unknown nonlinear
V (t ) {i 1 2 m} be a switching function and V (t ) be
dynamics in the system can be linearly parameterized. Then,
the agents set of group (i ) . Then (i) k {1 2 m}
the unknown nonlinear functions fi ( xi t ) , i 1 N and
means that agent i belongs to group k .
the unknown velocity v0 ( i ) (t ) of the leader can be
Since the communication relationships also exist in
different groups, a group in multi-agent system consists of the parameterized as
fi ( xi t ) iT ( xi t )i i 1 2 N (5)
set of inter-agent and the set of intra-agent. Let V((ti)) be the
and
set of inter-agent with V((ti)) V((ti)) . Agent i is called
v0 (i ) (t ) 0T (i ) (t )0 (i ) (6)
(i )
inter-agent if agent i V (t ) , and agent i is called intra-agent where 0 ( i ) (t ) , i ( xi t ) R m
are some basis function
(i ) (i )
if agent i V (t ) \V
(t ) . From [13], we have the following vectors and 0 ( i ) , i R are unknown constant parameter
m
33 www.erpublication.org
International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869 (O) 2454-4698 (P), Volume-4, Issue-3, March 2016
time instant, T0 tk 1 tk T , and T0 T are some positive leaders. When agent i is inter-agent, the state information of
constants. some group leaders is needed.
Suppose that each time interval [tk tk 1 ) k 01 2 , is B. Stability analysis
divided into a finite time interval sequence Let ei (t ) xi (t ) x0 (i ) (t ) , from Lemma 2, we have
[tk0 tk1 ) [tkl tkl 1 ) [tklk 1 tklk ) e(t ) x x 0
where tk tk0 tk 1 tklk for some integer lk 0 , f u v (11)
t l 1
k t 0 l lk 1 , 0 is a positive constant, the
l
k c( L (t ) B ( t ) )e T
lk 1
so-called dwell time. t t t 0
k
1
are the time instants at
k k 0
which the time-varying graph switches, in other words, in where f ( f1 f 2 f N )T , is a regressor
f
each such time subinterval [tkl tkl 1 ) the switching graph
0
G ( t ) is time invariant and connected. matrix, 0 diag{0 (1) 0 (2) 0 ( N ) } ,
Denote the stack column vector ( x1 x2 xN )T of f
f diag{1 2 N } , 0 0 0 col( 0 (i ) ) and
xi (t ) i 1 2 N by col( xi ) . Let x col( xi ) ,
f f f col( i ) , in which 0 (i ) 0 (i ) 0 (i ) and
u col(ui ) , 0 col( 0 (i ) ) , f col( i ) ,
0 col( 0 (i )) , f col( i ) , i 1 2 N . The adaptive i i i . We have the error dynamics of system as
follows:
group consensus algorithm is proposed as follows.
Control Laws: e cH (t ) e T
ui (t ) c aij (t )[ x j (t ) xi (t )] (12)
H (t ) e
j N i ( t ) c
N
L ( t ) B ( t ) , and the matrix H ( t ) of graph
cbi (t )[ xi (t ) x0 (i ) (t )] c lij x0 ( j ) (t )
where H (t )
j 1 G ( t ) has the well-known properties [22] as follows:
0 (i ) (t ) 0 (i ) i ( xi t ) i i V((ti))
T
T
(9) Lemma 4. (1) Matrix H ( t ) has nonnegative eigenvalues;
u (t ) c
i aij (t )[ x j (t ) xi (t )]
j N i ( t )
(2) Matrix H ( t ) is positive definite if and only if graph G ( t )
is connected.
cbi (t )[ xi (t ) x0 ( i ) (t )] 0T ( i ) (t ) 0 ( i )
Let p min ( H p ) , p P be the smallest non-zero
i ( xi t )T i, i V((ti)) \ V((ti)) , eigenvalue of positive semi-definite matrix H p . Then we
where c 0 is a constant number, bi (t ) 0 and let B (t ) be have min min{ p p P} is positive.
a diagonal matrix with diagonal elements We assume that the regression matrix is persistently
{b1 (t ) b2 (t ) bN (t )} . exciting (PE) [23], i.e., there exist two positive reals 0 and
Parameter adaptive laws: such that
0 ( i ) c 0 ( i ) (t ){ aij (t )[ x j (t ) xi (t )]
t 0
t
T d I 0 t 0 (13)
j N i ( t )
bi (t )[ xi (t ) x 0 ( i ) (t )] lij x0 ( j ) (t )}
lim 0 ( i ) 0 ( i ) 0
j 1 (10) t
(15)
i V((ti)) , lim i i 0, i 1 2 N
t
34 www.erpublication.org
Adaptive Group Consensus of Multi-agent in Networks with Switching Networks
U p H pU Tp p diag{1p p2 pN } (18) t
min
2 T
(19) Calculating the time derivative of the function ((t ) t )
ee
0. define in (34) at the time instant t [tk tk 1 ) , we have
Therefore, lim V (t ) V () exists. ((t ) t )
t
c t
Therefore, H p d
T T
t
lim eT ( s)e( s)ds 0 . (22)
t t Q1 Q2
From (12), (16), (19), it follows that e(t ) , 0(t ) , f (t ) From (18), we have
and e are all uniformly bounded, and from the assumption of H p U Tp pU p min I (31)
the theorem, 0 ( i ) , 0 (i ) , i and i are also uniformly Then
bounded. Therefore the function e (t )e (t ) is uniformly T t T
Q2 min TT d (32)
continuous. From Lemma 3, we have lim e( s )T e( s ) 0 , then c t
t
From (13), it follows that
lim e(t ) 0 (23)
T (33)
T
t 1 t T T t T
2 t d t
T
Thus, lim xi (t ) x0 (i ) (t ) 0, i 1 2 N . (t ) (t )
t
for any initial condition (0) , i.e, for any 0 , there exist From the PE condition (14), we have
T 0 such that
min 2 t T1
Q2 (35)
(t ) t T (25) c
Since e , 0 , f , 0 , f are all bounded, there exists
for any initial condition (e(0) (0)) . t T
To prove (25), we first prove that, given any 0 and a positive constant M such that l1 M e d . Then we
t
T 0 , for any e(0) and (0) , there exists t T such that have
35 www.erpublication.org
International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869 (O) 2454-4698 (P), Volume-4, Issue-3, March 2016
switching graph is time invariant and connected in the time
Q1 min 2 t T3 (35)
2c intervals [tkl tkl 1 ), l 0,1, , lk . Under the control law (9) and
From (29), (34) and (35), we have the adaptive law (10), we choose c 08 , 400 . The
simulation results are shown in Fig. 2, 3 and 4. Fig. 2 shows
min 2 0 t T4
((t ) t ) (36)
2c that the states of ten agents converge to the trajectories of the
where T4 max{T1 T2 T3 } . Thus, there exists a time interval two leaders, respectively, and Fig. 3 shows that over time the
tracking errors converge to zero. Fig. 4 shows that the
[tk j , tk j T ) with tk j T4 such that ((t ) t )
parameter convergence is also guaranteed. Note that the
interconnection graph of six agents is connected, and the
min 2 0 holds for t [tk j , tk j T ) due to the
4c regression matrix in this example satisfies the PE
sign-preserving property of continuous function. Integrating condition. Based on Theorem 1, it is expected that both
((t ) t ) from tk j to tk j T and selecting 1 consensus errors and parameter errors converge to zero
globally, uniformly and asymptotically.
min 2T , we have
4c
((tk j ) tk j ) ((tk j T ) tk j T ) 1 (37)
which contradicts (28). So, (26) holds. This completes the
proof of (25).
Because of lim e(t ) 0 , for any 0 there exists a time
t
T5 such that
c
e(t ) t T5 . (38) Fig. 1: The interconnection graph of six agents.
2
Since (25) holds, there exists a time T T5 such that
25
1
(T ) . (39)
2
20
From the initial conditions e(T ) and (T ) , according to
(16), (38) and (39), we have
Ten component s of x(t )
2
15
e(T ) (T ) t T
2
(40)
c
which implies (24). Therefore, the equilibrium point 10
V. SIMULATIONS 0
0 50 100 150 200 250 300 350 400
time(sec)
In this section, an example to illustrate the algorithm is
given to verify our theory results. Consider a system consists Fig. 2: The trajectories of six agents converge to the
of six agents, and assume that iT ( xi t ) [ x sin(t ) x cos(t )]T . trajectories of the two leaders.
Then the nonlinear dynamics of the six agents are 4
fi ( xi t ) [ x sin(t ) x cos(t )]i i 1 2 6, (41) 3.5
and we pick i [ 55 7
2 T
] . 3
The six agents are divided into two groups. We assume that 2.5
Ten component s of e(t )
36 www.erpublication.org
Adaptive Group Consensus of Multi-agent in Networks with Switching Networks
37 www.erpublication.org