L. Zhao et al. / Applied Mathematics and Computation 312 (2017) 23–35 25
Lemma 2 [35,37] . Consider the nonlinear system
˙
x = f (x (t)) , x (0) = 0 , f (0) = 0 , x ∈ R
n
(2)
where f : U
0
→ R
n
is continuous in an open neighborhood U
0
of the origin. Assume that system (2) possesses a unique solution in
forward time for all initial conditions. Assume that there exists a Lyapunov function V ( x ), scalars α, β, p, q, k ∈ R
+
, pk < 1 , gk > 1 ,
and 0 < ϑ < ∞ , such that
˙
V (x ) ≤−(αV (x )
p
+ βV (x )
g
)
k
+ ϑ, then the trajectory of system (2) is practical fixed-time stable, and
the residual set of the solution of system (2) is given by { lim
t→ T
x | V (x ) ≤min { α
−
1
p
(
ϑ
1 −θ
k
)
1
kp
} ,β
−
1
g
(
ϑ
1 −θ
k
)
1
kg
}} , where θ satisfies 0 < θ ≤
1
. The setting time is bounded as T ≤
1
α
k
θ
k
(1 −pk )
+
1
β
k
θ
k
(gk −1)
.
Lemma 3 [37] . If v ∈ R
+
and v > 1 , then for any x, y ∈ R , we have | x + y |
v
≤ 2
v −1
| x
v
+ y
v
| .
Lemma 4 [22] . For x
i
∈ R , i = 1 , 2 , . . . , n, 0 < p ≤ 1 ,
n
i =1
| x
i
|
p
≤
n
i =1
| x
i
|
p
≤ n
1 −p
n
i =1
| x
i
|
p
.
3. Main results
3.1. Error dynamics
The local neighborhood state errors for the i th follower are denoted as
e
1 i
=
n
j=1
a
ij
(q
i
− q
j
) + b
i
(q
i
− q
d
) , e
2 i
=
n
j=1
a
ij
(
˙
q
i
−
˙
q
j
) + b
i
(
˙
q
i
−
˙
q
d
) (3)
then we have
e
1
= (H I) θ
1
, e
2
= (H I) θ
2
(4)
where e
1
= [ e
T
11
, . . . , e
T
1 n
]
T
, e
2
= [ e
T
21
, . . . , e
T
2 n
]
T
, θ
1
= [ θ
T
11
, . . . , θ
T
1 n
]
T
, θ
2
= [ θ
T
21
, . . . , θ
T
2 n
]
T
, θ
1 i
= q
i
− q
d
, θ
2 i
=
˙
q
i
−
˙
q
d
. From the
definitions of (4) , we know that e
1
and e
2
can not be directly used for control law design for each agent, and can only
be used for stability analysis.
3.2. Fixed-time terminal sliding mode (FTTSM) design
Now, define the FTTSM vector as
S = [ s
T
1
, ... , s
T
n
]
T
(5)
where s
i
= [ s
i 1
, . . . , s
ip
]
T
∈ R
p
is given by
s
i
= e
2 i
+ α
i
(e
1 i
) , i ∈ V (6)
with α
i
(e
1 i
) = [ α
i 1
(e
1 i 1
) , ... , α
ip
(e
1 ip
)]
T
, and
α
iχ
(e
1 iχ
) =
⎧
⎨
⎩
sig (σ
1 i
sig (e
1 iχ
)
m
1
+ σ
2 i
sig (e
1 iχ
)
n
1
)
k
1
, if
¯
s
iχ
= 0 or
¯
s
iχ
= 0 ,
| e
1 iχ
| > φ
l
1 i
e
1 iχ
+ l
2 i
sig (e
1 iχ
)
2
, if
¯
s
iχ
= 0 , | e
1 iχ
| ≤ φ
(7)
χ = 1 , . . . , p,
¯
s
i
= [
¯
s
i 1
, . . . ,
¯
s
ip
]
T
,
¯
s
i
= e
2 i
+ sig (σ
1 i
sig (e
1 i
)
m
1
+ σ
2 i
sig (e
1 i
)
n
1
)
k
1
, m
1
, n
1
, k
1
∈ R
+
, 0 < m
1
k
1
< 1 , n
1
k
1
> 1 , l
1 i
=
(2 −k
1
)(σ
1 i
φ
m
1
−
1
k
1
+ σ
2 i
φ
n
1
−
1
k
1
)
k
1
, l
2 i
= (k
1
− 1)(σ
1 i
φ
m
1
−
2
k
1
+ σ
2 i
φ
n
1
−
2
k
1
)
k
1
, φ > 0 .
Remark 1. Note that the FTTSM (6) has the form of modified terminal sliding mode (TSM) in [42] for k
1
= 1 , n
1
= 0 , more-
over, (6) coincides with the modified fast TSM in [23,24] for k
1
= 1 . Actually, for
¯
s
iχ
= 0 , the FTTSM is switched to the
general linear sliding mode when e
1 i χ
enters the region | e
1 i χ
| ≤ φ. Thus, the singularity problem can be avoided for the
FTTSM (6) . Moreover, l
1 i
and l
2 i
are chosen such that α
i χ
and its time derivative continuous.
Remark 2. Compared with the TSM with the settling time estimated dependent on the initial conditions of systems in
[23–25] , the proposed FTTSM can guarantee the fixed-time convergence and the setting time can be estimated regardless to
initial conditions, which is more adopt to practical applications.
From (6) , we can obtain the following equation
˙
s
i
+ s
i
=
˙
e
2 i
+ ˙ α
i
+ e
2 i
+ α
i
(8)