Vous êtes sur la page 1sur 12

Nonlinear Dyn (2018) 92:1845–1856

https://doi.org/10.1007/s11071-018-4167-4

ORIGINAL PAPER

Adaptive NN finite-time tracking control of output


constrained nonlinear system with input saturation
Shiyi Zhao · Hongjing Liang · Peihao Du ·
Suwen Qi

Received: 15 October 2017 / Accepted: 10 February 2018 / Published online: 3 March 2018
© Springer Science+Business Media B.V., part of Springer Nature 2018

Abstract This paper considers the finite-time track- small neighborhood of zero and remains in the prede-
ing control problem for the strict-feedback nonlinear fined compact sets. The effectiveness of the proposed
continuous systems involving input saturation and out- scheme is confirmed via two simulation examples.
put constraints. A sequence of desired and auxiliary vir-
tual control signals and real control input is designed to Keywords Finite-time control · Barrier Lyapunov
derive a representation of the system estimation errors functions (BLFs) · Output constraints · Input
and stabilize the system. The proposed approach is fur- saturation · Neural networks (NNs)
ther developed via a finite-time stability theory, barrier
Lyapunov function, and neural network approximation
scheme to achieve an expected performance of the con- 1 Introduction
sidered system. According to the proposed scheme, we
solve the finite-time tracking control problem of the For the control systems with nonlinear parameters,
nonlinear systems with input saturation. Then, a the- fuzzy logic system [1–4], adaptive backstepping con-
orem is provided to address that all the signals and trol [5–8] and sliding mode control [3,9–12] have been
system states are bounded, and the system output is extensive studied to obtain the desired system perfor-
driven to track the reference signal in a finite time to a mance. Adaptive backstepping control technology is
particularly prominent on account of its adjustment for
S. Zhao · P. Du the adaptive parameters in the procedure of controller
School of Mathematics and Physics, Bohai University,
Jinzhou 121013, Liaoning, China
design by online learning [13–15]. On this basis, the
e-mail: shiyizhao2016@163.com estimates of the unknown nonlinear terms are disposed
by fuzzy logic systems (FLSs) [16–21] and neural net-
P. Du
e-mail: peihaodu2017@163.com works (NNs) [22–28]. To mention a few, the authors
in [29] proposed an adaptive NN consensus control
H. Liang (B) method for a class of nonlinear multiagent systems with
College of Engineering, Bohai University, Jinzhou 121013,
Liaoning, China
state time delay. A direct adaptive fuzzy controller for
e-mail: lianghongjing99@163.com a class of SISO uncertain affine nonlinear systems was
developed in [30]. In [31], Park et al. adopted the Lya-
S. Qi (B) punov approach and solved the global asymptotic sta-
School of Biomedical Engineering, Health Science Center,
Shenzhen University, Shenzhen 518060, Guangdong,
bility of equilibrium for continuous bidirectional asso-
China ciative memory NNs of neutral type. The authors in [32]
e-mail: qisuwen@szu.edu.cn solved the adaptive control problem of strict-feedback

123
1846 S. Zhao et al.

nonlinear systems by using multilayer neural networks results, the finite-time control method of input sat-
(MNNs). The stability analysis problem for neural net- uration system is more practical and meaningful in
works with time-varying delays was developed in [33]. the real engineering.
It is to be emphasized that there are many practi- (2) Constructively and theoretically, by contrast with
cal factors [34–43] need to be considered in the engi- most existing controller design procedure, we
neering. Output constraints and input saturation are design the desired and unknown control signals,
two of important factors, which will easily destabi- the virtual control signals and real control input by
lize stability and get bad performance of the systems. using neural network approximation, respectively.
For the output constraints, the authors in [44] inves- And we adopt NN approximation to approximate
tigated the control problem for SISO nonlinear sys- the intermediate control signals rather than most
tems in strict-feedback form with output constraints. methods to approximate the unknown functions,
In [45], the output tracking control for strict-feedback and therefore, the design parameters is greatly
nonlinear systems with time-varying output constraint reduced.
was presented. In addition, [11,46,47] proposed sev- (3) Different from the existing adaptive control
eral approaches to deal with input saturation occurred approaches, which can only ensure the tracking
in the system. To mention a few, the authors in [48] pro- error closing to zero in infinite time, the designed
posed an approximation-based adaptive tracking con- controller can efficiently insure the system perfor-
trol approach for a class of MIMO nonlinear systems mance in a finite time by adopting a finite-time sta-
with input saturation. In [49], for the uncertain nonlin- bility theory.
ear system in a strict-feedback form with an input satu-
The remainder of this paper is organized as follows.
ration and the unknown external disturbances, Mou et
In Sect. 2, the system description and the preparatory
al. presented a dynamic surface control (DSC) method.
work are proposed. Then, the desired and unknown
The authors in [50] developed two new robust adap-
control signals, the virtual control signals, and real con-
tive control algorithms by using a Nussbaum function
trol input are designed in Sect. 3. Two simulation exam-
and backstepping approaches. Moreover, for the finite-
ples results are given in Sect. 4 to prove the availability
time tracking control problem [51,52], [53] proposed
of the presented control technique. In the end, the con-
the new results on global finite-time stabilization for
clusions are made in Sect. 5.
a class of switched strict-feedback nonlinear systems.
The authors in [54] presented a finite-time tracking
issue for nonlinear quantized systems with unmeasur- 2 Problem formulation and preliminaries
able states. In [55], a continuous finite-time control
scheme is studied for rigid robotic manipulators using Consider the nonlinear strict-feedback system as fol-
a new form of terminal sliding modes. However, input lows

saturation and output constraints are not considered ⎨ ẋi = f i (x̄i ) + gi (x̄i ) xi+1 , i = 1, 2, . . . , n − 1
together in the continuous nonlinear systems, and it ẋn = f n (x̄n ) + gn (x̄n ) u (v (t)) , (1)

can also be known as the research for the system con- y = x1 ,
sidered input saturation and output constraints can be
where x = [x1 , x2 . . . , xn ]T , x̄i = [x1 , x2 . . . , xi ]T ∈
further studied so that can be better applied in the actual
R i are the state variables, y ∈ R represents the system
project, which motivates this study.
output and satisfies |y| < kc1 , ∀t > 0, kc1 is a pos-
The primary work of this paper is to design the finite-
itive constant. f i (x̄i ) and gi (x̄i ) , i = 1, 2, . . . , n are
time tracking controller of the nonlinear systems with
the unknown smooth nonlinear functions, and u (v (t))
an input saturation and the output constraints. Com-
expresses an input saturation defined as ( [56])
pared with many extant results, this paper has the main ⎧
advantages as follows. ⎨ u max , v ≥ u max
u = sat (v (t)) = v, u min < v < u max (2)

(1) It is the first time handling the finite-time tracking u min , v ≤ u min
control problem for the output constrained strict- where v is the input signal of the input saturation non-
feedback nonlinear systems with input saturation. linearity, u max > 0 and u min < 0 are the unknown
Compared with the general infinite-time control constants.

123
Adaptive NN finite-time tracking control 1847

For (2), a smooth piecewise function is adopted to Lemma 1 [54] Specific to the system ς̇ = f (ς, u),
denote the approximation of the input saturation func- for the function V (ς ) (smooth and positive definite),
tion and ⎧
defined as if there are some of the scalars c > 0, 0 < l < 1 and
 
⎨ u max · tanh v , v ≥ 0  > 0, so that
g (v) =  max
u
⎩ u min · tanh v , v < 0 V̇ (ς ) ≤ − cV l (ς ) + , t ≥ 0 (9)
u min

⎪ u
v − v
−e u max then, the nonlinear system ς̇ = f (ς, u) is SGPFS.
⎨ u max · e max v − v
, v≥0
e u max +e uvmax
= v −u (3) Lemma 2 [58] For any real variables z, ς , there are
⎪ u
⎩ u min · e minv
−e min
, v<0
− v positive constants μ, θ, ι, the inequality holds as fol-
e u min +e u min

and then, sat in (2) can be represented as lows:


u = sat (v) = g (v) + d (v) μ θ −μ
(4) |z|μ |α|θ ≤ ι |z|μ+θ + ι θ |α|μ+θ . (10)
μ+θ μ+θ
where d (v) = sat (v) − g (v) is a function bounded as
Lemma 3 [54] For z j ∈ R, k = 1, 2, . . . , n, 0 < l <
|d (v)| = |sat (v) − g (v)|
1, the inequality relation sets up as follows
≤ max {umax [1 − tanh (1)] , umin [tanh (1) − 1]} n l n l
= D. (5)
 
n
 l
 
z j  ≤ z j  ≤ n 1−l z j  . (11)
Otherwise, based on the theorem of the mean which k=1 k=1 k=1
shows there is a constant μ (0 < μ < 1) to make
Lemma 4 [59] RBF NN is employed to approximate
g (v) = g (v0 ) + gvμ (v − v0 ) (6) the smooth unknown nonlinear functions. For any con-
where tinuous and unknown smooth function f (X ) on the
∂g (v)  compact set
and an expected precision , there is a
gvμ = v=vμ ,
∂v RBF NN ΦiT ξi (X i ) so that
vμ = μv + (1 − μ) v0 .
f (X ) = Φ T ξ (X ) + δ (X ) , |δ (X )| ≤ .
Remark 1 Mean value theorem: For a function f (x),
if it is continuous in the interval [a, b] and derivable in Assumption 1 [60] The function gi (x̄i ) is unknown
the interval (a, b), then there exists at least one point but bounded as
 (a <  < b) in the interval such that: f (b)− f (a) = 0 < bm ≤ |gi (x̄i )| ≤ b M < ∞, i = 1, 2, . . . , n (12)

f ()(b − a) holds.
It is obvious that gi (x̄i ) is strict positive or strict neg-
By selecting v0 = 0, (6) can be rewritten as ative. In general, to assume that
g (v) = gvμ v (7)
0 < bm ≤ gi (x̄i ) ≤ b M .
Substituting (4) and (7) into (1), we have a new expres-
sion of the considered system as Assumption 2 [56] Consider gvμ in (7), there is an
⎧ unknown constant gm such that
⎨ ẋi = f i (x̄i ) + gi (x̄i ) xi+1 , i = 1, 2, . . . , n − 1
ẋn = f n (x̄n ) + gn (x̄n ) gvμ v (t) + gn (x̄n ) d (v) , 0 < gm ≤ gvμ ≤ 1. (13)

y = x1 ,
(8) Remark 2 Based on Assumptions 1 and 2, it can be
supposed that
In the following part, we first introduce some exist-
ing lemmas and assumptions for designing the adaptive 0 < b ≤ gi (x̄i ) , i = 1, 2, . . . , n − 1
NN controller. 0 < b ≤ gn (x̄n ) gvμ (14)
Definition 1 [57] The equilibrium position ς = 0 of where b = min {bm , bm gm } is an unknown constant.
the nonlinear system ς̇ = f (ς, u) is semi-global prac-
tical finite-time stable (SGPFS) if for all ς (t0 ) = ς0 , Assumption 3 The reference signal yd (t), and its nth-
there are ε > 0 and a deposit time T (ε, ς0 ) < ∞ such order time differential coefficients are both continuous
that ς (t) < ε, for all t ≥ t0 + T . and bounded.

123
1848 S. Zhao et al.

Assumption 4 [61] yr (t) and its jth-order differen- where η1 is the design parameter, kb1 = kc1 − B0 , In
( j)
tial coefficients yr (t), j = 1, . . . , n are assumed
 the set
e1 = {e1 : |e1 | < kb1 }, V1 is continuous. Then,
 ( j) 
to satisfy −yr ≤ yr (t) ≤ ȳr and yr (t) ≤ B j we obtain V̇1 as
where yr , ȳr¯, B0 , B1 , . . . , Bn are positive constants
[ f 1 (x1 ) + g1 x2 − ẏd ] − θ̃1 θ̂˙1
e1 1
¯ B0 ≤ kc1 , ȳr ≤ B0 ≤ kc1 . V̇1 =
and yr ≤ − e1
2
kb1 2 η1
¯ 2
e1   e1
= 2 ¯
f 1 + g1 x2 − g1 2
3 The controller design-based barrier Lyapunov kb1 − e12 2 − e2
kb1 1
functions 1 ˙
− θ̃1 θ̂1 (21)
η1
In this section, for system (8), a backstepping-based
scheme will be presented. It contains n steps. At Step where a new function is defined as
i (i = 1, 2, . . . , n), the desired and unknown control f¯1 = f 1 + g12 2
e1
− ẏd (22)
signals are firstly constructed to steady the first i sub- kb1 − e12
systems in theory. After that, a NN ΦiT ξi (X i ) will be
Design the virtual control signal as
adopted to approximate these unknown control signals,  
and then, the virtual controls signal can be constructed. e12l−1
The real controller v will be structured at Step n. To α̂1 = −g1−1
f¯1 +   (23)
2 − e2 l−1
kb1
proceed this course, we define a constant as 1

1 where l = 2n−1
2n+1 is the design parameter. For (23), (21)
θi = Ni Φi 2 , i = 1, 2, . . . , n (15) can be represented as
b
 l
as an unknown positive constant and θ̂i is the estimate e12 e1  
of θi , then, for any given bounded initial condition V̇1 = − 2 + 2 g1 x2 − α̂1
kb1 − e12 kb1 − e1
2
θ̂i (t0 ) ≥ 0, we have θ̂i (t) ≥ 0, for any t ≥ t0 , and 2
θ̃i = θi − θ̂i , where Ni is the number of neural network
− θ̃1 θ̂˙1
e1 1
nodes and satisfies ξiT ξi ≤ Ni . − g12 (24)
kb1 − e1
2 2 η 1
At Step i (i = 1, 2, . . . , n), the feasible virtual con-
trol signal is constructed as However, α̂1 cannot be used directly in practice. Thus,
1 e1 for any given 1 > 0, there exists a NN Φ1T ξ1 (X 1 ) such
α1 (X 1 ) = − 2 − e2 1
θ̂ (16) that
2a12 kb1 1
1 α̂1 = Φ1T ξ1 (X 1 ) + δ1 (X 1 ) |δ1 (X 1 )| ≤ 1 (25)
αi (X i ) = − 2 ei θ̂i , i = 2, . . . , n − 1 (17)
2ai where 1 is a positive constant. Then, we have
1
v = − 2 en θ̂n (18) e1
2an − 2 g1 α̂1
kb1 − e12
where e1 = x1 − yd , ei = xi − αi−1, X i = e1 e1
T =− 2 g1 Φ1T ξ1 − 2 g1 δ1
x̄iT , θ̂¯iT , ȳd
(i)T T ¯ (i)
, θ̂i = θ̂1 , θ̂2 , . . . , θ̂i , ȳd expresses kb1 − e1 2 kb1 − e12
the vector of yd and up to its ith-order time derivatives.  2 2
b e1 1 2 e1
The virtual control signals α̂i will be described at Step ≤ 2 2 − e2
θ1 + g1 2 − e2
2a1 kb1 2 kb1
i. 1 1
Step 1: Define the error vector as e1 = x1 − yd and b2M 2 1 2
+ a +  (26)
it has 2 1 2 1
ė1 = f 1 (x1 ) + g1 x2 − ẏd (19) with g1 > 0 and θ̂1 ≥ 0, the following inequality hold
Define a positive definite barrier Lyapunov function as
 2
1 k2 1 2 e1 b e1
V1 = log 2 b1 2 + θ̃ (20) g α ≤− 2
2 − e2 1 1 2 − e2
θ̂1 (27)
2 kb1 − e1 2η1 1 kb1 1 2a1 kb1 1

123
Adaptive NN finite-time tracking control 1849

Substituting (26) and (27) into (24), one has 1


V̇1 ≤ − σ1,1 V1l + σ1,2 + e22 (35)
 l 2
e12 e1  l 
V̇1 ≤ − 2 + 2 g1 (x2 − α1 ) where σ1,1 = min 2 , k1 , σ1,2 = k1
(1 − l) ι +
kb1 − e12 kb1 − e12 η1
b2M
2 k1 2
2η1 θ1 + 2 a12 + 21 12 .
1 2 e1 b2M 2 1 2
− g1 + a +  Step i (i = 2, . . . , n − 1): Defining the error variable
2 2 − e2
kb1 2 1 2 1

1
2 ⎤ as ei = xi − αi−1 , one has

− θ̂˙1 ⎦
1 ⎣ bη1 e1 ėi = f i (x̄i ) + gi xi+1 − α̇i−1 (36)
+ θ̃1 (28)
η1 2a12 kb1 2 − e2
1
where
Define variable
e2 = x2 − α1 (29)

i−1
∂αi−1

i−1
∂αi−1
α̇i−1 = f k (x̄k ) + gk (x̄k ) xk+1 + i−1 ,
based on the Young’s inequality, one has ∂ xk ∂ xk
2 k=1 k=1

e1 1 2 e1 1

i−1
∂αi−1 ˙
i
∂αi−1 (k+1)
g1 e2 ≤ g1 + e22 . (30) i−1 = θ̂k + y .
kb1 − e1
2 2 2 kb1 − e1
2 2 2
k=1 ∂ θ̂ k
(k) r
k=0 ∂ yr
Design the adaptive law as
2 Select a following positive Lyapunov function as
˙θ̂ = η1 b e1
− k1 θ̂1 (31) 1 2 1 2
1 2 − e2
2a12 kb1 Vi = ei + θ̃ (37)
1 2 2ηi i
with θ̂1 (t0 ) ≥ 0. Based on (29), (30) and (31), (28) can Then, there has
be rewritten as 

i−1
∂αi−1
 l V̇i = ei f i (x̄i ) + gi xi+1 − f k (x̄k )
∂ xk
e12 1 k1 b2 1 k=1
V̇1 ≤ − + e22 + θ̃1 θ̂1 + M a12 + 12 
2 − e2
kb1 2 η1 2 2

i−1
∂αi−1 1 ˙
1
− gk (x̄k ) xk+1 − i−1 − θ̃i θ̂i
(32) ∂ xk ηi
k=1
From  
= ei gi xi+1 + f¯i − gi2 ei2 − θ̃i θ̂˙i
1
1 1 (38)
θ̃1 θ̂1 ≤ − θ̃12 + θ12 (33) ηi
2 2
 2 where another new function is constructed as
 
and utilizing Lemma 2, let z = 1, α = 1
ϑ̃r,1  , and
2

i−1
∂αi−1

i−1
∂αi−1
l
μ = 1 − l, θ = l, ι = l , the following inequality
1−l f¯i = f i − f k (x̄k ) − gk (x̄k ) xk+1
∂ xk ∂ xk
can be obtained k=1 k=1
  − i−1 + gi2 ei (39)
1 2 l 1 2
θ̃ ≤ (1 − l) ι + θ̃ , (34)
2η1 1 2η1 1 Design virtual control signal as
Furthermore, based on (33), (34) and Lemma 3, (32)  
−1 ¯ 2l−1 1
will be repressed as α̂i = −gi f i + ei + ei (40)
 l   2
l 1 e12 1 2 l k1 By (40), (38) will be
V̇1 ≤ −2 − k 1 θ̃ + (1 − l) ι
2 − e2
2 kb1 2η1 1 η1
1   1
k1 2 b2M 2 1 2 1 2 V̇i = ei gi xi+1 − α̂i − ei2l − ei2
+ θ + a +  + e 2
2η1 1 2 1 2 1 2 2
− gi2 ei2 − θ̃i θ̂˙i
1
(41)
2
kb1 e2 ηi
≤ k 2 −e 2 when |e1 | < k b1 ,
1
It is obvious that log k 2 −e 2
 b1 b1 1 1
l  l Similar to (25), for any given i > 0, there exists a NN
k2 e12 ΦiT ξi (X i ) such that
and then for 0 < l < 1, log k 2 b1
−e2
≤ k 2 −e2
b1 1 b1 1
holds. Thus, one has α̂i = ΦiT ξi (X i ) + δi (X i ) , |δi (X i )| ≤ i ,

123
1850 S. Zhao et al.

where i is a positive constant. Then, we have Define a positive Lyapunov function as


b 1 1 1 2
− ei gi α̂i ≤ 2 ei2 θi ξiT ξi + gi2 ei2 Vn = en2 + θ̃ . (51)
2ai 2 2 2ηn n
b2M 2 1 2 Then, we have
+a + i (42) 
2 i 2 V̇n = en f n (x̄n ) + gn gvμ v (t) + gn d (v)
b
ei gi αi ≤ − 2 ei2 θ̂i (43)
2ai

n−1
∂αn−1

n−1
∂αn−1
Substituting (42) and (43) into (41), we have − f k (x̄k ) − gk (x̄k ) xk+1
  ∂ xk ∂ xk
k=1 k=1
1 bη i ˙ 
V̇i ≤ − ei2l + ei gi (xi+1 − αi ) + θ̃i ei2 − θ̂i
−n−1 − θ̃n θ̂˙n
1
ηi 2ai2
ηn
b2  
1 1 1
− gi2 ei2 − ei2 + M ai2 + i2 (44) ¯
= en f n + gn gvμ v (t) + gn d (v) − gn2 en2
2 2 2 2
− θ̃n θ̂˙n
1
Let (52)
ηn
ei+1 = xi+1 − αi , (45)
where
and based on the Yang’s inequality, we have

n−1
∂αn−1
1 1 2 f¯n = f n − f k (x̄k )
ei gi ei+1 ≤ gi2 ei2 + ei+1 (46) ∂ xk
2 2 k=1
Construct the adaptive law as

n−1
∂αn−1
− gk (x̄k ) xk+1 − n−1
θ̂˙i = 2 ei2 − ki θ̂i
bηi
(47) ∂ xk
k=1
2ai
Combining (45), (46) and (47), (44) can be rewritten as + gn2 en . (53)
1 2 ki 1 Design virtual control signal as
V̇i ≤ − ei2l + ei+1 + θ̃i θ̂i − ei2  
2 ηi 2  −1 1
α̂n = − gn gvμ f¯n + en2l−1 + en . (54)
b2M 2 1 2 2
+ a + i (48)
2 i 2 By (54), (52) will be
Similar to (33) and (34), then (48) can be rewritten as   1
    V̇n = −en2l + en gn gvμ v (t) − α̂n − en2
1 2 l 1 2 l ki 2
V̇i ≤ −2l ei − ki θ̃i + (1 − l) ι
+ en gn d (v) − gn2 en2 − θ̃n θ̂˙n .
2 2ηi ηi 1
(55)
ηn
ki 2 b2M 2 1 2 1 2 1 2
+ θ + a + i − ei + ei+1 Similar to (25), for any given n > 0, there exists a NN
2ηi i 2 i 2 2 2
ΦnT ξn (X n ) such that
1 1 2
≤ −σi,1 Vil + σi,2 − ei2 + ei+1 (49) α̂n = ΦnT ξn (X n ) + δn (X n ) |δn (X n )| ≤ n ,
2 2
 
where σi,1 = min 2l , ki , σi,2 = ηkii (1 − l) ι+ 2η
ki 2
θ + where n is a positive constant. Then, we have
i i
b2M b 1
2 ai2 + 21 i2 . − en gn gvμ α̂n ≤ 2 en2 θn ξnT ξn + gn2 en2
2an 2
Step n: Define error variable en = xn − αn−1 , we have b2M 2 1 2
ėn = f n (x̄n ) + gn gvμ v (t) + gn d (v) − α̇n−1 (50) + a +  (56)
2 n 2 n
where b
ei gn gvμ v ≤ − 2 en2 θ̂n . (57)

n−1
∂αn−1

n−1
∂αn−1 2an
α̇n−1 = f k (x̄k ) + gk (x̄k ) xk+1 Substituting (56) and (57) into (55), we have
∂ xk ∂ xk  
k=1 k=1
1 bηn 2 ˙
+ n−1 , V̇n ≤ −en + θ̃n
2l
e − θ̂n + en gn d (v)
ηn 2an2 n

n−1
∂αn−1 ˙
n
∂αn−1 (k+1)
n−1 = θ̂k + y 1 1 b2 1
(k) r − gn2 en2 − en2 + M an2 + n2 .
k=1 ∂ θ̂k
(58)
k=0 ∂ yr 2 2 2 2

123
Adaptive NN finite-time tracking control 1851

Based on the Yang’s inequality and (5), we have That is, after the finite time T ∗ , the tracking error
1 1 remains in a small neighborhood of the origin. Now,
en gn d (v) ≤ gn2 en2 + D 2 (59) the following theorem is given to sum up the above
2 2
Construct the adaptive law as backstepping calculations procedure.
Therefore, from the above calculations, it is obvious
θ̂˙n = 2 en2 − kn θ̂n
bηn
(60) that |x1 | ≤ |e1 | + |yr (t)| < kb1 + B0 . Define kb1 =
2an kc1 − B0 , and then, it has |x1 | < kc1 . Thus, the system
Combining (59) and (60), (58) can be rewritten as output is not violated by its limit.
kn 1
V̇n ≤ − en2l + θ̃n θ̂n − en2 Theorem 1 For the nonlinear system (1) with output
ηn 2
2 constraints and input saturation, with the aid of real
bM 2 1 2 1 2
a +  + D (61) controller (18), the virtual controllers (16), (17) and
2 n 2 n 2 adaptation laws (31), (47), (60), and then, all the sig-
Similar to (33) and (34), (61) can be rewritten as nals in the closed-loop system are SGPFS and the
 l  
l 1 2 1 2 l kn system output tracks the reference signal into a small
V̇n ≤ −2 e − kn θ̃ + (1 − l) ι
2 n 2ηn n ηn neighborhood of zero in a finite time.
kn 2 b2M 2 1 2 1 2 1 2
+ θ + a +  + D − en Remark 3 An adaptive NN finite-time tracking con-
2ηn n 2 n 2 n 2 2
trol method is studied in this paper for the output con-
1 2
≤ −σn,1 Vn + σn,2 − en
l
(62) strained nonlinear system with input saturation, it can
2 obtain the system stability in finite-time. However, we
 l 
where σn,1 = min 2 , kn , σn,2 = ηknn (1 − l) ι + have not considered more factors such as the external
kn 2 b2M 2 disturbances or time-varying constraints which usually
2ηn θn+ 21 n2 + 2 an + 21 D 2 .
appeared in the practical engineering system, and then,
Define
in future work, we will try to do more deeper work in
V = V1 + V2 + . . . + Vn (63) this direction and study these methods directly to the
Then, we have actual engineering system model.

V̇ = V̇1 + V̇2 + . . . + V̇n (64)


Combining (35), (49) and (62), one has
4 Simulation results

n
V̇ ≤ − σi,1 Vil + σi,2 . (65)
The following simulation examples results are put for-
i=1 i=1
ward to show the availability of the presented scheme.
Based on Lemma 3, we can obtain
V̇ ≤ − σ1 V l + σ2 (66)
  Example 1 Consider the nonlinear system as follows:
where
n σ1 = min σi,1 , i = 1, 2, . . . , n , σ2 = ⎧  
σ
i=1 i,2 . ⎪
⎪ ẋ1 = 2 + 0.5sin(x1 ) + x14 x2 
   1−l  ⎨
σ2 ẋ2 = x12 x23 + 2 + 0.25 cos(x22 ) x3
Let T ∗ = (1−l)ασ
1 1−l (0) − l
V , (67)
1 (1−α)σ1 ⎪ ẋ3 = x32 sin(x1 − x2 ) + u (v (t))


where V (0) denotes the starting value of V (t). y = x1
In accordance with Lemma 1, for ∀t ≥ T ∗ , V l ≤
σ2 where an input saturation is described as
(1−α)σ1 . In other words, all the signals of the considered
closed-loop systems are SGPFS. ⎧
⎨ 4, v(t) ≥ 4
Furthermore, according to the definition of V , for u = sat (v (t)) = v, − 4 < v(t) < 4 (68)
∀t ≥ T ∗ , we have ⎩
− 4, v(t) ≤ − 4
  1 2
1
σ2 l
−2 In order to deal with the unknown nonlinear terms, the
|y − yr | ≤ kb1 1 − e (1−α)σ1
.
membership functions are chosen as

123
1852 S. Zhao et al.

1.5 1.4
e1
y e2
yr 1.2
1
1

0.8
0.5
0.6

0 0.4

0.2
−0.5
0

−0.2
−1
−0.4

−1.5 −0.6
0 10 20 30 40 50 0 10 20 30 40 50
time[Sec] Time(Sec)

Fig. 1 Trajectories of y (t) and yr (t) Fig. 3 Trajectory of e1 and e2

0.8
x2 0.7
0.6 x3 θ̂1
0.6 θ̂2
0.4 θ̂3
0.5
0.2
0.4
0
0.3
−0.2
0.2
−0.4
0.1
−0.6
0
−0.8
0 10 20 30 40 50 −0.1
0 10 20 30 40 50
time[Sec]
Time(Sec)
Fig. 2 State trajectories x2 and x3
Fig. 4 Trajectories of adaptive laws θ̂1 , θ̂2 and θ̂3

 
(x1 − j)2
μ M J (x1 ) = exp − system are chosen as x1 (0) = 0.1, x2 (0) = 0.1,
1 2
  x3 (0) = 0.3 and [θ1 (0) , θ2 (0) , θ3 (0)]T = [0, 0, 0]T .
(x1 − j)2 (x2 − j)2 Figures 1, 2, 3, 4, 5 and 6 demonstrate corresponding
μ M J (x̄2 ) = exp − −
2 2 2 emulation results. Figure 1 shows trajectories of sys-
 
(x1 − j) 2 (x2 − j)2 (x3 − j)2 tem output y and a given reference signal yr (t), which
μ M J (x̄3 ) = exp − − − demonstrates that the tracking error can be assured to
3 2 2 2
be small enough and output y of the systems does not
j = −9, −7, −5, −3, −1, 0, 1, 3, 5, 7, 9.
violate the limit. Figure 2 depicts the trajectories of the
For the simulation, the reference signal is given as system states x2 and x3 , which reveals that the systems
yr (t) = sin (0.5t) + 0.5 sin (t), system output is lim- states are bounded. Figures 3 and 4 indicate the control
ited as |y| ≤ |yr | + |e1 | ≤ 1.5 + 0.5 = 2. In this errors and adaptive laws of the systems are bounded,
simulation, all the design parameters are chosen as and Figs. 5 and 6 show the system input saturation
a1 = 3, a2 = 2, a3 = 3, k1 = k2 = k3 = 7 u (v (t)) and the saturation control input v are bounded
and η1 = η2 = η3 = 15. The initial values of the as well.

123
Adaptive NN finite-time tracking control 1853

4 0.12
u y
yr
3 0.1

2 0.08

1
0.06

0
0.04
−1
0.02
−2
0
−3
−0.02
−4 0 10 20 30 40 50
0 10 20 30 40 50
time[Sec]
time[Sec]
Fig. 7 Trajectories of y (t) and yr (t)
Fig. 5 Trajectory of system input u (v (t))
0.1
10 x
2
v 0.08

0
0.06

−10
0.04

−20 0.02

−30 0

−0.02
−40

−0.04
−50
−0.06
0 10 20 30 40 50
−60 time[Sec]
0 10 20 30 40 50
time[Sec]
Fig. 8 State trajectory x2
Fig. 6 Trajectory of control input v
0.45
e1
e2
0.4
The simulation results reveal that in a finite time the 0.35
system output follows the expected reference signal
0.3
into a small enough bounded set, and that all of the
0.25
signals in the closed-loop systems are bounded.
0.2
Example 2 To further illustrate the effectiveness and 0.15
practicality of the proposed control method, a single- 0.1
link rigid robot system [3] is considered as follows:
0.05
J θ̈ = − (0.5mgl + Mgl) sin (θ ) + u, 0

where J = Ml 2+ 1
3 ml
2
is the moment of inertia, −0.05
0 10 20 30 40 50
θ ∈ 0, π2 is the joint rotation angle, m = 1.5 kg Time(Sec)
represents the mass of the load, g = 9.8 m/s 2 is the
gravity coefficient, M = 3 kg denotes the mass of the Fig. 9 Trajectory of e1 and e2
rigid link and l = 0.5 m is the length of the robot link.

123
1854 S. Zhao et al.

0.1 2
θ̂1 u
0.09 θ̂2
1
0.08

0.07
0
0.06

0.05 −1
0.04

0.03 −2

0.02
−3
0.01

0
0 10 20 30 40 50 −4
0 10 20 30 40 50
Time(Sec)
time[Sec]
Fig. 10 Trajectories of adaptive laws θ̂1 , θ̂2 and θ̂3
Fig. 11 Trajectory of system input u (v (t))

10
Define the new states variables as x1 = θ , x2 = θ̇ , v
and the system input is the input saturation described 8

as 6

⎨ 4, v(t) ≥ 4 4

u = sat (v (t)) = v, − 4 < v(t) < 4 2



− 4, v(t) ≤ − 4 0

we can construct the system as −2


⎧ −4
⎨ ẋ1 = x2
ẋ = 1J [− (0.5mgl + Mgl) sin (x1 ) + u (v (t))] −6
⎩ 2
y = x1 −8

In this simulation, the initial values of the states are −10


0 10 20 30 40 50
x1 (0) = 0.1, x2 (0) = 0.1 and [θ1 (0) , θ2 (0)]T = time[Sec]
[0, 0]T . The reference signal is yr (t) = 0.015 sin (0.8t)
+ 0.01, system output is limited as |y| ≤ |yr | + |e1 | ≤ Fig. 12 Trajectory of control input v
1.5 + 0.5 = 2. In this simulation, all the design param-
eters are chosen as a1 = 3, a2 = 2, k1 = k2 = 0.7
bounded set, all the closed-loop signals are bounded,
and η1 = η2 = 15. Figures 7, 8, 9, 10, 11, and 12
and the effectiveness of the proposed finite-time control
demonstrate corresponding emulation results. Figure 7 scheme is demonstrated.
shows trajectories of system output y and a given ref-
erence signal yr (t), which demonstrates that the track-
ing error can be assured to be small enough and output
y of the systems does not violate the limit. Figure 8 5 Conclusion
depicts the trajectories of the system states x2 , which
reveals that the systems states are bounded. Figures 9 This paper has proposed a finite-time tracking con-
and 10 indicate the control errors and adaptive laws of trol scheme for the output constrained and input satu-
the systems are bounded, and Figs. 11 and 12 show that rated nonlinear systems. By adopting the BLFs and NN
the system input saturation u (v (t)) and the saturation approximate approach, in accordance with the adap-
control input v are bounded as well. tive backstepping scheme and the finite-time stability
The simulation results show that in a finite-time sys- theory, the tracking control problem is studied with-
tem output follows the desired reference signal in a out the constrained condition of the unknown nonlinear

123
Adaptive NN finite-time tracking control 1855

functions terms. Under the proposed control method, 12. Utkin, V.I.: Sliding mode control design principles and appli-
the system performance can be achieved in a finite cations to electric drives. IEEE Trans. Industr. Electron.
40(1), 23–36 (1993)
time, and the tracking error of the system can reach 13. Liu, Y., Tong, S.: Adaptive fuzzy identification and control
a small enough neighborhood of zero concurrently. for a class of nonlinear pure-feedback MIMO systems with
These results can be demonstrated by two simulation unknown dead zones. IEEE Trans. Fuzzy Syst. 23(5), 1387–
examples in this paper. 1398 (2015)
14. Li, Y., Tong, S.: Adaptive fuzzy output-feedback control of
pure-feedback uncertain nonlinear systems with unknown
Funding This work is partially supported by the National Nat-
dead zone. IEEE Trans. Fuzzy Syst. 22(5), 1341–1347
ural Science Foundation of China (61703051) and the Depart-
(2014)
ment of Education of Liaoning Province (LZ2017001).
15. Li, Y., Tong, S., Li, T.: Adaptive fuzzy output feedback
dynamic surface control of interconnected nonlinear pure-
Compliance with ethical standards feedback systems. IEEE Trans. Cybern. 45(1), 138–149
(2015)
Conflict of interest The authors declare that they have no con- 16. Su, C.Y., Stepanenko, Y.: Adaptive control of a class of non-
flict of interest. linear systems with fuzzy logic. IEEE Trans. Fuzzy Syst.
2(4), 285–294 (1994)
17. Tsay, D.L., Chung, H.Y., Lee, C.J.: The adaptive control
References of nonlinear systems using the Sugeno-type of fuzzy logic.
IEEE Trans. Fuzzy Syst. 7(2), 225–229 (1999)
1. Chen, B., Liu, X., Tong, S.: Adaptive fuzzy output tracking 18. Wang, L.X.: Stable adaptive fuzzy control of nonlinear sys-
control of MIMO nonlinear uncertain systems. IEEE Trans. tems. In: Proceedings of the IEEE Conference on Decision
Fuzzy Syst. 15(2), 287–300 (2007) and Control, 1992, vol. 3, pp. 2511–2516. IEEE (2002)
2. Li, H., Wu, C., Shi, P., Gao, Y.: Control of nonlinear net- 19. Tong, S., Zhang, L., Li, Y.: Observed-based adaptive fuzzy
worked systems with packet dropouts: interval type-2 fuzzy decentralized tracking control for switched uncertain nonlin-
model-based approach. IEEE Trans. Cybern. 45(11), 2378– ear large-scale systems with dead zones. IEEE Trans. Syst.
2389 (2015) Man Cybern. Syst. 46(1), 37–47 (2015)
3. Li, H., Wang, J., Du, H., Karimi, H.R.: Adaptive sliding 20. Wang, L., Basin, M., Li, H., Lu, R.: Observer-based com-
mode control for Takagi-Sugeno fuzzy systems and its appli- posite adaptive fuzzy control for nonstrict-feedback sys-
cations. IEEE Trans. Fuzzy Syst. (2017). https://doi.org/10. tems with actuator failures. IEEE Trans. Fuzzy Syst. (2017).
1109/TFUZZ.2017.2686357 https://doi.org/10.1109/TFUZZ.2017.2774185
4. Wang, N., Sun, J.C., Meng, J.E.: Tracking-error-based uni- 21. Liang, H., Li, H., Yu, Z., Li, P., Wang, W.: Cooperative
versal adaptive fuzzy control for output tracking of nonlinear robust containment control for general discrete-time multi-
systems with completely unknown dynamics. IEEE Trans. agent systems with external disturbance. IET Control The-
Fuzzy Syst. PP(99), 1–1 (2017) ory Appl. 11(12), 1928–1937 (2017)
5. Wang, M., Wang, C.: Learning from adaptive neural 22. Wang, D., Huang, J.: Neural network-based adaptive
dynamic surface control of strict-feedback systems. IEEE dynamic surface control for a class of uncertain nonlinear
Trans. Neural Netw. Learn. Syst. 26(6), 1247–1259 (2015) systems in strict-feedback form. IEEE Trans. Neural Netw.
6. Liu, Z., Wang, F., Zhang, Y.: Adaptive visual tracking control 28(9), 2156–2167 (2016)
for manipulator with actuator fuzzy dead-zone constraint 23. Gong, J.Q., Yao, B.: Neural network adaptive robust control
and unmodeled dynamic. IEEE Trans. Syst. Man Cybern. of nonlinear systems in semi-strict feedback form. Automat-
Syst. 45(10), 1301–1312 (2015) ica 37(8), 1149–1160 (2001)
7. Liu, Y., Tong, S.: Adaptive fuzzy control for a class of nonlin- 24. Johnson, E.N., Calise, A.J.: Neural network adaptive con-
ear discrete-time systems with backlash. IEEE Trans. Fuzzy trol of systems with input saturation. In: Proceedings of the
Syst. 22(5), 1359–1365 (2014) American Control Conference, 2001, vol. 5, pp. 3527–3532.
8. Wang, N., Sun, J.C., Han, M., Zheng, Z., Er, M.J.: Adap- IEEE (2001)
tive approximation-based regulation control for a class of 25. Wang, D., Huang, J.: Adaptive neural network control for a
uncertain nonlinear systems without feedback linearizabil- class of uncertain nonlinear systems in pure-feedback form.
ity. IEEE Trans. Neural Netw. Learn. Syst. PP(99), 1–14 Automatica 38(8), 1365–1372 (2002)
(2017) 26. Hayakawa, T., Haddad, W.M., Hovakimyan, N.: Neural net-
9. Boiko, I., Fridman, L., Iriarte, R.: Analysis of chattering work adaptive control for a class of nonlinear uncertain
in continuous sliding mode control. In: Proceedings of the dynamical systems with asymptotic stability guarantees.
American Control Conference, 2005, vol. 4, pp. 2439–2444. IEEE Trans. Neural Netw. 19(1), 80 (2008)
IEEE (2005) 27. Deng, F., Guo, S., Zhou, R., Chen, J.: Sensor multifault diag-
10. Li, H., Wang, J., Lam, H.-K., Zhou, Q., Du, H.: Adaptive nosis with improved support vector machines. IEEE Trans.
sliding mode control for interval type-2 fuzzy systems. IEEE Autom. Sci. Eng. PP(99), 1–11 (2015)
Trans. Syst. Man Cybern. Syst. 46(12), 1654–1663 (2016) 28. Deng, F., Guan, S., Yue, X., Gu, X., Chen, J., Lv, J., Li,
11. Li, H., Wang, J., Shi, P.: Output-feedback based sliding mode J.: Energy-based sound source localization with low power
control for fuzzy systems with actuator saturation. IEEE consumption in wireless sensor networks. IEEE Trans. Ind.
Trans. Fuzzy Syst. 24(6), 1282–1293 (2016) Electron. PP(99), 1–1 (2017)

123
1856 S. Zhao et al.

29. Chen, C.P., Wen, G., Liu, Y., Wang, F.: Adaptive consen- national Conference on Control and Automation, pp. 2511–
sus control for a class of nonlinear multiagent time-delay 2516. IEEE (2010)
systems using neural networks. IEEE Trans. Neural Netw. 46. Li, T., Li, R., Li, J.: Decentralized adaptive neural con-
Learn. Syst. 25(6), 1217–1226 (2014) trol of nonlinear interconnected large-scale systems with
30. Pan, Y., Er, M.J., Huang, D., Wang, Q.: Adaptive fuzzy con- unknown time delays and input saturation. Neurocomput-
trol with guaranteed convergence of optimal approximation ing 74(14C15), 2277–2283 (2011)
error. IEEE Trans. Fuzzy Syst. 19(5), 807–818 (2011) 47. Li, Y., Tong, S., Li, T.: Composite adaptive fuzzy out-
31. Park, J.H., Park, C., Kwon, O., Lee, S.-M.: A new stability put feedback control design for uncertain nonlinear strict-
criterion for bidirectional associative memory neural net- feedback systems with input saturation. IEEE Trans.
works of neutral-type. Appl. Math. Comput. 199(2), 716– Cybern. 45(10), 2299 (2015)
722 (2008) 48. Zhou, Q., Shi, P., Tian, Y., Wang, M.: Approximation-based
32. Zhang, T., Ge, S.S., Hang, C.C.: Adaptive neural network adaptive tracking control for MIMO nonlinear systems with
control for strict-feedback nonlinear systems using back- input saturation. IEEE Trans. Cybern. 45(10), 2119–2128
stepping design. Automatica 36(12), 1835–1846 (2000) (2015)
33. Kwon, O., Park, J.H., Lee, S., Cha, E.: New augmented 49. Mou, C., Gang, T., Jiang, B.: Dynamic surface control using
Lyapunov–Krasovskii functional approach to stability anal- neural networks for a class of uncertain nonlinear systems
ysis of neural networks with time-varying delays. Nonlinear with input saturation. IEEE Trans. Neural Netw. Learn. Syst.
Dyn. 76(1), 221–236 (2014) 26(9), 2086 (2015)
34. Shen, H., Park, J.H., Wu, Z.-G.: Finite-time synchronization 50. Wen, C., Zhou, J., Liu, Z., Su, H.: Robust adaptive control of
control for uncertain Markov jump neural networks with uncertain nonlinear systems in the presence of input satura-
input constraints. Nonlinear Dyn. 77(4), 1709–1720 (2014) tion and external disturbance. IEEE Trans. Autom. Control
35. Li, T., Li, Z., Wang, D., Chen, C.L.P.: Output-feedback 56(7), 1672–1678 (2011)
adaptive neural control for stochastic nonlinear time-varying 51. Hu, Q., Jiang, B., Zhang, Y.: Observer-based output feed-
delay systems with unknown control directions. IEEE Trans. back attitude stabilization for spacecraft with finite-time
Neural Netw. Learn. Syst. 26(6), 1188–1201 (2015) convergence. IEEE Trans. Control Syst. Technol. PP(99),
36. Li, Z., Li, T., Feng, G.: Adaptive neural control for a class 1–9 (2017)
of stochastic nonlinear time-delay systems with unknown 52. Hu, Q., Tan, X., Akella, M.R.: Finite-time fault-tolerant
dead zone using dynamic surface technique. Int. J. Robust spacecraft attitude control with torque saturation. J. Guid.
Nonlinear Control 26(4), 759–781 (2016) Control Dyn. 40(10), 1–14 (2017)
37. Zhou, Q., Li, H., Wang, L., Lu, R.: Prescribed perfor- 53. Fu, J., Chai, T., Chai, T.: Global Finite-Time Stabilization of
mance observer-based adaptive fuzzy control for nonstrict- a Class of Switched Nonlinear Systems with the Powers of
feedback stochastic nonlinear systems, IEEE Trans. Fuzzy Positive Odd Rational Numbers. Pergamon Press Inc., New
Syst Man Cybern Syst. (2017). https://doi.org/10.1109/ York (2015)
TSMC.2017.2738155 54. Wang, F., Chen, B., Lin, C., Zhang, J.: Adaptive neural net-
38. Wang, B., Cheng, J., Al-Barakati, A., Abdullah, H.M.: A work finite-time output feedback control of quantized non-
mismatched membership function approach to sampled- linear systems. IEEE Trans. Cybern. PP(99), 1–10 (2017)
data stabilization for T-S fuzzy systems with time-varying 55. Yu, S., Yu, X., Stonier, R.: Continuous finite-time control
delayed signals. Sig. Process. 140, 161–170 (2017) for robotic manipulators with terminal sliding modes. Auto-
39. Cheng, J., Park, J.H., Zhang, L., Zhu, Y.: An asynchronous matica 41(11), 1957–1964 (2005)
operation approach to event-triggered control for fuzzy 56. Wang, H., Chen, B., Liu, X., Liu, K., Lin, C.: Robust adap-
Markovian jump systems with general switching policies. tive fuzzy tracking control for pure-feedback stochastic non-
IEEE Trans. Fuzzy Syst. 99, 1–1 (2016) linear systems with input constraints. IEEE Trans. Cybern.
40. Cheng, J., Park, J.H., Karimi, H.R., Shen, H.: A flexible 43(6), 2093–2104 (2013)
terminal approach to sampled-data exponentially synchro- 57. Zhu, Z., Xia, Y., Fu, M.: Attitude stabilization of rigid space-
nization of Markovian neural networks with time-varying craft with finite-time convergence. Int. J. Robust Nonlinear
delayed signals. IEEE Trans. Cybern. 99, 1–13 (2017) Control 21(6), 686–702 (2015)
41. Lee, T.H., Ju, H.P.: Improved criteria for sampled-data 58. Qian, C., Lin, W.: Non-Lipschitz continuous stabilizers for
synchronization of chaotic Lure systems using two new nonlinear systems with uncontrollable unstable lineariza-
approaches. Nonlinear Anal. Hybrid Syst. 24, 132–145 tion. Syst. Control Lett. 42(3), 185–200 (2001)
(2017) 59. Zhou, Q., Shi, P., Xu, S., Li, H.: Observer-based adaptive
42. Lee, T.H., Wu, Z., Ju, H.P.: Synchronization of a complex neural network control for nonlinear stochastic systems with
dynamical network with coupling time-varying delays via time delay. IEEE Trans. Neural Netw. Learn. Syst. 24(1), 71
sampled-data control. Appl. Math. Comput. 219(3), 1354– (2013)
1366 (2012) 60. Chen, B., Liu, X., Liu, K., Lin, C.: Direct adaptive fuzzy con-
43. Li, T.H., Ju, H.P., Xu, S.: Relaxed conditions for stability of trol of nonlinear strict-feedback systems. Automatica 45(6),
time-varying delay systems. Automatica 75, 11–15 (2017) 1530–1535 (2009)
44. Tee, K.P., Ge, S.S., Tay, E.H.: Barrier Lyapunov functions for 61. Liu, Y.J., Tong, S.: Barrier Lyapunov functions for Nuss-
the control of output-constrained nonlinear systems. Auto- baum gain adaptive control of full state constrained nonlin-
matica 45(4), 918–927 (2009) ear systems. Automatica 76, 143–152 (2017)
45. Tee, K.P., Ge, S.S., Li, H., Ren, B.: Control of nonlinear sys-
tems with time-varying output constraints. In: IEEE Inter-

123

Vous aimerez peut-être aussi