306 H. WANG, B. CHEN AND C. LIN
Assumption 2
The signs of g
i
.
Nx
i
, x
iC1
/
, i D 1, 2, :::, n, are known, and there exist constants b
m
, b
M
such that
for 1 6 i 6 n,
0<b
m
6
j
g
i
.
Nx
i
, x
iC1
/
j
6 b
M
< 1, 8
.
Nx
i
, x
iC1
/
2 R
i
R. (17)
Remark 3
As shown later, the constants b
m
and b
M
in Assumption 2 are just required for stability analysis;
their true values are unnecessary to be known for controller design. However, in [43], these constants
must be known for control design.
Assumption 3
The reference signal y
d
.t/ and its time derivatives up to the nth order are continuous and bounded.
In the following, RBF neural networks will be used to approximate the unknown smooth nonlinear
functions. In [45], it has been proved that with sufficiently large node number l, the RBF neural net-
works W
T
S.Z/ can approximate any continuous function f.Z/ over a compact set R
q
to
arbitrary accuracy ">0as
f.Z/D W
T
S.Z/ C ı.Z/, 8´ 2 2 R
q
, (18)
where W
is the ideal constant weight vector and defined as
W
WD arg min
W 2
N
R
N
sup
Z2
ˇ
ˇ
f.Z/ W
T
S.Z/
ˇ
ˇ
,
and ı.Z/ is the approximation error satisfying jı.Z/j 6 ", W
D Œw
1
, w
2
, :::, w
l
T
2 R
l
is the
weight vector and S.Z/ D Œs
1
.Z/, s
2
.Z/, :::, s
l
.Z/
T
is the basis function vector with l being the
number of the neural networks nodes and l>1. The basis function s
i
.Z/ in this paper is chosen as
the commonly used Gaussian function with the form:
s
i
.Z/ D exp
"
.
Z
i
/
T
.
Z
i
/
2
i
#
, i D 1, 2, :::, l,
where
i
D Œ
i1
,
i2
, :::,
iq
T
is the center of the receptive field and
i
is the width of the
Gaussian function.
3. ADAPTIVE NEURAL TRACKING CONTROL
In this section, an adaptive neural control scheme via backstepping technique is proposed for the
systems (4). The backstepping design procedure contains n steps and is developed based on the
following coordinate transformation:
´
i
D x
i
˛
i1
Nx
T
i
,
N
O
T
i
, Ny
.i/T
d
, i D 1, 2, , n, (19)
where
N
O
i
D
h
O
1
,
O
2
, :::,
O
i
i
T
, ˛
0
D y
d
,and˛
i1
is the virtual control signal, which will be
specified later.
O
i
is the estimation of unknown constant
i
which is defined as
i
D
b
2
M
b
m
k
W
i
k
2
, i D 1, 2, , n, (20)
where b
m
and b
M
are defined in Assumption 1 and W
i
will be specified later.
Copyright © 2012 John Wiley & Sons, Ltd. Int. J. Adapt. Control Signal Process. 2013; 27:302–322
DOI: 10.1002/acs