Nonlinear Analysis: Hybrid Systems 2 (2008) 862874

www.elsevier.com/locate/nahs
Neural network hybrid adaptive control for nonlinear uncertain
impulsive dynamical systems
Tomohisa Hayakawa
a,1
, Wassim M. Haddad
b,
, Konstantin Y. Volyanskyy
b,2
a
Department of Mechanical and Environmental Informatics, Tokyo Institute of Technology, Tokyo 332-0012, Japan
b
School of Aerospace Engineering, Georgia Institute of Technology, Atlanta, GA 30332-0150, United States
Received 30 October 2007; accepted 17 January 2008
Abstract
A neural network hybrid adaptive control framework for nonlinear uncertain hybrid dynamical systems is developed. The
proposed hybrid adaptive control framework is Lyapunov-based and guarantees partial asymptotic stability of the closed-loop
hybrid system; that is, asymptotic stability with respect to part of the closed-loop system states associated with the hybrid plant
states. A numerical example is provided to demonstrate the efcacy of the proposed hybrid adaptive stabilization approach.
c 2008 Elsevier Ltd. All rights reserved.
Keywords: Hybrid adaptive control; Neural networks; Nonlinear hybrid systems; Impulsive dynamical systems; System uncertainty; Stabilization;
Sector-bounded nonlinearities
1. Introduction
Modern complex engineering systems involve multiple modes of operation placing stringent demands on controller
design and implementation of increasing complexity. Such systems typically possess a multiechelon hierarchical
hybrid control architecture characterized by continuous-time dynamics at the lower levels of the hierarchy and
discrete-time dynamics at the higher levels of the hierarchy (see [13] and the numerous references therein). The
lower-level units directly interact with the dynamical system to be controlled while the higher-level units receive
information from the lower-level units as inputs and provide (possibly discrete) output commands which serve to
coordinate and reconcile the (sometimes competing) actions of the lower-level units. The hierarchical controller
organization reduces processor cost and controller complexity by breaking up the processing task into relatively small
pieces and decomposing the fast and slow control functions. Typically, the higher-level units perform logical checks
that determine system mode operation, while the lower-level units execute continuous-variable commands for a given
system mode of operation. The mathematical description of many of these systems can be characterized by impulsive
differential equations [37].

Corresponding author. Tel.: +1 404 894 1078; fax: +1 404 894 2760.
E-mail addresses: hayakawa@mei.titech.ac.jp (T. Hayakawa), wm.haddad@aerospace.gatech.edu (W.M. Haddad), gtg891s@mail.gatech.edu
(K.Y. Volyanskyy).
1
Tel.: +81 3 5734 2762; fax: +81 3 5734 2762.
2
Tel.: +1 404 894 1078; fax: +1 404 894 2760.
1751-570X/$ - see front matter c 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.nahs.2008.01.002
T. Hayakawa et al. / Nonlinear Analysis: Hybrid Systems 2 (2008) 862874 863
The purpose of feedback control is to achieve desirable system performance in the face of system uncertainty.
To this end, adaptive control along with robust control theory have been developed to address the problem of
system uncertainty in control-system design. In contrast to xed-gain robust controllers, which maintain specied
constants within the feedback control law to sustain robust performance, adaptive controllers directly or indirectly
adjust feedback gains to maintain closed-loop stability and improve performance in the face of system uncertainties.
Specically, indirect adaptive controllers utilize parameter update laws to identify unknown system parameters and
adjust feedback gains to account for system variation, while direct adaptive controllers directly adjust the controller
gains in response to plant variations. The inherent nonlinearities and systemuncertainties in hierarchical hybrid control
systems and the increasingly stringent performance requirements required for controlling such modern complex
embedded systems necessitates the development of hybrid adaptive nonlinear control methodologies.
In a recent paper [8], a hybrid adaptive control framework for adaptive stabilization of multivariable nonlinear
uncertain impulsive dynamical systems was developed. In particular, a Lyapunov-based hybrid adaptive control
framework was developed that guarantees partial asymptotic stability of the closed-loop system; that is, asymptotic
stability with respect to part of the closed-loop system states associated with the hybrid plant dynamics. Furthermore,
the remainder of the state associated with the adaptive controller gains was shown to be Lyapunov stable. As is the
case in the continuous and discrete-time adaptive control literature [913], the system errors in [8] are captured by
a constant linearly parameterized uncertainty model of a known structure but unknown variation. This uncertainty
characterization allows the system nonlinearities to be parameterized by a nite linear combination of basis functions
within a class of function approximators such as rational functions, spline functions, radial basis functions, sigmoidal
functions, and wavelets. However, this linear parametrization of basis functions cannot, in general, exactly capture the
unknown system nonlinearity.
Neural network-based adaptive control algorithms have been extensively developed in the literature, wherein
Lyapunov-like functions are used to ensure that the neural network controllers can guarantee ultimate boundedness of
the closed-loop system states rather than closed-loop asymptotic stability. Ultimate boundness ensures that the plant
states converge to a neighborhood of the origin (see, for example, [1416] for continuous-time systems and [1719]
for discrete-time systems). The reason why stability in the sense of Lyapunov is not guaranteed stems from the fact that
the uncertainties in the system dynamics cannot be perfectly captured by neural networks using the universal function
approximation property and the residual approximation error is characterized via a norm bound over a given compact
set. Ultimate boundedness guarantees, however, are often conservative since standard Lyapunov-like theorems that are
typically used to showultimate boundedness of the closed-loop hybrid system states provide only sufcient conditions,
while neural network controllers may possibly achieve plant state convergence to an equilibrium point.
In this paper, we develop a neural hybrid adaptive control framework for a class of nonlinear uncertain impulsive
dynamical systems which ensures state convergence to a Lyapunov stable equilibrium as well as boundedness of
the neural network weighting gains. Specically, the proposed framework is Lyapunov-based and guarantees partial
asymptotic stability of the closed-loop hybrid system; that is, Lyapunov stability of the overall closed-loop states and
convergence of the plant state. The neuroadaptive controllers are constructed without requiring explicit knowledge of
the hybrid system dynamics other than the fact that the plant dynamics are continuously differentiable and that the
approximation error of the unknown system nonlinearities lies in a small gain-type norm bounded conic sector over a
compact set. Hence, the overall neuroadaptive control framework captures the residual approximation error inherent
in linear parameterizations of system uncertainty via basis functions. Furthermore, the proposed neuroadaptive control
architecture is modular in the sense that if a nominal linear design model is available, then the neuroadaptive controller
can be augmented to the nominal design to account for system nonlinearities and system uncertainty.
Finally, we emphasize that we do not impose any linear growth condition on the system resetting (discrete)
dynamics. In the literature on classical (non-neural) adaptive control theory for discrete-time systems, it is typically
assumed that the nonlinear system dynamics have the linear growth rate which is necessary in proving Lyapunov
stability rather than practical stability (ultimate boundedness). Our novel characterization of the system uncertainties
(i.e., the small gain-type bound on the norm of the modeling error) allows us to prove asymptotic stability without
requiring a linear growth condition on the system dynamics.
2. Mathematical preliminaries
In this section, we introduce notation, denitions, and some key results concerning impulsive dynamical systems
[37,20]. Let R denote the set of real numbers, R
n
denote the set of n 1 real column vectors, ()
T
denote transpose,
864 T. Hayakawa et al. / Nonlinear Analysis: Hybrid Systems 2 (2008) 862874
()

denote the MoorePenrose generalized inverse, N denote the set of nonnegative integers, N
n
(resp., P
n
) denote
the set of n n nonnegative (resp., positive) denite matrices, and I
n
denote the n n identity matrix. Furthermore,
we write tr () for the trace operator, ln() for the natural log operator,
min
(M) (resp.,
max
(M)) for the minimum
(resp., maximum) eigenvalue of the Hermitian matrix M,
max
(M) for the maximum singular value of the matrix M,
V

(x) for the Fr echet derivative of V at x, and dist( p, M) for the smallest distance from a point p to any point in the
set M, that is, dist( p, M) inf
xM
p x.
In this paper, we consider controlled state-dependent [3] impulsive dynamical systems G of the form
x(t ) = f
c
(x(t )) + G
c
(x(t ))u
c
(t ), x(0) = x
0
, x(t ) Z
x
, (1)
x(t ) = f
d
(x(t )) + G
d
(x(t ))u
d
(t ), x(t ) Z
x
, (2)
where t 0, x(t ) D R
n
, D is an open set with 0 D, x(t ) x(t
+
) x(t ), u
c
(t ) U
c
R
m
c
,
u
d
(t
k
) U
d
R
m
d
, t
k
denotes the kth instant of time at which x(t ) intersects Z
x
for a particular trajectory x(t ),
f
c
: D R
n
is Lipschitz continuous and satises f
c
(0) = 0, G
c
: D R
nm
c
, f
d
: Z
x
R
n
is continuous,
G
d
: Z
x
R
nm
d
is such that rankG
d
(x) = m
d
, x Z
x
, and Z
x
D is the resetting set. Here, we assume
that u
c
() and u
d
() are restricted to the class of admissible inputs consisting of measurable functions such that
(u
c
(t ), u
d
(t
k
)) U
c
U
d
for all t 0 and k N
[0,t )
{k : 0 t
k
< t }, where the constrained set U
c
U
d
is given with (0, 0) U
c
U
d
. We refer to the differential equation (1) as the continuous-time dynamics, and we refer
to the difference equation (2) as the resetting law.
The equations of motion for the closed-loop impulsive dynamical system (1) and (2) with hybrid adaptive feedback
controllers u
c
() and u
d
() has the form

x(t ) =

f
c
( x(t )), x(0) = x
0
, x(t ) Z
x
, (3)
x(t ) =

f
d
( x(t )), x(t ) Z
x
, (4)
where t 0, x(t )

D R
n
, x(t ) denotes the closed-loop state involving the system state and the adaptive gains,

f
c
:

D R
n
and

f
d
:

D R
n
denote the closed-loop continuous-time and resetting dynamics, respectively, with

f
c
( x
e
) = 0, where x
e


D\ Z
x
denotes the closed-loop equilibrium point, and n denotes the dimension of the closed-
loop system state. A function x : I
x
0


D is a solution to the impulsive system (3) and (4) on the interval I
x
0
R
with initial condition x(0) = x
0
, where I
x
0
denotes the maximal interval of existence of a solution to (3) and (4), if
x() is left-continuous and x(t ) satises (3) and (4) for all t I
x
0
. For further discussion on solutions to impulsive
differential equations, see [36]. For convenience, we use the notation s(t, x
0
) to denote the solution x(t ) of (3) and
(4) at time t 0 with initial condition x(0) = x
0
.
In this paper, we assume that Assumptions A1 and A2 established in [3,7] hold; that is, the resetting set is such
that resetting removes x(t
k
) from the resetting set and no trajectory can intersect the interior of Z
x
. Hence, as shown
in [3,7], the resetting times are well dened and distinct. Since the resetting times are well dened and distinct and
since the solution to (3) exists and is unique it follows that the solution of the impulsive dynamical system (3) and (4)
also exists and is unique over a forward time interval. However, it is important to note that the analysis of impulsive
dynamical systems can be quite involved. In particular, such systems can exhibit Zenoness and beating as well as
conuence, wherein solutions exhibit innitely many resettings in a nite time, encounter the same resetting surface
a nite or innite number of times in zero time, and coincide after a certain point in time. In this paper we allow for
the possibility of conuence and Zeno solutions; however, A2 precludes the possibility of beating. Furthermore, since
not every bounded solution of an impulsive dynamical system over a forward time interval can be extended to innity
due to Zeno solutions, we assume that existence and uniqueness of solutions are satised in forward time. For details
see [3].
Next, we provide a key result from [3,7,20] involving an invariant set stability theorem for hybrid dynamical
systems. For the statement of this result the following key assumption is needed.
Assumption 2.1 ([3,7,20]). Let s(t, x
0
), t 0, denote the solution of (3) and (4) with initial condition x
0


D. Then
for every x
0


D, there exists a dense subset T
x
0
[0, ) such that [0, ) \ T
x
0
is (nitely or innitely) countable
and for every > 0 and t T
x
0
, there exists (, x
0
, t ) > 0 such that if x
0
y < (, x
0
, t ), y

D, then
s(t, x
0
) s(t, y) < .
T. Hayakawa et al. / Nonlinear Analysis: Hybrid Systems 2 (2008) 862874 865
Assumption 2.1 is a generalization of the standard continuous dependence property for dynamical systems with
continuous ows to dynamical systems with left-continuous ows. Specically, by letting T
x
0
= T
x
0
= [0, ),
where T
x
0
denotes the closure of the set T
x
0
, Assumption 2.1 specializes to the classical continuous dependence of
solutions of a given dynamical system with respect to the systems initial conditions x
0


D [21]. Since solutions
of impulsive dynamical systems are not continuous in time and solutions are not continuous functions of the
system initial conditions, Assumption 2.1 is needed to apply the hybrid invariance principle developed in [7,20]
to hybrid adaptive systems. Henceforth, we assume that the hybrid adaptive feedback controllers u
c
() and u
d
() are
such that closed-loop hybrid system (3) and (4) satises Assumption 2.1. Necessary and sufcient conditions that
guarantee that the nonlinear impulsive dynamical system

G satises Assumption 2.1 are given in [3,20]. A sufcient
condition that guarantees that the trajectories of the closed-loop nonlinear impulsive dynamical system (3) and (4)
satisfy Assumption 2.1 are Lipschitz continuity of

f
c
() and the existence of a continuously differentiable function
X :

D R such that the resetting set is given by Z
x
= { x

D : X( x) = 0}, where X

( x) = 0, x Z
x
,
and X

( x)

f
c
( x) = 0, x Z
x
. The last condition above ensures that the solution of the closed-loop hybrid system
is not tangent to the resetting set Z
x
for all initial conditions x
0


D. For further discussion on Assumption 2.1,
see [3,7,20].
The following theorem proven in [7,20] is needed to develop the main results of this paper.
Theorem 2.1 ([7,20]). Consider the nonlinear impulsive dynamical system

G given by (3) and (4), assume

D
c


D is
a compact positively invariant set with respect to (3) and (4), and assume that there exists a continuously differentiable
function V :

D
c
R such that
V

( x)

f
c
( x) 0, x

D
c
, x Z
x
, (5)
V( x +

f
d
( x)) V( x), x

D
c
, x Z
x
. (6)
Let R { x

D
c
: x Z
x
, V

( x)

f
c
( x) = 0} { x

D
c
: x Z
x
, V( x +

f
d
( x)) = V( x)} and let Mdenote the
largest invariant set contained in R. If x
0


D
c
, then x(t ) Mas t . Finally, if

D = R
n
and V( x) as
x , then all solutions x(t ), t 0, of (3) and (4) that are bounded approach Mas t for all x
0
R
n
.
3. Hybrid adaptive stabilization for nonlinear hybrid dynamical systems using neural networks
In this section, we consider the problem of neural hybrid adaptive stabilization for nonlinear uncertain hybrid
systems. Specically, we consider the controlled state-dependent impulsive dynamical system (1) and (2) with
D = R
n
, U
c
= R
m
c
, and U
d
= R
m
d
, where f
c
: R
n
R
n
and f
d
: R
n
R
n
are continuously differentiable
and satisfy f
c
(0) = 0 and f
d
(0) = 0, and G
c
: R
n
R
nm
c
and G
d
: R
n
R
nm
d
.
In this paper, we assume that f
c
() and f
d
() are unknown functions, and f
c
(), G
c
(), f
d
(), and G
d
() are given by
f
c
(x) = A
c
x +f
c
(x), G
c
(x) = B
c
G
cn
(x), (7)
f
d
(x) = (A
d
I
n
)x +f
d
(x), G
d
(x) = B
d
G
dn
(x), (8)
where A
c
R
nn
, A
d
R
nn
, B
c
R
nm
c
, and B
d
R
nm
d
are known matrices, G
cn
: R
n
R
m
c
m
c
and
G
dn
: R
n
R
m
d
m
d
are known matrix functions such that det G
cn
(x) = 0, x R
n
, and det G
dn
(x) = 0, x R
n
,
and f
c
: R
n
R
n
and f
d
: R
n
R
n
are unknown functions belonging to the uncertainty sets F
c
and F
d
,
respectively, given by
F
c
= {f
c
: R
n
R
n
: f
c
(0) = 0, f
c
(x) = B
c

c
(x), x R
n
}, (9)
F
d
= {f
d
: R
n
R
n
: f
d
(0) = 0, f
d
(x) = B
d

d
(x), x R
n
}, (10)
where
c
: R
n
R
m
c
and
d
: R
n
R
m
d
are uncertain continuously differentiable functions such that
c
(0) = 0
and
d
(0) = 0. It is important to note that since
c
(x) and
d
(x) are continuously differentiable and
c
(0) = 0 and

d
(0) = 0, it follows that there exist continuous matrix functions
c
: R
n
R
m
c
n
and
d
: R
n
R
m
d
n
such
that
c
(x) =
c
(x)x, x R
n
, and
d
(x) =
d
(x)x, x R
n
. Furthermore, we assume that the continuous matrix
functions
c
() and
d
() can be approximated over a compact set D
c
R
n
by a linear in the parameters neural
866 T. Hayakawa et al. / Nonlinear Analysis: Hybrid Systems 2 (2008) 862874
Fig. 3.1. Visualization of function
j
(), j = c, d.
network up to a desired accuracy so that
col
i
(
c
(x)) = W
T
ci

c
(x) +
ci
(x), x D
c
, i = 1, . . . , n, (11)
col
i
(
d
(x)) = W
T
di

d
(x) +
di
(x), x D
c
, i = 1, . . . , n, (12)
where col
i
(()) denotes the i th column of the matrix (), W
T
ci
R
m
c
s
c
and W
T
di
R
m
d
s
d
, i = 1, . . . , n,
are optimal unknown (constant) weights that minimize the approximation error over D
c
,
ci
: R
n
R
m
c
and

di
: R
n
R
m
d
, i = 1, . . . , n, are modeling errors such that
max
(
c
(x))
1
c
and
max
(
d
(x))
1
d
,
x R
n
, where
c
(x) [
c1
(x), . . . ,
cn
(x)],
d
(x) [
d1
(x), . . . ,
dn
(x)], and
c
,
d
> 0, and
c
: R
n
R
s
c
and

d
: R
n
R
s
d
are given basis functions such that each component of
c
() and
d
() takes values between 0 and 1.
Next, dening

c
(x)
c
(x) W
T
c
[x
c
(x)], (13)

d
(x)
d
(x) W
T
d
[x
d
(x)], (14)
where W
T
c
[W
T
c1
, . . . , W
T
cn
] R
m
c
ns
c
, W
T
d
[W
T
d1
, . . . , W
T
dn
] R
m
d
ns
d
, and denotes the Kronecker product,
it follows from (11) and (12), and the CauchySchwarz inequality that

T
j
(x)
j
(x) =
j
(x)x W
T
j
(x
j
(x))
2
=
j
(x)x
j
(x)x
2
=
j
(x)x
2

2
j
x
T
x, x D
c
, j = c, d, (15)
where denotes the Euclidean norm and
j
(x) [W
T
j 1

j
(x), . . . , W
T
j n

j
(x)], j = c, d. This corresponds to a
nonlinear small gain-type norm bounded uncertainty characterization for
j
(), j = c, d (see Fig. 3.1).
Theorem 3.1. Consider the nonlinear uncertain hybrid dynamical system G given by (1) and (2) where f
c
(), G
c
(),
f
d
(), and G
d
() are given by (7) and (8), and f
c
: R
n
R
n
and f
d
: R
n
R
n
belong to the uncertainty sets
F
c
and F
d
, respectively. For given
c
,
d
> 0, assume there exists a positive-denite matrix P R
nn
such that
0 = A
T
cs
P + PA
cs
+
2
c
PB
c
B
T
c
P + I
n
+ R
c
, (16)
P = A
T
d
PA
d
A
T
d
PB
d
(B
T
d
PB
d
)
1
B
T
d
PA
d
+( +)I
n
+ R
d
, (17)
where A
cs
A
c
+ B
c
K
c
, K
c
R
m
c
n
, R
c
R
nn
and R
d
R
nn
are positive denite, > 0, and satises

2
d
_

max
(B
T
d
PB
d
) +a
1 + x
T
Px
c +[x
d
(x)]
T
[x
d
(x)]
_
, x Z
x
, (18)
T. Hayakawa et al. / Nonlinear Analysis: Hybrid Systems 2 (2008) 862874 867
where
a = max{c, n/
min
(P)}
max
_
B
T
d
PB
d
_
I
m
+
1

2
d
B
T
d
PB
d
__
(19)
and c > 0. Finally, let A
ds
A
d
+B
d
K
d
, where K
d
(B
T
d
PB
d
)
1
B
T
d
PA
d
, and let Q
c
R
m
c
m
c
and Y R
ns
c
ns
c
be positive denite. Then the neural hybrid adaptive feedback control law
u
c
(t ) = G
1
cn
(x(t ))
_
K
c
x(t )

W
T
c
(t )[x(t )
c
(x(t ))]
_
, x(t ) Z
x
, (20)
u
d
(t ) = G
1
dn
(x(t ))
_
K
d
x(t )

W
T
d
(t )[x(t )
d
(x(t ))]
_
, x(t ) Z
x
, (21)
where

W
T
c
(t ) R
m
c
ns
c
, t 0,

W
T
d
(t ) R
m
d
ns
d
, t 0, and
c
: R
n
R
s
c
and
d
: R
n
R
s
d
are given basis
functions, with update laws

W
T
c
(t ) =
1
1 + x(t )
T
Px(t )
Q
c
B
T
c
Px(t )[x(t )
c
(x(t ))]
T
Y,

W
T
c
(0) =

W
T
c0
, x(t ) Z
x
, (22)

W
T
c
(t ) = 0, x(t ) Z
x
, (23)

W
T
d
(t ) = 0,

W
T
d
(0) =

W
T
d0
, x(t ) Z
x
, (24)

W
T
d
(t ) =
1
c +[x(t )
d
(x(t ))]
T
[x(t )
d
(x(t ))]
B

d
[x(t
+
) A
ds
x(t )][x(t )
d
(x(t ))]
T
, x(t ) Z
x
,
(25)
where

W
T
c
(t )

W
T
c
(t
+
)

W
T
c
(t ) and

W
T
d
(t )

W
T
d
(t
+
)

W
T
d
(t ), guarantees that there exists a positively
invariant set D

R
n
R
m
c
ns
c
R
m
d
ns
d
such that (0, W
T
c
, W
T
d
) D

, where W
T
c
R
m
c
ns
c
and W
T
d
R
m
d
ns
d
,
and the solution (x(t ),

W
T
c
(t ),

W
T
d
(t )) (0, W
T
c
, W
T
d
) of the closed-loop system given by (1), (2) and (20)(25) is
Lyapunov stable and x(t ) 0 as t for all f
c
() F
c
, f
d
() F
d
, and (x
0
,

W
T
c0
,

W
T
d0
) D

.
Proof. First, note that
A
T
ds
PB
d
B
T
d
PA
ds
= (A
d
+ B
d
K
d
)
T
PB
d
B
T
d
P(A
d
+ B
d
K
d
)
= (A
d
B
d
(B
T
d
PB
d
)
1
B
T
d
PA
d
)
T
PB
d
B
T
d
P(A
d
B
d
(B
T
d
PB
d
)
1
B
T
d
PA
d
)
= 0, (26)
and hence, since A
T
ds
PB
d
B
T
d
PA
ds
is nonnegative denite, A
T
ds
PB
d
= 0. Furthermore, note that
P = A
T
ds
PA
ds
+( +)I
n
+ R
d
. (27)
Now, with u
c
(t ), t 0, and u
d
(t
k
), k N, given by (20) and (21), respectively, it follows from (7) and (8) that the
closed-loop hybrid system (1) and (2) is given by
x(t ) = f
c
(x(t )) + B
c
_
K
c
x(t )

W
T
c
(t )[x(t )
c
(x(t ))]
_
, x(0) = x
0
, x(t ) Z
x
, (28)
x(t ) = f
d
(x(t )) + B
d
_
K
d
x(t )

W
T
d
(t )[x(t )
d
(x(t ))]
_
, x(t ) Z
x
, (29)
or, equivalently, using (11) and (12),
x(t ) = A
cs
x(t ) + B
c
_

c
(x(t ))

W
T
c
(t )[x(t )
c
(x(t ))]
_
, x(0) = x
0
, x(t ) Z
x
, (30)
x(t ) = (A
ds
I
n
)x(t ) + B
d
_

d
(x(t ))

W
T
d
(t )[x(t )
d
(x(t ))]
_
, x(t ) Z
x
, (31)
868 T. Hayakawa et al. / Nonlinear Analysis: Hybrid Systems 2 (2008) 862874
where

W
T
c
(t )

W
T
c
(t )W
T
c
and

W
T
d
(t )

W
T
d
(t )W
T
d
. Furthermore, dene
d
(x) x
d
(x) and note that adding
and subtracting W
T
d
to and from (25) and using (31) it follows that

W
T
d
(t
+
) =

W
T
d
(t ) +
1
c +
T
d
(x(t ))
d
(x(t ))
B

d
_
B
d
[
d
(x(t ))

W
T
d
(t )
d
(x(t ))]
_
[x(t )
d
(x(t ))]
T
=

W
T
d
(t ) +
1
c +
T
d
(x(t ))
d
(x(t ))
[
d
(x(t ))

W
T
d
(t )
d
(x(t ))]
T
d
(x(t )), x(t ) Z
x
. (32)
To show Lyapunov stability of the closed-loop hybrid system (22)(24) and (30)(32), consider the Lyapunov
function candidate
V(x,

W
T
c
,

W
T
d
) = ln(1 + x
T
Px) +tr Q
1
c

W
T
c
Y
1

W
c
+atr

W
d

W
T
d
. (33)
Note that V(0, W
T
c
, W
T
d
) = 0 and, since P, Q
c
, and Y are positive denite and a > 0, V(x,

W
T
c
,

W
T
d
) > 0 for all
(x,

W
T
c
,

W
T
d
) = (0, W
T
c
, W
T
d
). In addition, V(x,

W
T
c
,

W
T
d
) is radially unbounded. Now, letting x(t ) denote the solution
to (30) and using (22) and (24), it follows that the Lyapunov derivative along the closed-loop system trajectories over
the time interval t (t
k
, t
k+1
], k N, is given by

V(x(t ),

W
T
c
(t ),

W
T
d
(t )) =
2x
T
(t )P
1 + x
T
(t )Px(t )
_
A
cs
x(t ) + B
c
_

c
(x(t ))

W
T
c
(t )[x(t )
c
(x(t ))]
__
+2tr Q
1
c

W
T
c
(t )Y
1

W
c
(t )
x
T
(t )(R
c
+
2
PB
c
B
T
c
P + I
n
)x(t )
+2x
T
(t )PB
c
_

c
(x(t ))

W
T
c
(t )[x(t )
c
(x(t ))]
_
+2tr

W
T
(t )
_
B
T
c
Px(t )[x(t )
c
(x(t ))]
T
_
T
= x
T
(t )R
c
x(t ) x
T
(t )(
2
PB
c
B
T
c
P + I
n
)x(t ) +2x
T
(t )PB
c

c
(x(t ))
x
T
(t )R
c
x(t ) [
1
B
T
c
Px(t )
c
(x(t ))]
T
[
1
B
T
c
Px(t )
c
(x(t ))]
x
T
(t )R
c
x(t )
0, t
k
< t t
k+1
. (34)
Next, using (23), (27) and (32), the Lyapunov difference along the closed-loop system trajectories at the resetting
times t
k
, k N, is given by
V(x(t
k
),

W
T
c
(t
k
),

W
T
d
(t
k
)) V(x(t
+
k
),

W
T
c
(t
+
k
),

W
T
d
(t
+
k
)) V(x(t
k
),

W
T
c
(t
k
),

W
T
d
(t
k
))
= ln
_
1 +
_
A
ds
x(t
k
) + B
d
[
d
(x(t
k
))

W
T
d
(t
k
)[x(t
k
)
d
(x(t
k
))]]
_
P
_
A
ds
x(t
k
) + B
d
[
d
(x(t
k
))

W
T
d
(t
k
)[x(t
k
)
d
(x(t
k
))]]
__
+atr
_

W
T
d
(t
k
) +
1
c +
T
d
(x(t
k
))
d
(x(t
k
))
_

d
(x(t
k
))

W
T
d
(t
k
)
d
(x(t
k
))
_

T
d
(x(t
k
))
_
T

_

W
T
d
(t
k
) +
1
c +
T
d
(x(t
k
))
d
(x(t
k
))
_

d
(x(t
k
))

W
T
d
(t
k
)
d
(x(t
k
))
_

T
d
(x(t
k
))
_
ln(1 + x
T
(t
k
)Px(t
k
)) atr

W
d
(t
k
)

W
T
d
(t
k
)
= ln
_
1 +
_
x
T
(t
k
)A
T
ds
PA
ds
x(t
k
) +2x
T
(t
k
)A
T
ds
PB
d

d
(x(t
k
))
2x
T
(t
k
)A
T
ds
PB
d

W
T
d
(t
k
)
d
(x(t
k
)) +
T
d
(x(t
k
))B
d
PB
d

d
(x(t
k
))
2
T
d
(x(t
k
))B
d
PB
d

W
T
d
(t
k
)
d
(x(t
k
)) +
T
d
(x(t
k
))

W
d
(t
k
)B
d
PB
d

W
T
d
(t
k
)
d
(x(t
k
))
T. Hayakawa et al. / Nonlinear Analysis: Hybrid Systems 2 (2008) 862874 869
x
T
(t
k
)Px(t
k
)
_ _
1 + x
T
(t
k
)Px(t
k
)
_
1
_
+atr

W
d
(t
k
)

W
T
d
(t
k
)
+
2a
c +
T
d
(x(t
k
))
d
(x(t
k
))
tr W
d
(t
k
)
_

d
(x(t
k
))

W
T
d
(t
k
)
d
(x(t
k
))
_

T
d
(x(t
k
))
+
a
(c +
T
d
(x(t
k
))
d
(x(t
k
)))
2
tr
d
(x(t
k
))
_

T
d
(x(t
k
))
T
d
(x(t
k
))

W
d
(t
k
)
_ _

d
(x(t
k
))

W
T
d
(t
k
)
d
(x(t
k
))
_

T
d
(x(t
k
)) atr

W
d
(t
k
)

W
T
d
(t
k
)

_
x
T
(t
k
)(( +)I
n
+ R
d
)x(t
k
) +
T
d
(x(t
k
))B
T
d
PB
d

d
(x(t
k
)) 2
T
d
(x(t
k
))B
T
d
PB
d

W
T
d
(t
k
)
d
(x(t
k
))
+
T
d
(x(t
k
))

W
d
(t
k
)B
T
d
PB
d

W
T
d
(t
k
)
d
(x(t
k
))
_ _
1 + x
T
(t
k
)Px(t
k
)
_
1
+
2a
c +
T
d
(x(t
k
))
d
(x(t
k
))
tr

W
d
(t
k
)
_

d
(x(t
k
))

W
T
d
(t
k
)
d
(x(t
k
))
_

T
d
(x(t
k
))
+
a
c +
T
d
(x(t
k
))
d
(x(t
k
))
tr
_

T
d
(x(t
k
))
T
d
(x(t
k
))

W
d
(t
k
)
_ _

d
(x(t
k
))

W
T
d
(t
k
)
d
(x(t
k
))
_

_
x
T
(t
k
)(( +)I
n
+ R
d
)x(t
k
) +
T
d
(x(t
k
))B
T
d
PB
d

d
(x(t
k
)) 2
T
d
(x(t
k
))B
T
d
PB
d

W
T
d
(t
k
)
d
(x(t
k
))
+
T
d
(x(t
k
))

W
d
(t
k
)B
T
d
PB
d

W
T
d
(t
k
)
d
(x(t
k
))
_ _
1 + x
T
(t
k
)Px(t
k
)
_
1
+
2a
c +
T
d
(x(t
k
))
d
(x(t
k
))
tr

W
d
(t
k
)
_

d
(x(t
k
))

W
T
d
(t
k
)
d
(x(t
k
))
_

T
d
(x(t
k
))
+
a
c +
T
d
(x(t
k
))
d
(x(t
k
))
tr
_

T
d
(x(t
k
))
T
d
(x(t
k
))

W
d
(t
k
)
_ _

d
(x(t
k
))

W
T
d
(t
k
)
d
(x(t
k
))
_
, (35)
where in (35) we used ln a ln b = ln
a
b
and ln(1 +d) d for a, b > 0, and d > 1, respectively, and

T
d

d
c+
T
d

d
< 1.
Furthermore, note that
T
d
(x)
d
(x) nx
T
x.
Now, dening
1

2
d
(B
T
d
PB
d
)
2
, it follows from (35) that
V(x(t
k
),

W
T
c
(t
k
),

W
T
d
(t
k
))
_
x
T
(t
k
)R
d
x(t
k
) x
T
(t
k
)x(t
k
) [x
T
(t
k
)x(t
k
)
2
d

T
d
(x(t
k
))
d
(x(t
k
))]

T
d
(x(t
k
)),
T
d
(x(t
k
))

W
d
(t
k
)
_
_

2
d
I
n
B
T
d
PB
d
B
T
d
PB
d

_ _

d
(x(t
k
))

W
T
d
(t
k
)
d
(x(t
k
))
_
+
T
d
(x(t
k
))

W
d
(t
k
)

W
T
d
(t
k
)
d
(x(t
k
))
+
T
d
(x(t
k
))

W
d
(t
k
)B
T
d
PB
d

W
T
d
(t
k
)
d
(x(t
k
))
_
_
1 + x
T
(t
k
)Px(t
k
)
_
1

a
c +
T
d
(x(t
k
))
d
(x(t
k
))

T
d
(x(t
k
))

W
d
(t
k
)

W
T
d
(t
k
)
d
(x(t
k
))
+
a
c +
T
d
(x(t
k
))
d
(x(t
k
))

T
d
(x(t
k
))
d
(x(t
k
))

x
T
(t
k
)R
d
x(t
k
)
1 + x
T
(t
k
)Px(t
k
)

T
d
(x(t
k
))

W
d
(t
k
)

R
d1
(x(t
k
))

W
T
d

d
(x(t
k
))
(c +
T
d
(x(t
k
))
d
(x(t
k
)))(1 + x
T
(t
k
)Px(t
k
))

T
d
(x(t
k
))

R
d2
(x(t
k
))
d
(x(t
k
))
(c +
T
d
(x(t
k
))
d
(x(t
k
)))(1 + x
T
(t
k
)Px(t
k
))
, (36)
where
870 T. Hayakawa et al. / Nonlinear Analysis: Hybrid Systems 2 (2008) 862874

R
d1
(x) a(1 + x
T
Px)I
m
(B
T
d
PB
d
+)(c +
T
d
(x)
d
(x))
a(1 + x
T
Px)I
m
B
T
d
PB
d
_
I
m
+
1

2
d
B
T
d
PB
d
_
(c +nx
T
x)
0, x D
c
, (37)
and

R
d2
(x)
2
d
(c +
T
d
(x)
d
(x))I
m
B
T
d
PB
d
(c +
T
d
(x)
d
(x)) a(1 + x
T
Px)I
m
(c +
T
d
(x)
d
(x))
_

2
d

max
(B
T
d
PB
d
) a
1 + x
T
Px
c +
T
d
(x)
d
(x)
_
I
m
0, x D
c
. (38)
Hence, the Lyapunov difference given by (36) yields
V(x(t
k
),

W
T
c
(t
k
),

W
T
d
(t
k
))
x
T
(t
k
)R
d
x(t
k
)
1 + x
T
(t
k
)Px(t
k
)

T
d
(x(t
k
))

W
d
(t
k
)

R
d
(x(t
k
))

W
T
d

d
(x(t
k
))
(c +
T
d
(x(t
k
))
d
(x(t
k
)))(1 + x
T
(t
k
)Px(t
k
))

x
T
(t
k
)R
d
x(t
k
)
1 + x
T
(t
k
)Px(t
k
)
0, k N. (39)
Next, let

_
(x,

W
T
c
,

W
T
d
) R
n
R
m
c
ns
c
R
m
d
ns
d
: V(x,

W
T
c
,

W
T
d
)
_
, (40)
where is the maximum value such that

D

D
c
R
m
c
ns
c
R
m
d
ns
d
. Since V(x(t
k
),

W
T
c
(t
k
),

W
T
d
(t
k
)) 0 for
all (x(t
k
),

W
T
c
(t
k
),

W
T
d
(t
k
))

D

and k N, it follows that

D

is positively invariant. Next, since

D

is positively
invariant, it follows that
D

_
(x,

W
T
c
,

W
T
d
) R
n
R
m
c
ns
c
R
m
d
ns
d
: (x,

W
T
c
W
T
c
,

W
T
d
W
T
d
)

D

_
(41)
is also positively invariant. Now, it follows from Theorem 2.1 of [3] that (34) and (39) imply that the solution
(x(t ),

W
T
c
(t ),

W
T
d
(t )) (0, W
T
c
, W
T
d
) to (22)(24), (30) and (32) is Lyapunov stable. Furthermore, since R
c
> 0
and R
d
> 0, it follows from Theorem 2.1, with R = M = {(x,

W
T
c
,

W
T
d
) R
n
R
m
c
s
c
R
m
d
s
d
: x = 0}, that
x(t ) 0 as t for all x
0
R
n
.
Remark 3.1. Note that the conditions in Theorem 3.1 imply partial asymptotic stability, that is, the solution
(x(t ),

W
T
c
(t ),

W
T
d
(t )) (0, W
T
c
, W
T
d
) of the overall closed-loop system is Lyapunov stable and x(t ) 0 as t .
Hence, it follows from (22) and (23) that

W
T
c
(t ) 0 as t . Furthermore, if x(t ), t 0, intersects Z
x
innitely
many times, then it follows from (24) and (25) that

W
d
(t
+
k
)

W
d
(t
k
) 0 as k .
Remark 3.2. Since the Lyapunov function used in the proof of Theorem 3.1 is a class K

function, in the case where

the neural network approximation holds in R
n
, the control law (20) and (21) ensures global asymptotic stability with
respect to x. However, the existence of a global neural network approximator for an uncertain nonlinear map cannot
in general be established. Hence, as is common in the neural network literature, for a given arbitrarily large compact
set D
c
R
n
, we assume that there exists an approximator for the unknown nonlinear map up to a desired accuracy
(in the sense of (11) and (12)). In the case where
c
() and
d
() are continuous on R
n
, it follows from the Stone-
Weierstrass theorem that
c
() and
d
() can be approximated over an arbitrarily large compact set D
c
. In this case,
our neuroadaptive hybrid controller guarantees semiglobal partial asymptotic stability.
Remark 3.3. Note that the neuroadaptive hybrid controller (20) and (21) can be constructed to guarantee partial
asymptotic stability using standard linear H

theory. Specically, it follows from standard continuous-time H

theory [22] that G

c
(s)

<
c
, where G(s) = E
c
(s I
n
A
cs
)
1
B
c
and E
c
is such that E
T
c
E
c
= I
n
+ R
c
, if and only
T. Hayakawa et al. / Nonlinear Analysis: Hybrid Systems 2 (2008) 862874 871
if there exists a positive-denite matrix P satisfying the bounded real Riccati equation (16). It is important to note
that
c
> 0 and
d
> 0, which characterize the approximation error (13) and (14), respectively, over D
c
, can be made
arbitrarily large provided that we take a large number of basis functions in the parameterization of the uncertainty

c
() and
d
(). In this case, noting that
1+x
T
Px
c+[x
d
(x)]
T
[x
d
(x)]
in (18) is a bounded positive function, it can be shown
that there always exist and such that the conditions (16)(19) are satised.
It is important to note that the hybrid adaptive control law (20)(25) does not require explicit knowledge of the
optimal weighting matrices W
c
, W
d
, and the positive constants and . Theorem 3.1 simply requires the existence of
W
c
, W
d
, , and such that (16) and (17) hold. Furthermore, no specic structure on the nonlinear dynamics f
c
(x) and
f
d
(x) is required to apply Theorem 3.1 other than the assumption that f
c
(x) and f
d
(x) are continuously differentiable
and that the approximation error of the uncertain system nonlinearities lie in a small gain-type norm bounded conic
sector. Finally, in the case where the pair (A
d
, B
d
) is in controllable canonical form and R
d
in (17) is diagonal, it
follows that A
ds
=
_
A
0
0
m
d
n
_
, where A
0
R
(nm
d
)n
is a known matrix of zeros and ones capturing the multivariable
controllable canonical form representation [23], and hence, the update law (25) is simplied as

W
T
d
(t ) =
1
c +[x(t )
d
(x(t ))]
T
[x(t )
d
(x(t ))]
B

d
x(t )[x(t )
d
(x(t ))]
T
, x(t ) Z
x
, (42)
since B

d
A
ds
= 0.
4. Illustrative numerical example
In this section, we present a numerical example to demonstrate the utility of the proposed neural hybrid adaptive
control framework for hybrid adaptive stabilization. Specically, consider the nonlinear uncertain controlled hybrid
system given by (1) and (2) with n = 2, x = [x
1
, x
2
]
T
,
f
c
(x) =
_
x
2

f
c
(x)
_
, G
c
(x) =
_
0
b
c
_
, f
d
(x) =
_
x
1
+ x
2

f
d
(x)
_
, G
d
(x) =
_
0
b
d
_
, (43)
where

f
c
: R
2
R and

f
d
: R
2
R are unknown, continuously differentiable functions. Furthermore, assume that
the resetting set Z
x
is given by
Z
x
= {x R
2
: X(x) = 0, x
2
> 0}, (44)
where X : R
2
R is a continuously differentiable function given by X(x) = x
1
. Here, we assume that f
c
(x) and
f
d
(x) are unknown and can be written in the form of (7) and (8) with
A
c
= A
d
=
_
0 1
0 0
_
,
f
c
(x) = [0,

f
c
(x)]
T
, f
d
(x) = [0,

f
d
(x)]
T
, B
c
= [0, b
c
]
T
, B
d
= [0, b
d
]
T
, G
cn
(x) = G
dn
(x) = 1. We assume
that f
c
(x) and f
d
(x) are unknown and can be written as f
c
(x) = B
c

c
(x) and f
d
(x) = B
d

d
(x), where

c
(x) =
1
b
c

f
c
(x) and
d
(x) =
1
b
d

f
d
(x).
Next, let K
c
=
1
b
c
[k
c1
, k
c2
] and K
d
=
1
b
d
[k
d1
, k
d2
], where k
c1
, k
c2
, k
d1
, and k
d2
are arbitrary scalars, such that
A
cs
= A
c
+ B
c
K
c
=
_
0 1
k
c1
k
c2
_
,
A
ds
= A
d
+ B
d
K
d
=
_
0 1
k
d1
k
d2
_
.
Now, with the proper choice of k
c1
, k
c2
, k
d1
, and k
d2
, it follows from Theorem 3.1 that if there exists P > 0 satisfying
(16) and (17), then the neural hybrid adaptive feedback controller (20) and (21) guarantees x(t ) 0 as t .
Specically, here we choose k
c1
= 1, k
c2
= 1, k
d1
= 0.2, k
d2
= 0.5,
c
= 10,
d
= 20, b
c
= 3, b
d
= 1.4,
c = 1, = 1,
d
(x) = [tanh(0.1x
2
), . . . , tanh(0.6x
2
)]
T
, and
872 T. Hayakawa et al. / Nonlinear Analysis: Hybrid Systems 2 (2008) 862874
Fig. 4.1. Phase portraits of uncontrolled and controlled hybrid system.
Fig. 4.2. State trajectories versus time.
R
c
=
_
2.6947 2.4323
2.4323 5.8019
_
, R
d
=
_
8.0196 2.0334
2.0334 1.0569
_
, (45)
so that P satisfying (16) and (17) is given by
P =
_
10.0196 2.0334
2.0334 12.7523
_
.
With

f
c
(x) = a
1
x
1
a
2
(x
2
1
a
3
)x
2
,

f
d
(x) = x
2
a
4
x
2
1
a
5
x
3
2
1+x
2
2
a
6
x
3
2
, a
1
= 1, a
2
= 2, a
3
= 1, a
4
= 5,
a
5
= 2, a
6
= 8, Y = 0.02I
3
,
c
(x) =
_
1
1+e

1
x
1
, . . . ,
1
1+e
3
1
x
1
,
1
1+e

2
x
2
, . . . ,
1
1+e
3
2
x
2
_
, and initial conditions
T. Hayakawa et al. / Nonlinear Analysis: Hybrid Systems 2 (2008) 862874 873
Fig. 4.3. Control signals versus time.
Fig. 4.4. Adaptive gain history versus time.
x(0) = [1, 1]
T
,

W
T
c
(0) = 0
16
, and

W
T
d
(0) = 0
16
, Fig. 4.1 shows the phase portraits of the uncontrolled and
controlled hybrid system. Figs. 4.2 and 4.3 show the state trajectories versus time and the control signals versus time,
respectively. Finally, Fig. 4.4 shows the adaptive gain history versus time.
874 T. Hayakawa et al. / Nonlinear Analysis: Hybrid Systems 2 (2008) 862874
5. Conclusion
A direct hybrid neuroadaptive nonlinear control framework for hybrid nonlinear uncertain dynamical systems was
developed. Using Lyapunov methods the proposed framework was shown to guarantee partial asymptotic stability
of the closed-loop hybrid system; that is, asymptotic stability with respect to part of the closed-loop system states
associated with the hybrid plant dynamics. In the case where the nonlinear hybrid system is represented in normal
form, the nonlinear hybrid adaptive controller was constructed without requiring knowledge of the system dynamics.
Finally, a numerical example was presented to show the utility of the proposed hybrid adaptive stabilization scheme.
Acknowledgments
This research was supported in part by the Japan Science and Technology Agency under CREST program and the
Air Force Ofce of Scientic Research under Grant FA9550-06-1-0240.
References
[1] P.J. Antsaklis, A. Nerode (Eds.), Special issue on hybrid control systems, IEEE Trans. Autom. Control 43 (4) (1998).
[2] A.S. Morse, C.C. Pantelides, S. Sastry, J.M. Schumacher (Eds.), Special issue on hybrid control systems, Automatica 35 (3) (1999).
[3] W.M. Haddad, V. Chellaboina, S.G. Nersesov, Impulsive and Hybrid Dynamical Systems: Stability, Dissipativity, and Control, Princeton
University Press, Princeton, NJ, 2006.
[4] V. Lakshmikantham, D.D. Bainov, P.S. Simeonov, Theory of Impulsive Differential Equations, World Scientic, Singapore, 1989.
[5] D.D. Bainov, P.S. Simeonov, Systems with Impulse Effect: Stability, Theory and Applications, Ellis Horwood Limited, England, 1989.
[6] A.M. Samoilenko, N. Perestyuk, Impulsive Differential Equations, World Scientic, Singapore, 1995.
[7] W.M. Haddad, V. Chellaboina, N.A. Kablar, Nonlinear impulsive dynamical systems part I: Stability and dissipativity, Int. J. Control 74
(2001) 16311658.
[8] W.M. Haddad, T. Hayakawa, S.G. Nersesov, V. Chellaboina, Hybrid adaptive control for nonlinear impulsive dynamical systems, Int. J. Adapt.
Control Signal Process. 19 (6) (2005) 445469.
[9] K.J.

Astr om, B. Wittenmark, Adaptive Control, Addison-Wesley, Reading, MA, 1989.
[10] P.A. Ioannou, J. Sun, Robust Adaptive Control, Prentice-Hall, Upper Saddle River, NJ, 1996.
[11] K.S. Narendra, A.M. Annaswamy, Stable Adaptive Systems, Prentice-Hall, Englewood Cliffs, NJ, 1989.
[12] M. Krsti c, I. Kanellakopoulos, P.V. Kokotovi c, Nonlinear and Adaptive Control Design, Wiley, New York, 1995.
[13] G.C. Goodwin, K.S. Sin, Adaptive Filtering Prediction and Control, Englewood Cliffs, NJ, Prentice-Hall, 1984.
[14] F.L. Lewis, S. Jagannathan, A. Yesildirak, Neural Network Control of Robot Manipulators and Nonlinear Systems, Taylor & Francis, London,
UK, 1999.
[15] J. Spooner, M. Maggiore, R. Ordonez, K. Passino, Stable Adaptive Control and Estimation for Nonlinear Systems: Neural and Fuzzy
Approximator Techniques, John Wiley & Sons, New York, NY, 2002.
[16] S.S. Ge, C. Wang, Adaptive neural control of uncertain MIMO nonlinear systems, IEEE Trans. Neural Networks 15 (3) (2004) 674692.
[17] F.C. Chen, H.K. Khalil, Adaptive control of a class of nonlinear discrete-time systems using neural networks, IEEE Trans. Autom. Control 40
(5) (1995) 791801.
[18] S. Jagannathan, F.L. Lewis, Discrete-time neural net controller for a class of nonlinear dynamical systems, IEEE Trans. Autom. Control 41
(11) (1996) 16931699.
[19] S.S. Ge, T.H. Lee, G.Y. Li, J. Zhang, Adaptive NN control for a class of discrete-time non-linear systems, Int. J. Control 76 (4) (2003)
334354.
[20] V. Chellaboina, S.P. Bhat, W.M. Haddad, An invariance principle for nonlinear hybrid and impulsive dynamical systems, Nonlinear Anal.
TMA 53 (2003) 527550.
[21] M. Vidyasagar, Nonlinear Systems Analysis, Prentice-Hall, Englewood Cliffs, NJ, 1993.
[22] J.C. Willems, Least squares stationary optimal control and the algebraic Riccati equation, IEEE Trans. Autom. Control 16 (6) (1971) 621634.
[23] C.-T. Chen, Linear System Theory and Design, Holt, Rinehart, and Winston, New York, 1984.