Sei sulla pagina 1di 7

2011 Third International Conference on Computational Intelligence, Modelling & Simulation

Fuzzy Gain-Scheduling Nonlinear Parametric Uncertain System


Ebrahim A. Mattar
Department of Electrical and Electronics Engineering
College of Engineering
University of Bahrain
P.O. Box 13184
Kingdom of Bahrain
ebrgallaf@eng.uob.bh

Abstract This paper has presented two main issues related to


H robust fuzzy control. The first has been fuzzy modeling of
nonlinear dynamical systems, whereas the second was directed
towards H fuzzy gain-scheduling control systems. Regarding
fuzzy modeling, that was achieved by employing TakagiSugeno (T-S) fuzzy modeling technique. Employed (T-S)
modeling technique was able to cluster an entire nonlinear
global model into linear sub-models. With respect to the H
fuzzy gain-scheduling, the paper first presented an approach
for designing H fuzzy controller for disturbance rejection via
defining a suitable Lyapunov potential function of the fuzzy
model, hence designing a controller by reducing the problem
to a standard Linear Matrix Inequalities (LMI) formulation.
H fuzzy gain-scheduling was achieved via treating the (T-S)
fuzzy sub-models as a Linear Parameter Varying (LPV)
system, hence synthesizing a scheduling controller for variation
in parameters.
Keywords- Hfuzzy; robust control; Takagi-Sugeno, Time
Varying systems

I.
INTRODUCTION
Based on the stability conditions, model-based control of
T-S systems has been developed for the discrete case [1], [5],
[7] in addition to the continuous case [9], [10]. Tanaka and
Sugeno [5] have provided a sufficient condition for the
asymptotic stability of a fuzzy system in the sense of
Lyapunov through the existence of a common Lyapunov
function for all the subsystems. Tanaka and Sano [7] have
extended this to robust stability in case of systems with
premise-parameter uncertainty. Tanaka et al. [3] suggest the
idea of using linear matrix inequalities LMI for finding the
common Lyapunov P matrix where in [2] an iterative
algorithm for the choice of such common P matrix is
proposed.
A further and a significant step has also been taken to
utilize Lyapunov function based control design techniques
to the control synthesis problem for T-S models. The socalled Parallel Distributed Compensation (PDC) ( J. Li et al.
[11], [12], [13], D. Niemann et al. [9], H. Wang et al. [14])
is one such control design framework that has been
proposed and developed over the last few years. It has been
shown that within the framework of T-S fuzzy model and
PDC control design, design conditions for the stability and
performance of a system can be stated in terms of the
feasibility of a set of linear matrix inequalities (LMIs) ( J. Li
et al. [11], [12], D. Niemann et al. [9], H. Wang et al. [14]).

978-0-7695-4562-2/11 $26.00 2011 IEEE


DOI 10.1109/CIMSim.2011.31

Khaled H. Al Mutib
Department of Computer Science
College of Computer and Information Sciences
King Saud University
P. O. Box 51178
Kingdom of Saudi Arabia
mutib@ccis.ksu.edu.sa

This is a significant finding in the sense that there exist


very efficient numerical algorithms for determining the
feasibility of LMIs, so even large-scale analysis and design
problems are computationally tractable. In the conventional
optimal control, the plant model must be known beforehand.
However, the designer has to solve a Hamilton-Jacobi
equation, which is a nonlinear partial differential equation
[15]. Only some very special nonlinear systems have a
closed-form solution. A hybrid fuzzy controller is
introduced to stabilize the nonlinear system, and at the same
time to eliminate effects of external disturbance below a
prescribed level, so that a desired H control performance
can be guaranteed, B. Chen [16]. In this approach only a
linear fuzzy control design is used, although, the same
H control performance is achieved. This attempt is made
to create a bridge between two important control design
techniques, i.e., robust and fuzzy control design, so as to
supply H control design with more intelligence and fuzzy
control design with better performance respectively (B.
Chen et al. [4], B. Chen [16], K. Tanaka et al.[3]).
In recent years, LMIs have emerged as powerful
mathematical tools for the control engineers. However,
many researchers expressed their results in LMI rather than
classical AREs.
Within the development of fast
optimization algorithms for LMIs, the classical AREs
results are more easily solved through their counterparts
expressed in LMIs. For instance, J. Li et al. [18] shows the
relation between LMIs and AREs through absolute stability
criteria, robustness analysis and optimal control. This
analysis of robust stability of fuzzy control systems via
quadratic stabilization, H control theory is solved by LMI
techniques. This LMI based technique is not only an
analysis tool; it can also be utilized to automate the
controller design process. There are a few numbers of
research such as, P. Korba et al. [23], [17] are proposed a
fuzzy controller via gain-scheduling method for the T-S
fuzzy model based on Lyapunov method and LMI
techniques.
II. T-S FUZZY MODELS
A rule with scheduling variables
Rule i:
then

127

j (t )

can be written as :

if 1 (t ) is Mi1 and j (t ) is Mij


(1)
x& i (t ) = A i x(t ) + B i u(t ) + w (t), y i (t ) = C i x(t )

For a given pair of vectors x(t) and u(t), the final output
of the fuzzy system is hence inferred as a weighted sum of
the contributing sub-models :
r

x& (t ) =

( (t )){ A x(t ) + B u(t ) }


i

(2)

i =1
r

( (t ))
i

i =1
r

y(t ) =

( (t )){C x(t ) + D u(t )}


i

(3)

i =1
r

( (t ))

III. LPV: LINEAR PARAMETER VARYING


SYSTEMS
Firstly, LTI systems are described as :
x& = Ax + Bu
(8)
y = Cx + Du
Secondly, Linear Time-Varying (LTV) are described by:
x& = A(t ) x + B(t ) u
(9)
y = C(t ) x + D(t ) u

i =1

th

i ( (t )) is degree of fulfillment of an i rule. Equ. (2) and

Equ. (3) can be written as :


r

x& (t ) = hi ( (t )){A i x(t ) + B i u(t ) }


y (t ) =

i =1
r

h ( (t )){ C x(t ) + D u(t )}


i

(4)
(5)

Finally, Linear Parameter-Varying (LPV) systems, where


the state-space model entries A(.), B(.), C(.) and D(.) are
explicit functions of a time-varying parameter (t ) :
x& = A ( (t )) x + B( (t )) u
(10)
y = C( (t )) x + D( (t )) u

i =1

T-S fuzzy model can also be regarded as a quasi-linear


system, i.e., a linear system in both x(t) and u(t) whose
matrices A(.), B(.), C(.) and D(.) are not constant, but
varying as the operating condition of the system changes.
This is given as:
x& (t ) = A ( (t )) x (t ) + B ( (t )) u (t )
(6)
y (t ) = C( (t )) x (t ) + D( (t )) u (t )

From Equs. (2-3) one can observe that for all possible
values of (t ) , these are bounded within a polytope whose
vertices are the matrices of the individual rules. The
parameter dependence is affine; that is, the fuzzy state
model A( (t )), B( (t )), C( (t )), D( (t )) depends affinally on
(t ) . This means the time-varying parameter (t ) varies in
a polytope of vertices 1 , 2 ,...,

Co

{S i

: i = 1, 2 ,..., L } =

L
i =1

Bi
: i = 1,..., L
D i

CONTROL

We are interested in a class of LPV systems where : (i )


Parameter-dependence is affine, that is, state-space matrices
A ( (t )), B( (t )), C( (t )), D( (t )) depend affinely on (t ) . (ii )
The time-varying parameter (t ) varies in a polytope of
vertices 1 , 2 ,..., L . That is:
(13)
(t ) := Co { 1 , 2 , ..., L }

(7)

and
A i
Si =
C i

u = C K ( (t )) x + D K ( (t )) y

IV. LPV H

i (t ) S i : i =1 i (t ) = 1, i (t ) 0
L

y = C( (t )) x + D( (t ))u

(t ) , t 0. One of the most significant potentiality


of the LPV framework is the derivation of LPV or selfscheduled controllers. Such controllers have the same
parameter-dependence as the system and can be described in
state-space form as :
x& = A K ( (t )) x + B K ( (t )) y
(12)

Time varying parameter (t ) is defined in terms of,


in which L is the number of
(t ) = Co{ 1 , 2 ,..., L }
vertices in a polytope. A state-space of Equ. (6) are given in
terms of a number of models depending on (t ) :
A i
A ( (t )) B ( (t ))
C ( (t )) D ( (t )) Co C

Hence, LPV system is well-defined whenever its


parameter-dependence and its operating domain are fixed.
This can be expressed more formally by the set of statespace relations as,
x& = A( (t )) x + B( (t ))u
(11)

Bi
D i

Scheduling variables are function of the state, hence


(t ) = (x(t )) and x& = A( x )x + B( x ) u .

these vertices represent the extremal values of the


parameters. This description encompasses many practical
situations. From this characterization, it is clear that the
state-space matrices A( (t )), B( (t )), C( (t )), D( (t )) involved
in a polytope of matrices whose vertices are the images of
the vertices 1 , 2 ,..., L . In other words our fuzzy model is
put as :
(14)
A i B i
A ( ) B( )
C( ) D( ) Co C D : i = 1,..., L

i
i

A i
C
i
128

B i A( i ) B( i )
=
D i C( i ) D( i )

(15)

For LPV system defined by the state-space equation as :


x& = A( (t )) x + B( (t )) u
y = C( (t )) x + D( (t )) u

(16)

has quadratic H Performance if and only if there


exists a Lyapunov function V ( x ) such that V ( x ) = x T Px
with P > 0 that establishes global stability and the L2 gain
of the input/output map is bounded by . That is :
(17)
y < u
2

along all possible parameter trajectories (t ) in .

V. LFT :

LINEAR

(B
be

CONTROLLER

T
2

T
and (C 2 , D 21 ) , respectively, the
, D12

synthesized.

There

exists an

H controller can
LPV

controller

guaranteeing Quadratic H Performance along all


parameter trajectories in the polytope [22]. This is

achieved if and only if there exist two symmetric matrices


n n

(R, S) in
formulation of :

satisfying the system of (2r +1) LMIs

Ai R + RATi
NR 0
0 I C1i R

B1Ti

An augmented LPV plant need to be considered, mapping


exogenous inputs w and control inputs u to the controlled
outputs z and measured outputs y. Such formulation is given
by the standard Linear Fractional Transformation (LFT) and
given in terms of LPV as :

RC1Ti B1i
N 0
I D11i R < 0 i = 1,2,..., L
0 I
T
D11i I

AiS + SATi SCT1i B1i


NS 0
N 0
I D11i S < 0 i =1,2,...,L
0 I C1i S
0 I

BT
T
D11i I
1i

R I
I S 0

(18)

z = C1 ( (t )) x + D11 ( (t )) w + D12 ( (t )) u

Considering a continuous LPV polytopic system of Equ.


(18) and working under the three assumptions in section 6.4,
and by letting NR and NS denote bases of the null space of

Fractional TRANSFORMATION

x& = A ( (t )) x + B1 ( (t )) w + B 2 ( (t )) u

LMI FORMULATION OF LPV H

VI.

(21)

(22)

(23)

y = C 2 ( (t )) x + D 21 ( (t )) w + D 22 ( (t )) u

(t ) := Co{1 , 2 ,..., L }, for t 0 .


assumed to be polytopic, i.e. :

A plant is further

Ai B1i B2i
A((t)) B1((t)) B2((t))
C ((t)) D ((t)) D ((t))P:= CoC D D , i =1,2...,L
1i 11i 12i
11
12
1

C D D
C2((t)) D21((t)) D22((t))
2i 21i 22i

where r is the number of vertices. Since these conditions


are LMIs in the variables R and S, they are convex and fall
into the scope of efficient convex optimization techniques.

(19)

in which A i , B1i ,... denote the values of A( (t )), B1 ( (t )),... at


the vertices (t ) = i of the parameter polytope.
With
reference to Equ. (19) the problem dimensions are thus
given by : A( (t )) nn , D11 ( (t )) p m , D 22 ( (t )) p m
1

With such notations and assumptions, the H control


problem for LPV systems can be stated as follows;
Synthesis a LPV robust controller of state-space form of :
x& = A K ( (t )) x + B K ( (t )) u
y = C K ( (t )) x + D K ( (t )) u

Figure 1. Output feedback H fuzzy controller.

(20)

which guarantees Quadratic H Performance for the


closed-loop system, in such away to ensure the following
control objectives: The closed-loop system is quadratically
stable over . The L2-induced of the operator mapping w
into z bounded by for all possible trajectories (t ) in .
To find the controller of Equ. (20), the LMI technique has
been used in this respect as will be presented in Section(V).

For , R and S solutions to the LMIs Equs. (21-23),


there always exist LPV polytopic controllers solving the
problem [25]. In turn, such controllers are described by a
system of LMIs from which one can extract a particular
solution by algebraic manipulations. More precisely, along
some trajectory (t ) in the polytope , i.e :
r

(t ) = i (t ) i
i =1

129

(24)

The state-space matrices:


A K ( (t )), B K ( (t )), C K ( (t )), D K ( (t )) of the LPV polytopic

controller Equ. (28) into the state equation of the fuzzy


model Equ. (4). The closed-loop system is then given by :
r

controllers are then read as :

x& (t ) = hi ( (t ))hj ( (t )){(Ai Bi (Ci + Ki )) x(t ) + Bi r(t )}

(30)

i=1 j =1

A K ( (t )) B K ( (t )) r
A Ki
C ( (t )) D ( (t )) = i (t ) C
i
=
1
K
K

Ki

B Ki
D Ki

(25)

i s are computed according to the convex decomposition


Equ. (24). State-space data matrices A Ki , B Ki , C Ki , D Ki are
computed off-line, the LPV controller matrices
A K ( ), B K ( ), C K ( ), D K ( ) must be updated in real time
depending on the parameter measurement (t ) .
VII.

G ij + G ji

i =1

i< j

x& (t ) = hi ( (t )) hi ( (t )) G ii x(t ) + 2 hi ( (t )) h j ( (t ))

FUZZY GAIN-SCHERDULER

Rule i: if 1 (t ) is Mi1 and j (t ) is Mij


THEN u i (t ) = r(t ) y (t ) K i x(t )

(26)

(27)

i =1
r

( (t ))
i

i =1

and can be expressed as :


r

u (t ) = hi ( (t )){r (t ) y (t ) K i x(t )}
i =1

= r(t ) [K ( (t )) + C( (t ))] x(t )

(28)

If the scheduling vector (t ) is a function of the state x(t),


then u(t) represents a nonlinear gain-scheduled control law.
Hence, the goal of the controller design is to determine the
constant matrix Ki such that the desired dynamic of the
closed-loop system and the desired steady-state, inputoutput behavior are obtained. Designing the state-feedback
gains Ki requires dealing with the system dynamics and
hence ensuring stability. This problem is solved by means
of LMI. For the T-S fuzzy controller in Equ. (26), a feedforward gains Vi can be also included in the design and can
be given as :

V i = C i ( A i + B i K

1
Bi)
i )

i =1

j =1

(31)

(32)

i =1

Fuzzy H Gain-Scheduler :
With reference to Fig. 3., the system receives input signal
u(t), which has already computed based on the controller
gain K. It will be necessary to choose appropriate values of
Ki in. In section 6 we introduced the H control under
which the system to be controlled described as LPV system.
Fuzzy gain-scheduling is introduced here in twofold, The
first is via the use of different fuzzy state space sub-models
(i.e. T-S models) to designate the parameters that are
changing. This will formulate the bases of constructing a
suitable LPV system based on fuzzy models for the entire
operating region.
H controller is then synthesized in
according to LPV systems.
The entire fuzzy system can then be described as:
x& (t ) = A w (.) x(t ) + B w (.) r (t )

(34)

y w (t ) = C w (.) x (t )

(35)

Based on LPV theory, fuzzy state-space are given by:


r

A w (.) = hi ( (t ))h j ( (t )) A i B i (Ci + K j )

(36)

i =1 j =1

B w (.) = hi ( (t )) B i

(29)

Closed-Loop Fuzzy Dynamics :


The closed-loop system consisting of the fuzzy model and
the fuzzy controller is obtained by substituting the fuzzy

x(t )

For the particular case of common matrices Bi, i.e., Bi=B for
all sub-models i=1,2,,r and for the shared rules, the
following simplified description of the entire closed loop
system can be derived as :
r
x& (t ) =
h ( (t )) { (A B K ) x (t ) + B r (t ) } (33)

where K is the controller gain and 1 (t ) is the scheduling


variable. With r(t) as the reference, the controllers output
can be written as,

( (t )){r(t ) y (t ) K x(t )}

h ( (t )) h ( (t )) B r (t )
G ij = (A i B i K j )

For a reference input r and in the continuous-time case, the


T-S fuzzy control rules have the form of :

u(t ) =

It is assumed throughout this section that; the weight of each


rule in the fuzzy controller is equal to that of the
corresponding rule in the fuzzy model. This assumption is
easy to satisfy since all weighting factors of the controller
can be simply taken over from the known fuzzy model.
Then, Equ. (30) can be rewritten as :

i =1
r

C w (.) = hi ( (t )) C i
i =1

These equations are expressed as,

130

(37)
(38)

A w (.) = hi ( (t ))A iw

(39)

i =1
r

B w (.) = hi ( (t )) B iw

(40)

i =1
r

C w (.) = hi ( (t ))C iw

(41)

parameters are required, the figure will be complicated to


visualize. From the figure, the parametric box represent the
changes in system parameters of two dimensional, whereas
the shown membership function show the extremes in
parameters variation which have been obtained from the
modeling procedure.

i =1

The fuzzy controller depends of two parts: First (u1) and


is based on the state variables of the controlled system
whereas (u2) is based on the difference between the
reference input r(t) and system output y(t) :
r

u 1 (t ) = hi ( (t ))K i x(t )

(42)

i =1

u 2 (t ) = r (t ) hi ( (t ))C i x(t )

(43)

i =1

The entire gain-scheduled control law is then determined


by,
(44)
u = u 2 ( x ) u1 ( x )
r

u(t ) = r(t ) hi ( (t ))C i x(t ) hi ( (t ))K i x(t )


i =1

i =1
r

u(t ) = r (t ) hi ( (t ))[C i + K i ] x(t )

VIII. A CASE STUDY:


A NONLINEAR ANTENNA SIMULATION
This section presents an application of the Fuzzy H
Gain-Scheduling to the multi-input multi-output antenna
dynamic system. It is considered as nonlinear systems,
where the system parameters are assumed to be measured in
real-time. Based on linearization of the dynamic system
around specific operating conditions, LPV representation
are developed. A gain-scheduled H control presented in
section (VII) is immediately applicable. With reference to
the T-S fuzzy model, the associated fuzzy linearized
dynamics of the antenna system variables are re-described
by replacing varying parameter by variables in a state-space
representation in the form of :
&

T b & & & (I I)sin (2 )

I sin 2 ( ) + I cos 2 ( )

z& =

&
2
T b & 1 (I I)

&
sin
(
2

(45)

i =1

u(t ) = r(t ) K w (.) x(t )

K w (.) = hi ( (t )) [C i + K i ]

(46)

i =1

as the state feedback-gain matrix. The closed-loop behavior


of the fuzzy controller is then written as,
(47)

x& (t ) = A w (.) x(t ) + B w (.)r (t )

(50)

1 0 0 0 e (t )
T
y=
z + e (t ) z = [ & & ]
0
0
1
0

(51)

In Equ. (47) A w (t ) is a time-varying system matrix,


satisfying the LPV in Equ. (7),

A w (t ) Co A ijw : i = 1, 2 ,..., L ,

(48)

for the entire time domain the matrices A


follows,

A ijw = A iw B iw C iw + K iw

ij
w

are formed as
Isotropic

(49)

Non-isotropic

Equ. (49) gives the transformations (A iw , B iw , K iw ) for the


synthesis of the controller, denoted as Fuzzy Scheduler, into
the standard problem (A i , B i , K j ) . The gain-scheduler

0.3375

0.3375

0.2 Nms2

0.3375

0.3375

0.5

0.5

0
-20
1

fuzzy control was shown in Fig. 3., with reference input r(t)
applied to the system. Once the system parameters start to
change, the fuzzy controller rather select K from a large
range computed to satisfy the H performance. At the
same time, the H controller K is computed to satisfy the
extremals of system parameter variations as expressed by
Equ. (21). In addition, Fig. 3. shows the fuzzy scheduling
variable (t ) that makes the controller switches from one
controller gain Ki to another.
The figure shows the
variation of two parameters, Mi and Mj, and when more

-10

10

0.5

0.2 Nms

0.2 Nms2
0.02 Nms2

-10

10

-10

10

0.5

0
-20
1

-10

10

0.5
0
-4
1

0
-20
1

I'

0
-20
1
0.5

-2

0
-4

-2

0.5
0
-5

10

Figure 2. Extracted membership functions of all inputs regressors


associated with the azimuth angle.

131

Varying parameters 1 , 2 and 1 are all functions of the


antenna system motion, therefore available to measurement
in real time. Fig. 3. displays the associated variations in the
linear system parameters. The antenna system was even
found to be very close to the marginally stable localities,
depending on the sign of 1 . Parametric changes (i.e.
1 , 2 and 1 ), shown in Fig. 4, are found to be driving
the system performance to be sever.

These points are regularly distributed over the entire region


of operation. For the system formulation of the nonlinear
antenna introduced above and the interconnection shown in
Fig. 3., we are in a position to perform the H gainscheduled controller synthesis. The optimization algorithm
converged and produced a polytopic LPV controller with a
minimum , with structure of :
K ( s , (t )) = Co {K

-3

alpha 10

(s , (t )),

i = 1, 2 , 3 , 4 }

x 10

1
0
-1

100

200

300

400

500

600

100

200

300

400

500

600

100

200

300
Time (s)

400

500

600

beta 1

0.014

0.012

0.01

beta 2

0.03

0.02

0.01

Figure 5. Step response of the antenna system


subject to local simulation of 18 points.

Figure 3. Linearized antenna Parameter ( , , ).


10
1
2

Figure 6. Step response of the antenna system


subject to local simulation of 18 points.

Figure 4.

Predefined parameter trajectory for


the antenna system.

There are abrupt changes in parameters as functions of the


antenna system motion that range over a large operating
domain. Stability properties of the system are greatly
influential. For further validations of the final LPV
controllers, (18) operating points have been selected in the
parameter range.

Fig. 4. shows the antenna parameter variations of the


linearized antenna system parameters 10 , 1, 2 .
Superimposing the time responses corresponding to 18
different testing points is first considered. Using the
concept of polytopic, the dynamics of the antenna system
is modeled in terms of four corners.
For each of these
points, one determines its polytopic coordinate, then the
corresponding system model and controller. Furthermore,
in order to perform such real simulations, one parameter
trajectory has been examined. Such trajectory varies
continuously in the operating domain and converges to a
predefined value within one second time. Fig. 4. shows
such pre-defined parameters trajectory used in this analysis.
132

The actual response due to such parameter trajectory has to


be computed. The system was able to converge to the
required set-points with an achieved stability even under
changes and uncertain in the parameters. This confirms the
H fuzzy gain-scheduling. In this simulation, changes in
model parameters were computed by the pre-defined
trajectory, and the H controller was switching from one
set of controller gain K1 to another set K2 depending on the
model parameters.

IX. CONCLUSIONS
In the gain-scheduling H synthesis, the nonlinear systems
under concern were treated as uncertain system with
parametric range of variations (i.e. region of variation).
Hence, scheduled H controller has been synthesized for
such parameter varying systems. Knowledge of real time
parameter changes is obtained via the employed fuzzy
modeling technique. From the designed H gain-scheduling
for both dynamic systems under concern, it was found that
the scheduled H controller where able to adjust controller
gain parameters to follow the changes in the performance
requirements
for
scheduled
characteristics.
H

Synthesizing H fuzzy gain-scheduling is indeed a new


paradigm in the area of fuzzy control. Considering the
global model of a nonlinear fuzzy system to schedule the
H controller is one approach, however other approaches
are to be investigated.

REFERENCES

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]

[17]

[18]

[19]
[20]

[1]

[2]

[3]

[4]

[5]

[6]

[7]

Tanaka K. and Sugeno M., Stability Analysis and Design of Fuzzy


Control Systems, Fuzzy Sets & systems, vol. 16, No.4, pp.277-240,
1992.
Joh J., Chen Y. and Langari R., On the Stability Issues of Linear
Takagi-Sugeno Fuzzy Models, IEEE Transactions on Fuzzy
Systems, vol.6, No.3, pp.402-410, August 1998.
Tanaka K., Ikeda T. and Wang H., Robust Stabilization of a Class of
Uncertain Nonlinear
Systems via Fuzzy Control: Quadratic

Stabilizability, H Control Theory, and Linear Matrix Inequalities,


IEEE Transactions on Fuzzy Systems, vol.4, No.1, pp.1-13, February
1998.
Chen B., Tseng C. and Uang H., Robustness Design of Nonlinear
Dynamic Systems via Fuzzy Linear Control, IEEE Transactions on
Fuzzy Systems, vol. 7, No.5, pp.571-585, October 1999.
Tanaka K. and Sugeno M., Stability Analysis and Design of Fuzzy
Control Systems, Fuzzy Sets and Systems, Vol. 45, No. 2, pp. 135156, 1992.
Lo J. and Chen Y., Stability Issues on Takagi-Sugeno Fuzzy Model,
Parametric Approach, IEEE Transaction of Fuzzy Systems, vol. 7,
No. 5, pp. 597-607, October 1999.
Tanaka K. and Sano M., Trajectory Stabilization of a Model Car
via Fuzzy Control , Fuzzy Sets and Systems, Vol. 70, pp. 155-170,
1995.

[21]
[22]
[23]

[24]

[25]

133

Sugeno M., On the Stability of Fuzzy Systems Expresses by Fuzzy


Rules with Singleton Consequents, IEEE Transaction of Fuzzy
Systems, vol. 7, No. 2, pp. 201-224, April 1999.
Niemann D., Li J., Wang H. and Tanaka K., Parallel Distributed
Compensation for Takagi-Sugeno Fuzzy Models: New Stability
Conditions and Dynamic Feedback Designs, IFAC 1999 (Beijing),
pp.207-212, July 1999.
Wang H., Tanaka K. and Griffin M.,
An Approach to Fuzzy
Control of Nonlinear Systems: Stability and Design Issues, IEEE
Transaction of Fuzzy Systems, vol. 4, pp. 14-23, February 1996.
Li J., Niemann D., Wang H., Tanaka K., Parallel Distributed
Compensation for Takagi-Sugeno Fuzzy Models: Multiobjective
Controller Design, Proc. American Control Conference, (San
Diego), pp.1832-1836, 1999.
Li J., Niemann D., Wang H., Tanaka K., Multiobjective Dynamic
Feedback Control of Takagi-Sugeno Model via LMIs, Joint Conf. Of
Information Science, (North Carolina), pp 159-162, 1998.
Li J., Niemann D., Wang H. and Tanaka K., Dynamic Parallel
Distributed Compensation for Takagi-Sugeno Fuzzy Systems: An
LMI
approach, Proce. American Control Conf. (San Diego),
pp.1832-1836, 1999.
Wang H., Niemann D., and Tanaka K., T-S Fuzzy Model with
Linear Rule Consequence and PDC Controller: A Universal
Framework for Nonlinear Control Systems, Proc. 9th IEEE Int.
Conf. on Fuzzy Systems, 2000.

Isidori A. and Asolfi A., Disturbance Attenuation and H Control


via Measurement Feedback in Nonlinear Systems, IEEE Trans.
Automatic control, vol. 37, pp.1283-1293, September 1992.

Chen B., H Tracking Design of Uncertain Nonlinear SISO


Systems: Adaptive Fuzzy Approach, IEEE Transactions on Fuzzy
Systems, vol. 4, No.1, pp.32-43, February1996.
Korba P., Werner H., and Frank P. LMI-Based Fuzzy GainScheduling for the TORA Nonlinear Benchmark Control Problem,
In J. Maciejovski, editor, Control 2000, Cambridge, UK, 4-7
September 2000.
Li J., Wang H. and Bushnell L., On the Relationship Between LMIs
and AREs: Applications to Absolute Stability Criteria, Robustness
Analysis and Optimal Control, Submitted to Proc. 39th CDC,
Sydney, Australia, 2000.
Doyle J., Lecture Notes on Advances in Multivariable Control,
ONR/Honeywell Workshop, Minneapolis, October, 1984.
Zhou K., Doyle J. and Glover K., Robust and Optimal Control,
Prentice Hall, 1996.
Zhou K. and Doyle J., Essentials of Robust Control, Prentice Hall,
1998.
Gahinet P., Nemirovski A., Laub A. and Chilali M., Linear Matrix
Inequalities Toolbox, The Math works, 1995.

Chilali M. and Gahint P., H Design with Pole Placement


Constraints: an LMI Approach, Proc. Conf. Dec. Contr., pp. 553558, 1994.
Korba P., Babuska R., Vebruggen H. and Frank P., Fuzzy Gain
Scheduling: Controller and Observer Design by Means Of the
Lyapunov Method and Convex optimization Techniques, to be
appear in IEEE Trans. On Fuzzy Systems.

Apkarian P. and Biannic J., Gain-Scheduled H Control of a


Missile via Linear Matrix Inequalities, Journal of Guidance Control
and Dynamic vol. 18, No. 3, pp.532-538.

Potrebbero piacerti anche