See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/256660821

Design of distributed H∞ fuzzy controllers with constraint for nonlinear

hyperbolic PDE systems

Article in Automatica · October 2012

DOI: 10.1016/j.automatica.2012.06.043

CITATIONS READS

116 242

3 authors, including:

Huai-Ning Wu Jun-Wei Wang

Beihang University (BUAA) University of Science and Technology Beijing
279 PUBLICATIONS 8,787 CITATIONS 109 PUBLICATIONS 2,302 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Jun-Wei Wang on 26 February 2018.

The user has requested enhancement of the downloaded file.

This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright
 Author's personal copy

Automatica 48 (2012) 2535–2543

Contents lists available at SciVerse ScienceDirect

Automatica
journal homepage: www.elsevier.com/locate/automatica

Brief paper

Design of distributed H∞ fuzzy controllers with constraint for nonlinear

hyperbolic PDE systems✩
Huai-Ning Wu a,1 , Jun-Wei Wang a , Han-Xiong Li b,c
a
Science and Technology on Aircraft Control Laboratory, School of Automation Science and Electrical Engineering, Beihang University (Beijing University of Aeronautics and
Astronautics), Beijing 100191, China
b
Department of Systems Engineering and Engineering Management, City University of Hong Kong, Tat Chee Avenue, Kowloon, Hong Kong Special Administrative Region
c
The State Key Laboratory of High Performance Complex Manufacturing, Central South University, Changsha, Hunan, China

article info abstract

Article history: This paper investigates the problem of designing a distributed H∞ fuzzy controller with constraint for
Received 11 September 2011 a class of nonlinear spatially distributed processes modeled by first-order hyperbolic partial differential
Received in revised form equations (PDEs). The purpose of this paper is to design a distributed fuzzy state feedback controller such
9 March 2012
that the closed-loop PDE system is exponentially stable with a prescribed H∞ performance of disturbance
Accepted 12 April 2012
Available online 18 July 2012
attenuation, while the control constraint is respected. Initially, a Takagi–Sugeno (T–S) hyperbolic PDE
model is proposed to accurately represent the nonlinear PDE system. Then, based on the T–S fuzzy PDE
Keywords:
model, a distributed H∞ fuzzy controller design with constraint is developed in terms of a set of coupled
H∞ control differential/algebraic linear matrix inequalities (D/ALMIs) in space. Furthermore, a suboptimal distributed
Takagi–Sugeno (T–S) models H∞ fuzzy controller with constraint is proposed to minimize the level of attenuation. The finite difference
Spatially distributed processes method in space and the existing linear matrix inequality (LMI) optimization techniques are employed to
Control constraint approximately solve the suboptimal fuzzy control design problem. Finally, the proposed design method is
Exponential stability applied to the distributed control of a nonlinear system described by two coupled first-order hyperbolic
Linear matrix inequalities (LMIs) PDEs to illustrate its effectiveness.
© 2012 Elsevier Ltd. All rights reserved.

1. Introduction most industrial processes are inherently distributed in space so

that their behavior depends on spatial position as well as time,
Since it can combine the merits of both fuzzy logic theory such as heat conduction, fluid flow, and chemical reactor processes
and linear system theory, the so-called Takagi–Sugeno (T–S) fuzzy (Christofides, 2001; Keulen, 1993; Li & Qi, 2010; Ray, 1981).
model (Takagi & Sugeno, 1985) has been widely employed to deal The mathematical models of these spatially distributed processes
with the control design problem of nonlinear systems over the usually take the form of partial differential equations (PDEs). Due
past decade (Feng, 2010; Gao & Chen, 2007; Guerra & Vermeiren, to the infinite-dimensional nature of PDE systems, the existing T–S
2004; Lee, Park, & Joo, 2011; Mozelli, Palhares, Souza, & Mendes, fuzzy-model-based control methods for nonlinear ODE systems
2009; Tanaka, Hori, & Wang, 2003; Tanaka & Wang, 2001; Tuan, are very difficult to be directly applied to the control design of
Apkarian, Narikiyo, & Yamamoto, 2001; Wu, 2008). However, nonlinear PDE systems.
the existing results are mainly developed for nonlinear systems The well-known classification of PDE systems could be hyper-
modeled by ordinary differential equations (ODEs). In practice, bolic, parabolic, and elliptic, etc. (Ray, 1981), according to the prop-
erties of the spatial differential operator (SDO). The main feature
of parabolic PDE systems is that the eigenspectrum of the SDO can
✩ This work was supported in part by the National Basic Research Program of be partitioned into a finite-dimensional slow one and an infinite-
China (973 Program) (2012CB720003); in part by the National Natural Science
dimensional stable fast complement, which implies that their
Foundation of China under Grant 61074057, Grant 61121003, and Grant 51175519; dominant dynamic behavior can be approximately described by
in part by the Innovation Foundation of BUAA for Ph.D. Graduates; and in part finite-dimensional ODE systems. Therefore, the standard approach
by the General Research Fund project from Research Grants Council of Hong to the control of parabolic PDE systems involves the application of
Kong, SAR, under CityU: 117310. The material in this paper was not presented at
Galerkin’s method to the PDE system to derive an approximate ODE
any conference. This paper was recommended for publication in revised form by
Associate Editor Akira Kojima under the direction of Editor Ian R. Petersen. system, which is subsequently used as the basis for controller syn-
E-mail addresses: [email protected] (H.-N. Wu), [email protected] thesis (Balas, 1979; Curtain, 1982; Ray, 1981). Recently, the above
(J.-W. Wang), [email protected] (H.-X. Li). T–S fuzzy-model-based control technique together with Galerkin’s
1 Tel.: +86 10 82317301; fax: +86 10 82317332. method has been successfully applied to the finite-dimensional
0005-1098/$ – see front matter © 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.automatica.2012.06.043
 Author's personal copy

2536 H.-N. Wu et al. / Automatica 48 (2012) 2535–2543

fuzzy control design of nonlinear parabolic PDE systems (Chen & Notations. ℜ, ℜn and ℜm×n denote the set of all real numbers,
Chang, 2009; Wu & Li, 2007, 2008). A mode-constrained fuzzy con- n-dimensional Euclidean space and the set of all m × n matrices,
trol design with guaranteed quadratic performance and an H∞ respectively. ∥ · ∥ and ⟨·, ·⟩ℜn denote the Euclidean norm and inner
fuzzy observer-based control design with actuator saturation were product for vectors, respectively. An identity matrix, of appropriate
proposed for a class of semi-linear parabolic PDE systems in Wu dimensions, will be denoted by I . For a symmetric matrix M ,
and Li (2007, 2008), respectively, where the singular perturbation M > (≥, <, ≤)0 means that it is positive definite (positive semi-
theory (Khalil, 2002) is used to ensure the stability of the closed- definite, negative definite, negative semi-definite, respectively).
loop PDE system. Chen and Chang (2009) then presented a robust The space-varying symmetric matrix function M (x) > (≥, <, ≤)
H∞ fuzzy observer-based control design for a class of quasilinear 0, x ∈ [l1 , l2 ] means that it is positive definite (positive semi-
parabolic PDE systems, in which the standard Galerkin method is definite, negative definite, negative semi-definite, respectively) for
used to obtain a fuzzy infinite-dimensional ODE model consisting each x ∈ [l1 , l2 ]. H n , L2 ([l1 , l2 ]; ℜn ) is a Hilbert space of n-
of a finite-dimensional slow subsystem to be controlled and a cou- dimensional square integrable vector function ϑ(x, t ) ∈ ℜn , x ∈
pled infinite-dimensional fast subsystem to be tolerated, and the [l1 , l2 ] ⊂ ℜ, ∀t ≥ 0 with the inner product and norm:
stability of the closed-loop system is guaranteed by the small gain  l2
theorem. ⟨ϑ1 (·, t ), ϑ2 (·, t )⟩ = ⟨ϑ1 (x, t ), ϑ2 (x, t )⟩ℜn dx and
The distinct feature of hyperbolic PDE systems is that all l1

the eigenmodes of the SDO contain the same, or nearly the ∥ϑ1 (·, t )∥2 = ⟨ϑ1 (·, t ), ϑ1 (·, t )⟩1/2
same, amount of energy, and thus an infinite number of modes
where ϑ 1 (·, t ), ϑ2(·, t )  ∈ H n . For a space-varying matrix
are required to accurately describe their dynamic behavior.
function S (x) = sij (x) m×n , x ∈ [l1 , l2 ], we define ∥S (x)∥2F ,
This feature prohibits the application of modal decomposition m n  l2
techniques to the hyperbolic PDE system to derive an approximate i=1 j =1 s2ij (x)dx. The superscript ‘T ’ is used for the transpose
l1
ODE model and suggests addressing the control problem on of a vector or a matrix. The symbol ‘∗’ is used as an ellipsis in matrix
the basis of the infinite-dimensional model itself. Following expressions that are induced by symmetry.
this suggestion, some control approaches have been developed
for hyperbolic PDE systems in the past decades, including the 2. Problem formulation and preliminaries
linear quadratic optimal control method via spectral factorization
(Aksikas, Winkin, & Dochain, 2007), the sliding mode control We consider nonlinear spatially distributed processes given by:
method (Sira-Ramirez, 1989) and the model predictive control
yt (x, t ) = Θ(x)yx (x, t ) + f (y (x, t ), x)

method (Shang, Frobes, & Guay, 2004) on the basis of equivalent
ODE realizations obtained by the method of characteristics, and
+Gu (y (x, t ), x)u(x, t ) + Gw (y (x, t ), x)w (x, t ) (1)
z (x, t ) = h(y (x, t ), x) + D(y (x, t ), x)u(x, t )
the nonlinear control method (Christofides, 2001; Christofides
& Daoutidis, 1996) through a combination of PDE theory and subject to the boundary condition
geometric control techniques. More recently, Wang, Wu, and H1 y (l1 , t ) + H2 y (l2 , t ) = d (2)
Li (2011) have developed a distributed fuzzy control design
to exponentially stabilize a class of nonlinear hyperbolic PDE and the initial condition
systems based directly on the T–S fuzzy PDE model. However, y (x, 0) = y0 (x) (3)
for control engineers, it is usually desirable to design a control
where yt (x, t ) = ∂ y (x, t )/∂ t, yx (x, t ) = ∂ y (x, t )/∂ x, y (x, t ) ∈
system which is not only stable but also attenuates the effect
ℜn is the state, x ∈ [l1 , l2 ] ⊂ ℜ and t ∈ [0, ∞) denote
of external disturbances impinging on the processes. To the best
position and time, respectively, u(x, t ) ∈ ℜm is the control
of our knowledge, the distributed H∞ fuzzy control problem of
 ∞ w (x, t )2 ∈ ℜ is the exogenous
p
input, disturbance satisfying
nonlinear hyperbolic PDE systems has not been addressed yet,
∥w (·, t )∥2 dt < ∞. z (x, t ) ∈ ℜq is the controlled output.
which motivates the present study. 0
Θ(x) = diag{θ1 (x), . . . , θn (x)} is a real known diagonal n ×
In this study, we will deal with the problem of designing a
n matrix function of x. f (y (x, t ), x), h(y (x, t ), x), Gu (y (x, t ), x),
distributed H∞ fuzzy controller with constraint for a class of
Gw (y (x, t ), x), and D(y (x, t ), x) are locally Lipschitz continuous
nonlinear hyperbolic PDE systems, where the control actuators
vector or matrix functions in y (x, t ) with appropriate dimensions,
are continuously distributed in space. Initially, a T–S hyperbolic and f (y (x, t ), x) satisfies f (0, x) = 0 for all x ∈ [l1 , l2 ]. H1 and
PDE model is proposed to accurately represent the nonlinear H2 are real known n × n matrices. d ∈ ℜn is a known constant
PDE system. Then, based on the T–S fuzzy PDE model and the vector, and y0 (x) ∈ ℜn is a known initial state. Whenever it is not
parallel distributed compensation (PDC) scheme (Tanaka & Wang, confusing, we drop the arguments x and t of y (x, t ), w (x, t ), etc.
2001), a distributed H∞ fuzzy controller design is developed such Choose H n as the state space and the trajectory segment
that the closed-loop PDE system is exponentially stable with a y (·, t ) = {y (x, t ), x ∈ [l1 , l2 ]} as the state. Define the operator
prescribed H∞ performance of disturbance attenuation, while the A in H n as A , Θ(x)∂/∂ x with its domain
control constraint is respected. The outcome of the distributed
D (A) = y ∈ H n : y is absolutely continuous,

H∞ fuzzy control problem with constraint is presented in terms
∂ y /∂ x ∈ H n and H1 y (l1 ) + H2 y (l2 ) = d .

of a set of coupled differential/algebraic linear matrix inequalities
(D/ALMIs) in space. Furthermore, a suboptimal distributed H∞
Let H m be control input space and the trajectory segment u(·, t ) =
fuzzy controller with constraint is proposed to minimize the level {u(x, t ), x ∈ [l1 , l2 ]} as the control input. Similarly, H p and
of attenuation. The suboptimal fuzzy control problem can be w (·, t ) = {w (x, t ), x ∈ [l1 , l2 ]} stand for the exogenous distur-
solved approximately by using the finite difference method in bance input space and the exogenous disturbance input, respec-
space and the existing linear matrix inequality (LMI) optimization tively. Then the state equation of the model (1) can be rewritten as
techniques (Boyd, Ghaoui EI, Feron, & Balakrishnan, 1994; Gahinet, the following nonlinear abstract differential equation on H n (Pazy,
Nemirovskii, Laub, & Chilali, 1995). Finally, a simulation study on 1983):
the distributed control of a nonlinear system described by two
coupled hyperbolic PDEs is given to show the effectiveness of the ẏ (t ) = Ay (t ) + f˜ (y (t )) + G̃u (y (t ))u(t ) + G̃w (y (t ))w (t ),
proposed design method. t ≥0 y (0) = y0 (4)
 Author's personal copy

H.-N. Wu et al. / Automatica 48 (2012) 2535–2543 2537

where f˜ (y (t )) , f (y (·, t ), ·), G̃u (y (t )) , Gu (y (·, t ), ·), G̃w (y (t )) , When controlling a real process, it is usually desirable to design
Gw (y (·, t ), ·), y0 , y0 (·) ∈ D (A), and the operator A is assumed a control system which is not only stable but also guarantees an
to generate a C0 -semigroup on H n . adequate level of performance. In order to attenuate the effect of
Since the operator A generates a C0 -semigroup on H n , by w (x, t ), H∞ control will be employed to deal with the disturbance
applying Theorem 1.7 of Chapter 6 in Pazy (1983) and considering attenuation problem in this study. We consider the following H∞
the term f˜ (y (t )) is locally Lipschitz continuous in y (t ), we can control performance for the fuzzy system (6) and (7):
easily prove the local existence of the unique classical solution to  ∞
the system (4) with u(t ) = 0 and w (t ) = 0. For more details ∥z (·, t )∥22 dt ≤ ⟨y0 (·), P (·)y0 (·)⟩
regarding this issue, one can refer to Pazy (1983). 0
In this study, we assume that the nonlinear PDE system (1) can  ∞
be exactly represented by the following T–S fuzzy PDE model: +γ2 ∥w (·, t )∥22 dt (11)
Plant rule i: 0

IF ξ1 (x, t ) is Fi1 and · · · and ξl (x, t ) is Fil where P (x) > 0 is a real continuous space-varying n × n
matrix defined on [l1 , l2 ] and γ > 0 is a prescribed level
yt = Ay + Ai (x)y + Gu,i (x)u + Gw,i (x)w

of disturbance attenuation. In order to attain a lower level of
THEN z = Ci (x)y + Di (x)u, (5)
disturbance attenuation, the control input to the process is often
i ∈ S , {1, 2, . . . , r }
required to have a very high magnitude. This, of course, is seldom
where Fij , i ∈ S, j = 1, 2, . . . , l are fuzzy sets, Ai (x), Gu,i (x), possible in practice, and, in a more realistic setting, the control
Gw,i (x), Ci (x), and Di (x), for i ∈ S are real continuous space- input must be constrained in magnitude. This consideration leads
varying matrices with appropriate dimensions, and r is the number us to impose the following magnitude constraint on control input
of IF–THEN rules. ξ1 (x, t ), . . . , ξl (x, t ) are the premise variables, in this study:
which are assumed to be functions of state variables and/or space
in this study. It has been shown that an exact T–S fuzzy ODE model ∥u(·, t )∥22 ≤ umax (12)
construction from a given nonlinear dynamical ODE model can
be obtained by the sector nonlinearity approach (Tanaka & Wang, where umax is a given positive scalar.
2001). For simplicity, when u(x, t ) ≡ 0, the fuzzy PDE system (6) is
By applying the center-average defuzzifier, product interfer- referred to as an unforced fuzzy PDE system, while when w (x, t ) ≡
ence and singleton fuzzier, the overall dynamics of T–S fuzzy PDE 0, it is referred to as a disturbance-free fuzzy PDE system. We
system (5) can be expressed as introduce the following definitions:
r

yt = Ay + hi (ξ){Ai (x)y + Gu,i (x)u + Gw,i (x)w } Definition 1. Given a constant ρ > 0. The unforced disturbance-
i =1 free fuzzy PDE system of (6) (i.e., u(x, t ) ≡ 0 and w (x, t ) ≡ 0) is
= Ay + A(ξ, x)y + Gu (ξ, x)u + Gw (ξ, x)w (6) said to be ρ -exponentially stable (ES) or ES with decay rate ρ , if there
r exists a constant σ > 0 such that the following expression holds:
z = hi (ξ){Ci (x)y + Di (x)u} = C (ξ, x)y + D(ξ, x)u (7)
i =1 ∥y (·, t )∥22 ≤ σ exp(−2ρ t ) ∥y0 (·)∥22 , ∀t ≥ 0. (13)

where ξ = [ξ1 · · · ξl ]T and M (ξ, x) = hi (ξ)Mi (x) with

r
i=1
M ∈ {A, Gu , Gw , C , D} Definition 2. Given constants ρ > 0 and γ > 0. The unforced
l
 r
 fuzzy PDE system of (6) and (7) is said to be ρ -ES with γ -
ωi (ξ) = Fij (ξj ), hi (ξ) = ωi (ξ) ωi (ξ). disturbance attenuation if the response z (x, t ) satisfies (11) and the
j =1 i=1 disturbance-free system is ρ -ES.
Fij (ξj ) is the grade of the membership of ξj in Fij . This paper assumes Substituting (10) into (6) and (7) leads to the following closed-
that ωi (ξ) ≥ 0, i ∈ S and i=1 ωi (ξ) > 0 for all x ∈ [l1 , l2 ] and
r loop system:
t ≥ 0. Then we have
yt = Ay + Ac (ξ, x)y + Gw (ξ, x)w (14)
r
z = Cc (ξ, x)y (15)

hi (ξ) ≥ 0, i ∈ S and hi (ξ) = 1 (8)
i =1
where Ac (ξ, x) , A(ξ, x)+ Gu (ξ, x)K (ξ, x) and Cc (ξ, x) , C (ξ, x)+
for all x ∈ [l1 , l2 ] and t ≥ 0. D(ξ, x)K (ξ, x).
Based on the PDC scheme (Tanaka & Wang, 2001), we consider
the following fuzzy control law for the system (6) and (7): Therefore, the problem under consideration is to find a dis-
Control rule j: tributed H∞ fuzzy controller of the form (10) for the fuzzy PDE sys-
tem (6) and (7) such that the closed-loop system (14) and (15) is
IF ξ1 is Fj1 and · · · and ξl is Fjl
ρ -ES with γ -disturbance attenuation, while the control constraint
THEN u = Kj (x)y , j∈S (9) in (12) is respected.
where Kj (x), x ∈ [l1 , l2 ], j ∈ S are real continuous space-varying
m × n matrices to be determined. 3. Design of distributed H∞ fuzzy controllers with constraint
The overall fuzzy controller is represented by
r
 Let us choose the Lyapunov function candidate as
u= hj (ξ)Kj (x)y = K (ξ, x)y (10)
 l2
j =1
V (t ) = y T (x, t )P (x)y (x, t )dx (16)
where K (ξ, x) = j=1 hj (ξ)Kj (x).
r
l1
 Author's personal copy

2538 H.-N. Wu et al. / Automatica 48 (2012) 2535–2543

where P (x) > 0 is the same as that in (11). Differentiating V (t ) Therefore, we have the following theorem:
in (16) along the solution of the closed-loop system (14) and (15),
yields Theorem 1. Consider the fuzzy system (6) and (7). For some given
 l2 scalars ρ > 0 and γ > 0, there exists a fuzzy controller of
dV (t )/dt = [y T P (x)Ay + (Ay )T P (x)y ]dx the form (10) such that the closed-loop system is ρ -ES with γ -
l1 disturbance attenuation, if there exist a diagonal matrix Q (x) ,
 l2 diag {q1 (x), q2 (x), . . . , qn (x)} > 0 and m × n matrices Ni (x), i ∈ S
+ y T [P (x)Ac (ξ, x) + ∗]ydx + Γ1 (17) satisfying the following DLMIs for all x ∈ [l1 , l2 ]:
l1

Ξii (x) < 0, i∈S

l 
where Γ1 , 2 l 2 y T P (x)Gw (ξ, x)wdx. Considering (4) and apply-
1 1 1 (29)
Ξii (x) + Ξij (x) + Ξji (x) < 0, i ̸= j, i, j ∈ S

ing integration by parts, we have
r −1 2
 l2
x =l
y T P (x)Aydx = y T (x, t )P (x)Θ(x)y (x, t ) |x=l21 where
l1
Ξ (x) + Λij (x) + ℘(x)
 
 l2
∗ ∗
−

y T P (x)Θ(x)

ydx. (18) Ξij (x) ,  T
Gw, i ( x) −γ 2 I ∗ (30)
l1
x
Ci (x)Q (x) + Dj (x)Ni (x) 0 −I
We introduce the Dirac delta function δ(x), which is defined as with Ξ (x) , dQ (x)/dxΘ(x), and
follows (Levine, 2009)
 Λij (x) , [Ai (x)Q (x) + Gu,j (x)Ni (x) + ∗],
0 x ̸= 0
δ(x) = (19) ℘(x) , 2ρ Q (x) − Q (x)dΘ(x)/dx
∞ x=0
+ [δ(x − l2 ) − δ(x − l1 )]Q (x)Θ(x).
and
In this case, the gain matrices of a suitable fuzzy controller can be
 x2
constructed as
δ(x)dx = 1, if 0 ∈ [x1 , x2 ] (20)
x1
Ki (x) = Ni (x)Q −1 (x), i ∈ S. (31)
with the following fundamental property:
x2
Proof. Let

f (x)δ(x − x0 )dx = f (x0 ), x0 ∈ [x1 , x2 ], ∀f (x). (21)
x1 P (x) = Q −1 (x), Ni (x) = Ki (x)Q (x), i ∈ S. (32)
Using (21), (18) can be rewritten as
Using Theorem 2 in Tuan et al. (2001), we can conclude that if
l2 l2 DLMIs (29) hold for some diagonal matrix Q (x) > 0 and some
 
y P (x)Aydx = −
T
y P (x)Θ(x) x ydx + Γ2
T
 
(22) matrices Ni (x), i ∈ S, then the following inequality is fulfilled:
l1 l1

Ψ̄ (ξ, x)
 
l
where Γ2 , l 2 [δ(x − l2 ) − δ(x − l1 )]y T P (x)Θ(x)ydx. ∗ ∗
1
In order to obtain an H∞ stabilization condition in terms of
G T (ξ, x)P (x)
w −γ 2 I ∗<0 (33)
DLMIs in space, we restrict ourselves to the case that the matrix Cc (ξ, x) 0 −I
P (x) given in (11) is diagonal, i.e.,
where Ψ̄ (ξ, x) is given by (27). By Schur complement, it follows
P (x) = diag{p1 (x), p2 (x), . . . , pn (x)}. (23) that (33) is equivalent to
In this case, for all x ∈ [l1 , l2 ], the equality P (x)Θ(x) = Θ(x)P (x)
Ψ̄ (ξ, x) + CcT (ξ, x)Cc (ξ, x)
 
∗
holds clearly. Using this equality, we obtain ∆, < 0. (34)
GwT
(ξ, x)P (x) −γ 2 I
 l2  l2
y T P (x)Θ(x) (Ay )T P (x)ydx + Γ3
 
ydx = (24) From (15), (26) and (34), we have
x
l1 l1
 l2 dV (t )/dt + 2ρ V (t ) + ∥z ∥22 − γ 2 ∥w ∥22
where Γ3 , l y d(P (x)Θ(x))/dx ydx. Substituting (24) into and
T
 l2
1 T
yT wT ∆ yT wT dx ≤ 0.
  
rearranging (22), we get = (35)
l1
 l2
[y T P (x)Ay + (Ay )T P (x)y ]dx = Γ2 − Γ3 . (25) Integrating (35) from t = 0 to t = ∞ yields
l1  ∞
Using (17) and (25), we can obtain 2ρ V (t )dt + V (∞) − V (0)
0
 l2  ∞
dV (t )/dt + 2ρ V (t ) = y T Ψ̄ (ξ, x)ydx + Γ1 (26) + ∥z ∥2 − γ 2 ∥w ∥22 dt ≤ 0.
 2 
(36)
l1
0
where
Since V (t ) ≥ 0, t ≥ 0, we obtain (11) from (36).
Ψ̄ (ξ, x) , Φ (x) + 2ρ P (x) + [P (x)Ac (ξ, x) + ∗] (27) Now, we prove that the disturbance-free system of (14) is ρ -ES.
When w (x, t ) = 0, (26) can be simplified as
in which
d(P (x)Θ(x))
 l2
Φ (x) , − + [δ(x − l2 ) − δ(x − l1 )]P (x)Θ(x). (28) dV (t )/dt + 2ρ V (t ) = y T Ψ̄ (ξ, x)ydx. (37)
dx l1
 Author's personal copy

H.-N. Wu et al. / Automatica 48 (2012) 2535–2543 2539

It is clear from (34) that Ψ̄ (ξ, x) < 0 for all x ∈ [l1 , l2 ], which Therefore, based on Theorem 1, we have the following result,
implies which shows how the control constraint (12) can be enforced at
all time t > 0 for the distributed H∞ fuzzy controller (10).
dV (t )/dt + 2ρ V (t ) < 0 for 0 ̸= y (·, t ) ∈ H n . (38)
Theorem 2. Assume that there exist a diagonal matrix Q (x) > 0
Hence, by using Lemma 1 in Wang et al. (2011), we can easily ob-
and matrices Ni (x), i ∈ S such that DLMIs (29) are satisfied for all
tain from (38) that the disturbance-free system of (14) and (15)
x ∈ [l1 , l2 ]. The control constraint (12) is enforced at all time t > 0
is ρ -ES on the Hilbert space H n . Moreover, from (32), we have
for the distributed H∞ fuzzy controller (10), if the following algebraic
(31). 
LMIs (ALMIs) hold for all x ∈ [l1 , l2 ]:
Remark 1. It is obvious that the model (1) involves a special case
(l2 − l1 )−1 (ϕ − wmax )
 
∗
Gu (y (x, t ), x) = Gu (x) and D(y (x, t ), x) = D(x). In such a case, ≥0 (43)
y0 (x) Q (x)
by letting Gu,j (x) = Gu (x), Dj (x) = D(x), j ∈ S, DLMIs (29) in
Q (x)
 
Theorem 1 can be simplified into the following DLMIs: ∗
≥ 0, i ∈ S. (44)
Ni (x) ϕ −1 umax I
Ξ (x) + Λi (x) + ℘(x)
 
∗ ∗
 T
Gw, i (x) −γ 2 I ∗  < 0, i∈S (39)
Proof. By Schur complement and considering (32), it follows that
Ci (x)Q (x) + D(x)Ni (x) 0 −I (43) and (44) are respectively equivalent to
where Λi (x) , [Ai (x)Q (x) + Gu (x)Ni (x) + ∗].
(ϕ − wmax )(l2 − l1 )−1 − y0T (x)P (x)y0 (x) ≥ 0 (45)
Remark 2. Notice that DLMIs (29) are established by introducing and
the Dirac delta function defined in (19) and (20). The expression
(29) is similar to the continuous (in time and space) differential ϕ u− 1
max P
−1
(x)KiT (x)Ki (x)P −1 (x) − P −1 (x) ≤ 0, i ∈ S. (46)
Riccati equation involving a Dirac delta function for the optimal Obviously, integrating (45) from x = l1 to x = l2 yields (42),
point-wise control of the hyperbolic and parabolic PDE systems which can ensure (41). Pre- and post-multiplying both sides of (46)
(Ray, 1981). Strictly speaking, however, the Dirac delta is not 1 1

a true function in the analysis sense, because any real function by ϕ − 2 umax
2
P (x), respectively, we can get
that is equal to zero everywhere but a single point must have
KiT (x)Ki (x) ≤ ϕ −1 umax P (x), i ∈ S. (47)
total integral zero. While for many purposes the Dirac delta can
(x) − (x) × (Ki (x) −
 
be manipulated as a function, formally it can be defined as a In addition, it follows from the fact KiT KjT
distribution that is also a measure. The Dirac delta function is Kj (x)) ≥ 0 that
regarded as a simplest generalized function, which is extensively,
especially in physics and engineering, used to describe physical KiT (x)Kj (x) + KjT (x)Ki (x) ≤ KiT (x)Ki (x) + KjT (x)Kj (x). (48)
phenomena such as point changes. Due to its property given by
Using (10) and (48), we can obtain that for any t > 0
(21), the Dirac delta function has been widely used to describe
the point-wise control/sensor problem in distributed parameter  l2 r

control theory (e.g., see Demetriou (2010), and Ray (1981), and the ∥u∥22 ≤ hi (ξ)y T KiT (x)Ki (x)ydx. (49)
references therein) and applied to introduce the boundary control l1 i=1
into the differential equation of PDE systems (e.g., see Ray (1981)).
By using (47) and considering (8), it follows from (49) that
In general, the Dirac delta function δ(x − xb ), xb = l1 or l2 can
l2 r
be approximated by the following form (Demetriou, 2010; Levine,
 
2009): ∥u∥ ≤ ϕ umax
2
2
−1
y T P (x)y hi (ξ)dx
l1 i =1
0,

otherwise
δε (x − xb ) ,
 ε ε = ϕ −1 umax V (t ). (50)
ε ,
−1
if x ∈ xb − , xb +
2 2 Therefore, from (41) and (50), we can conclude that the control
where ε is a sufficiently small positive constant. constraint (12) is enforced at all time t > 0. 

From the proof of Theorem 1, we know that if there exist a Remark 3. It must be stressed that the obtained results assume
diagonal matrix Q (x) > 0 and matrices Ni (x), i ∈ S such that DLMIs distributed sensing and actuation. Although this is normally
(29) are satisfied for all x ∈ [l1 , l2 ], then it can be concluded that recognized as a critical drawback, with recent advances in
(35) holds. The integration of (35) from zero to t leads immediately technological developments of micro-electro-mechanical systems,
to the following inequality: it becomes feasible to manufacture large arrays of micro-sensors
 t  t and actuators with integrated control circuitry, which can be used
V (t ) + 2ρ V (τ ) + ∥z ∥ 2
dτ ≤ γ 2
∥w ∥22 dτ + V (0) for the implementation of distributed feedback control loops in
 
2 (40)
0 0 some practical applications (see Bamieh, Paganini, and Dahleh
for all t > 0. ∞
(2002) and the references therein). In addition, the developed
Assume that the disturbance energy is bounded as 0 ∥w ∥22 dτ framework of analysis paves the way for further improvements,
t such as point-wise (or boundary) control and/or sensing that will
≤ γ −2 wmax , where wmax is a positive scalar. Since 0
(2ρ V (τ ) + be addressed in future research activities.
∥z ∥22 )dτ ≥ 0, the inequality (40) implies
To make the level of attenuation γ as small as possible, we can
V (t ) ≤ ϕ for all t > 0 (41) consider the following minimization problem for some given scalar
ρ > 0:
where ϕ > 0 is a scalar, if the following inequality holds:
 l2 min ϖ
ϖ ,Q (x),Ni (x)
wmax + y0T (x)P (x)y0 (x)dx ≤ ϕ. (42) (51)
l1 subject to D/ALMIs (29) (or (39)), (43) and (44)
 Author's personal copy

2540 H.-N. Wu et al. / Automatica 48 (2012) 2535–2543

where ϖ = γ 2 and Q (x) > 0. Clearly, it is very hard to solve 4. Computer simulation
the above optimization problem because of its infinite dimensional
nature. However, by using the finite difference method in space We consider the following nonlinear hyperbolic PDE system
and the existing LMI optimization techniques, the optimization with distributed disturbance and control inputs:
problem (51) can be approximately solved as follows.
ȳ1,t = −υ1 (x)ȳ1,x + β1 (x)ȳ1 + r1 (x)ȳ1 ȳ2 + b(x)u + 0.25w (57)
Firstly, discretize the interval [l1 , l2 ] into space instances
{xk , k ∈ N, x0 = l1 , xnz = l2 } of the same distance, where ȳ2,t = −υ2 (x)ȳ2,x + β2 (x)ȳ2 + r2 (x)ȳ1 ȳ2 + 0.1w (58)
xk − xk−1 , φ = (l2 − l1 )/nz , N , {0, 1, 2, . . . , nz }, and consider z = ȳ1 (59)
the backward difference for the spatial derivative, i.e., dQ (x)/dx ≈
(Q (xk )− Q (xk−1 ))/φ . Then, for some given scalar ρ > 0, the space- subject to the boundary conditions:
varying DLMIs (29) and (39) can be respectively approximated ȳ1 (0, t ) = 2, ȳ2 (0, t ) = 1.1, t ≥0 (60)
by the following discrete-space LMIs with respect to matrices
Q (x−1 ) > 0, Q (xk ) > 0, and Ni (xk ), k ∈ N, i ∈ S: and the initial conditions:
ȳ1 (x, 0) = 2 + 0.1 sin(2π x),
Σii (xk ) < 0,

ȳ2 (x, 0) = 1.1 + 0.2 sin(π x), x ∈ [0, 1],

k ∈ N, i ∈ S (61)



1 1 (52)
Σii (xk ) + Σij (xk ) + Σji (xk ) < 0,

where ȳ1 and ȳ2 are the state variables; u stands for the distributed
control input; w is the exogenous disturbance; υi (x), βi (x), ri (x),

r − 1 2

k ∈ N, i ̸= j, i, j ∈ S

i = 1, 2 and b(x), x ∈ [0, 1] are the process parameters. The
Ξ̃ (xk ) + Λi (xk ) + ℘(xk )
 
∗ ∗ nonlinear term ȳ1 ȳ2 in the system (57)–(61) is of Lotka–Volterra
 Gw,T
i ( xk ) −ϖ I ∗  < 0, type. This system is a widely used model of biological systems
Ci (xk )Q (xk ) + D(xk )Ni (xk ) 0 −I and networks, predator–prey and competing species interaction
(Fister, 2001; Goh, Leitmann, & Vincent, 1974) with ȳ1 and ȳ2
k ∈ N, i ∈ S (53) denoting the predator and prey, respectively.
where Ξ̃ (xk ) , (Q (xk ) − Q (xk−1 ))Θ(xk )/φ and Define the following state transformation:
 T  T
Ξ̃ (xk ) + Λij (xk ) + ℘(xk ) , ȳ1 − ȳ1d
 
∗ ∗ y = y1 y2 ȳ2 − ȳ2d (62)
Σij (xk ) ,  T
Gw, i (xk ) −ϖ I ∗ . where ȳ1d and ȳ2d are the desired constant steady-state of ȳ1 and
Ci (xk )Q (xk ) + Dj (xk )Ni (xk ) 0 −I ȳ2 , respectively. Then the system (57)–(61) can be represented as
The space-varying ALMIs (43) and (44) can be respectively approx- the form of (1)–(3) with
imated by the following discrete-space LMIs with respect to matri- Θ(x) = −diag{υ1 (x), υ2 (x)}, H1 = diag{1, 1},
ces Q (xk ) > 0 and Ni (xk ), k ∈ N, i ∈ S: T
Gu (y , x) = Gu (x) = b(x) 0 , H2 = diag{0, 0},


(z2 − z1 ) (ϕ − wmax )
 −1

∗ β1 (x)y1 + r1 (x)f0 (y , x)
 
≥ 0, k ∈ N (54) f ( y , x) = , h(y , x) = y1 ,
y0 (xk ) Q (xk ) β2 (x)y2 + r2 (x)f0 (y , x)
Q (xk )
  T
∗ Gw (y , x) = Gw (x) = 0.25 0.1 ,

D(y , x) = 0,
≥ 0, k ∈ N, i ∈ S. (55)
Ni (xk ) ϕ umax I
−1
T
d = 2 − ȳ1d 1.1 − ȳ2d , l1 = 0, l2 = 1,

Therefore, for some given scalar ρ > 0, a suboptimal dis- T  T
y0 = y10 y20 , ȳ1 (x, 0) − ȳ1d ȳ2 (x, 0) − ȳ2d

tributed H∞ fuzzy controller design with constraint can be for- (63)
mulated as the following approximate LMI optimization problem
where f0 (y , x) , y1 y2 + ȳ1d y2 + ȳ2d y1 .
under a suitable choice of the positive integer nz :
In this example, the values of model parameters are given as
min ϖ follows:
U
(56)
subject to LMIs (52) (or (53)), (54) and (55) υ1 (x) = 0.1, υ2 (x) = 0.2, β1 (x) = 0.8 sin(2x),
r1 (x) = 1, r2 (x) = −1, β2 (x) = 0.5 cos(2x),
where U , {ϖ , Q (x−1 ) > 0, Q (xk ) > 0, Ni (xk ), k ∈ N, i ∈ S},
and b(x) = 1, x ∈ [0, 1].
which can be solved efficiently using the LMI optimization tech-
niques (Boyd et al., 1994; Gahinet et al., 1995). Using the solution Using these parameters, Fig. 1 shows the open-loop profiles of the
ϖ , Q (xk ), and Ni (xk ), k ∈ N, i ∈ S of the problem (56), the corre- evolution of ȳ1 and ȳ2 in the absence of w ≡ 0. In this situation, we
sponding optimized level of attenuation and the approximate con- find that (11) does not hold, even though the open-loop system is
trol gain matrices Ki (xk ), k ∈ N, i ∈ S can be obtained as stable. Moreover, it is easy to verify that the operator A with the
√ parameters given in (63) can generate a C0 -semigroup on H 2 .
γ = ϖ, Ki (xk ) = Ni (xk )Q −1 (xk ), k ∈ N, i ∈ S. The nonlinear term f (y , x) can be rewritten as

β1 (x) + r1 (x)(ȳ2d + y2 ) r1 (x)ȳ1d

 
f ( y , x) = y.
Remark 4. Notice that the approximation error of space-varying r2 (x)(ȳ2d + y2 ) β2 (x) + r2 (x)ȳ1d
DLMIs (29) and (39) is caused by the backward difference approxi-
dQ (x) Q (xk )−Q (xk−1 ) For simplicity, we assume that y2 ∈ [α1 , α2 ], where α1 = −0.2
mation for the spatial derivative (i.e., dx ≈ ). Based
φ and α2 = 0.5. Under this assumption, y2 can be represented by
on the error analysis result of the standard Euler method (Burden
& Faires, 2000), the error bound can be expressed as o , 0.5φ M y2 = h1 (y2 ) · α1 + h2 (y2 ) · α2 (64)
for small values of φ , where M , maxx∈[l1 ,l2 ] d2 Q (x)/dx2 F is a
where h1 (y2 ), h2 (y2 ) ∈ [0, 1] and
known constant. Obviously, this error bound converges to zero as
φ → 0. h1 (y2 ) + h2 (y2 ) = 1. (65)
 Author's personal copy

H.-N. Wu et al. / Automatica 48 (2012) 2535–2543 2541

320
240
y1 160
80
0
0
0.25 60
0.5 30 45
0.75 1 0 15

1.5
1
y2
0.5
0
0
0.25 45 60
0.5 30
x 0.75 1 0 15 Fig. 2. The Lyapunov matrix P (x) , diag{p1 (x), p2 (x)}.
t

Fig. 1. Open-loop profiles of evolution of ȳ1 (x, t ) and ȳ2 (x, t ).

By solving Eqs. (64) and (65), we get

α2 − y2 y2 − α1
h1 (y2 ) = , h2 (y2 ) = 1 − h1 (y2 ) = .
α2 − α1 α2 − α1
The terms h1 (y2 ) and h2 (y2 ) can be interpreted as the member-
ships. Define the fuzzy sets as ‘‘Big’’ and ‘‘Small’’. Then, the system
(1)–(3) with (63) is exactly represented by the following T–S fuzzy
PDE model of two rules:
Plant rule 1:

IF y2 is ‘‘Small’’
yt = Ay + A1 (x)y + Gu,1 (x)u + Gw,1 (x)w

THEN
z = C1 (x)y + D1 (x)u.
Fig. 3. The control matrices Ki (x) , ki,1 (x) ki,2 (x) , i ∈ {1, 2}.
 

Plant rule 2:

IF y2 is ‘‘Big’’ 2.4
yt = Ay + A2 (x)y + Gu,2 (x)u + Gw,2 (x)w

2.2
THEN y1
z = C2 (x)y + D2 (x)u 2
1.8
0
where 0.25 0.5 6 8 10
0.75 4
1 0 2
β1 (x) + r1 (x)(ȳ2d + α1 ) r1 (x)ȳ1d
 
A1 (x) = ,
r2 (x)(ȳ2d + α1 ) β2 (x) + r2 (x)ȳ1d
1.3
β1 (x) + r1 (x)(ȳ2d + α2 ) r1 (x)ȳ1d
 
A2 (x) = , 1.2
r2 (x)(ȳ2d + α2 ) β2 (x) + r2 (x)ȳ1d y2
1.1
T
Gu,i (x) = Gu (x) = b(x) 0 , Ci (x) = 1 0 , 1
  
0
0.25 8 10
T 0.5 6
Gw,i (x) = Gw (x) = 0.25 0.1 , Di (x) = 0, i ∈ {1, 2}. 4

x 0.75 1 0 2 t

The desired steady-state profiles ȳ1d and ȳ2d are chosen as ȳ1d = 2 Fig. 4. Closed-loop profiles of evolution of ȳ1 (x, t ) and ȳ2 (x, t ).
and ȳ2d = 1.1.
To implement the fuzzy controller, we use 200 equally dis-
t ≥ 10. Moreover, we can obtain maxt ∈[0,10] ∥u∥22 = 0.1204 < 1.5
tributed points on the spatial domain [0, 1] (i.e., nz = 200) at
and
which the states are observed and the distributed control action
is added, and thus φ = 0.005. Let ϕ = 1, wmax = 0.86, umax = 1.5,  10  10

ρ = 0.15, ε = 0.005, and y0 (x) = 0.1 sin(2π x)


0.2 sin(π x) .
T ∥z ∥22 dτ − ⟨y0 (·), P (·)y0 (·)⟩ − γ 2 ∥w∥22 dτ ≈ −0.1509
0 0
By solving the LMI optimization problem √ (56), the optimized level
of attenuation is obtained as γ = ϖ = 0.7444, the ap- which mean that u(x, t ) satisfies ∥u(·, t )∥22 < 1.5 and the control
proximate Lyapunov matrix P (xk ) = Q −1 (xk ) and control gain performance (11) is ensured, respectively.
matrices Ki (xk ), k ∈ N, i ∈ S can be obtained as shown in If the constraint ∥u(·, t )∥22 < 1.5 is not considered in the
Figs. 2 and 3. Using the Lyapunov matrix shown in Fig. 2, we get
1 T control design, by solving LMIs (53) with γ = 0.7444, then we
0 0
y (x)P (x)y0 (x)dx ≈ 0.0999. can obtain an unconstrained fuzzy controller (10). Applying this
Now, we apply the fuzzy controller (10) with the control controller to the PDE system (57)–(61) with the same w , we can get
gain matrices shown in Fig. 3 to the system (57)–(61).∞ Choose maxt ∈[0,20] ∥u∥22 = 4.6721 > 1.5, which implies that the control
w = (0.5 + x) cos(2t ) exp(−0.6t ), which satisfies 0 ∥w∥22 dt ≤ constraint does not hold. This simulation result indicates that it
0.9028 < γ −2 wmax . The actual closed-loop profiles of evolution may lead to a larger control action if we want to achieve the lower
of ȳ1 and ȳ2 are shown in Fig. 4, where we can see that ȳ1 and ȳ2 level of disturbance attenuation while not considering the control
numerically converge to the desired steady-state ȳ1d and ȳ2d , when constraint.
 Author's personal copy

2542 H.-N. Wu et al. / Automatica 48 (2012) 2535–2543

5. Conclusions Shang, H., Frobes, J. F., & Guay, M. (2004). Model predictive control for
quasilinear hyperbolic distributed parameter systems. Industrial & Engineering
and Chemistry Research, 43(9), 2140–2149.
In this paper, we have studied the problem of designing a Sira-Ramirez, H. (1989). Distributed sliding mode control in systems described
distributed H∞ fuzzy controller with constraint for a class of by quasilinear partial differential equations. Systems & Control Letters, 13(2),
nonlinear hyperbolic PDE systems. The outcome of the distributed 177–181.
Takagi, T., & Sugeno, M. (1985). Fuzzy identification of systems and its applications
H∞ fuzzy control problem with constraint is presented in terms
to modeling and control. IEEE Transactions on Systems, Man, Cybernetics, 15(1),
of a set of coupled D/ALMIs in space. A suboptimal distributed 116–132.
H∞ fuzzy controller with constraint is proposed to minimize Tanaka, K., Hori, T., & Wang, H. O. (2003). A multiple Lyapunov function approach to
the level of attenuation. The suboptimal fuzzy control design stabilization of fuzzy control systems. IEEE Transactions on Fuzzy Systems, 11(4),
582–589.
problem can be solved approximately by using the finite difference Tanaka, T., & Wang, H. O. (2001). Fuzzy control systems design and analysis: a linear
method in space and the existing LMI optimization techniques. matrix inequality approach. New York: Wiley.
Finally, the proposed design method is applied to the distributed Tuan, H. D., Apkarian, P., Narikiyo, T., & Yamamoto, Y. (2001). Parameterized linear
control of a nonlinear first-order hyperbolic PDE system, and the matrix inequality techniques in fuzzy control system design. IEEE Transactions
on Fuzzy Systems, 9(2), 324–332.
achieved simulation results show the effectiveness of the proposed Wang, J.-W., Wu, H.-N., & Li, H.-X. (2011). Distributed fuzzy control design of
controller. nonlinear hyperbolic PDE systems with application to nonisothermal plug-flow
reactor. IEEE Transactions on Fuzzy Systems, 19(3), 514–526.
Wu, H.-N. (2008). An ILMI approach to robust H2 static output feedback fuzzy
Acknowledgments control for uncertain discrete-time nonlinear systems. Automatica, 44(9),
2333–2339.
The authors gratefully acknowledge the helpful comments and Wu, H.-N., & Li, H.-X. (2007). Finite-dimensional constrained fuzzy control of a class
suggestions of the Associate Editor and anonymous reviewers, of nonlinear distributed process systems. IEEE Transactions on Systems, Man,
Cybernetics-Part B: Cybernetics, 37(5), 1422–1430.
which have improved the presentation of this paper. Wu, H.-N., & Li, H.-X. (2008). H∞ fuzzy observer-based control for a class
of nonlinear distributed parameter systems with control constraints. IEEE
Transactions on Fuzzy Systems, 16(2), 502–516.
References

Aksikas, I., Winkin, J. J., & Dochain, D. (2007). Optimal LQ-feedback regulation
of a nonisothermal plug flow reactor model by spectral factorization. IEEE Huai-Ning Wu was born in Anhui, China, on November
Transactions on Automatic Control, 52(7), 1179–1193. 15, 1972. He received the B.E. degree in automation from
Balas, M. J. (1979). Feedback control of linear diffusion processes. International Shandong Institute of Building Materials Industry, Jinan,
Journal of Control, 29(3), 523–534. China and the Ph.D. degree in control theory and control
Bamieh, B., Paganini, F., & Dahleh, M. A. (2002). Distributed control of spatially engineering from Xi’an Jiaotong University, Xi’an, China,
invariant systems. IEEE Transactions on Automatic Control, 47(7), 1091–1107. in 1992 and 1997, respectively.
Boyd, S., Ghaoui EI, L., Feron, E., & Balakrishnan, V. (1994). Linear matrix inequalities From August 1997 to July 1999, he was a Postdoctoral
in system and control theory. PA: SIAM. Researcher in the Department of Electronic Engineering
Burden, R. L., & Faires, J. D. (2000). Numerical analysis (7th ed.). New York: at Beijing Institute of Technology, Beijing, China. In
Brooks/Cole. August 1999, he joined the School of Automation Science
Chen, B.-S., & Chang, Y.-T. (2009). Fuzzy state-space modeling and robust observer- and Electrical Engineering, Beihang University (formerly
based control design for nonlinear partial differential systems. IEEE Transactions Beijing University of Aeronautics and Astronautics), Beijing. From December
on Fuzzy Systems, 17(5), 1025–1043. 2005 to May 2006, he was a Senior Research Associate with the Department of
Christofides, P. D. (2001). Nonlinear and robust control of PDE systems: methods and Manufacturing Engineering and Engineering Management (MEEM), City University
applications to transport-reaction processes. Boston, MA: Birkhäuser. of Hong Kong, Kowloon, Hong Kong. From October to December during 2006–2008
Christofides, P. D., & Daoutidis, P. (1996). Feedback control of hyperbolic PDE and from July to August in 2010, he was a Research Fellow with the Department of
systems. AIChE Journal, 42(11), 3063–3086. MEEM, City University of Hong Kong. From July to August in 2011, he was a Research
Curtain, R. F. (1982). Finite-dimensional compensator design for parabolic Fellow with the Department of Systems Engineering and Engineering Management,
distributed systems with point sensors and boundary input. IEEE Transactions City University of Hong Kong. He is currently a Professor with Beihang University.
on Automatic Control, 27(1), 98–104. His current research interests include robust control and filtering, fault-tolerant
Demetriou, M. A. (2010). Guidance of mobile actuator-plus-sensor networks control, distributed parameter systems, and fuzzy/neural modeling and control. He
for improved control and estimation of distributed parameter systems. IEEE is a member of the Committee of Technical Process Failure Diagnosis and Safety,
Transactions on Automatic Control, 55(7), 1570–1584. Chinese Association of Automation.
Feng, G. (2010). Analysis and synthesis of fuzzy control systems: a model based
approach. Boca Raton, FL: CRC.
Fister, K. R. (2001). Optimal control of harvesting with boundary control in a
predator–prey system. Applied Analysis, 77(1–2), 11–28. Jun-Wei Wang received the B.Sc. degree in Mathematics
Gahinet, P., Nemirovskii, A., Laub, A. J., & Chilali, M. (1995). LMI control toolbox for and Applied Mathematics and the M.Sc. degree in System
use with Matlab. Natick, MA: The Math.Works Inc. Theory from Harbin Engineering University, Harbin, China,
Gao, H., & Chen, T. (2007). Stabilization of nonlinear systems under variable in 2007 and 2009, respectively. He is now studying
sampling: a fuzzy control approach. IEEE Transactions on Fuzzy Systems, 15(5), for the Ph.D. degree in Control Science and Engineering
972–983. in Beihang University (Beijing University of Aeronautics
Goh, B. S., Leitmann, G., & Vincent, T. L. (1974). Optimal control of a prey–predator and Astronautics), Beijing, China. His research interests
system. Mathematical Biosciences, 19(3–4), 263–286. include robust control and filtering, stochastic systems,
Guerra, T. M., & Vermeiren, L. (2004). LMI-based relaxed nonquadratic stabilization distributed parameter systems, and fuzzy modeling and
conditions for nonlinear systems in the Takagi–Sugeno’s form. Automatica, control.
40(5), 823–829.
Keulen, van B. (1993). H∞ -control for distributed parameter systems: a state-space
approach. Boston, MA: Birkhauser.
Khalil, H. K. (2002). Nonlinear systems (3rd ed.). Upper Saddle River, NJ: Prentice-
Hall. Han-Xiong Li received the B.E. degree in aerospace
Lee, D. H., Park, J. B., & Joo, Y. H. (2011). Approaches to extended non-quadratic engineering from the National University of Defence
stability and stabilization conditions for discrete-time Takagi–Sugeno fuzzy Technology, Changsha, China, in 1982, the M.E. degree in
systems. Automatica, 47(3), 534–538. electrical engineering from Delft University of Technology,
Levine, I. N. (2009). Quantum chemistry (6th ed.). New Jersey: Pearson Prentice Hall. Delft, The Netherlands, in 1991, and the Ph.D. degree in
Li, H.-X., & Qi, C. (2010). Modeling of distributed parameter systems for applications- electrical engineering from the University of Auckland,
a synthesized review from time–space separation. Journal of Process Control, Auckland, New Zealand, in 1997.
20(8), 891–901. Currently, he is a full professor in the Department
Mozelli, L. A., Palhares, R. M., Souza, F. O., & Mendes, E. M. A. M. (2009). Reducing of Systems Engineering and Engineering Management,
conservativeness in recent stability conditions of TS fuzzy systems. Automatica, the City University of Hong Kong. Over the past twenty
45(6), 1580–1583. years, he has had opportunities to work in different fields,
Pazy, A. (1983). Semigroups of linear operators and applications to partial differential including military service, industry, and academia. His research experience and
equations. New York: Springer-Verlag. accomplishment include, fuzzy-PID for process control, a pioneering 3-domain
Ray, W. H. (1981). Advanced process control. New York: McGraw-Hill. fuzzy logic system for modeling and control, intelligent modeling and control of
 Author's personal copy

H.-N. Wu et al. / Automatica 48 (2012) 2535–2543 2543

spatio–temporal dynamic systems, with application to the thermal cure process Dr. Li serves as Associate Editor of IEEE Transactions on Systems, Man &
and fluid dispensing for IC packaging. He published over 100 SCI journal papers Cybernetics, part-B, and IEEE Transactions on Industrial Electronics. He was awarded
(nearly half of them in IEEE Transactions and ASME Transactions) with h-index 22. the Distinguished Young Scholar (overseas) by the China National Science
His current research interests are in intelligent modeling and control, integrated Foundation in 2004, a Chang Jiang chair professor by the Ministry of Education,
process design and control, and distributed parameter systems. China in 2006. He is a fellow of the IEEE.

View publication stats