www.ietdl.

org
Published in IET Control Theory and Applications
Received on 30th May 2011
Revised on 21st January 2012
doi: 10.1049/iet-cta.2011.0310

ISSN 1751-8644

Fuzzy control of non-linear systems using

parameter-dependent polynomial fuzzy model
H.K. Lam1 L.D. Seneviratne1 X. Ban2
1 Division of Engineering, King’s College London, Strand, London WC2R 2LS, UK
2 Center for ControlTheory and GuidanceTechnology, Harbin Institute of Technology, Harbin, People’s Republic of China
E-mail: [email protected]

Abstract: This study investigates the stability of parameter-dependent polynomial-fuzzy-model-based (PDPFMB) control
systems. A parameter-dependent polynomial fuzzy model is proposed to represent a non-linear plant. A parameter-dependent
polynomial fuzzy controller is then employed to stabilise the non-linear plant. The stability of PDPFMB control systems
is investigated using the sum-of-squares (SOS) technique based on a parameter-dependent Lyapunov function candidate.
With the consideration of the information of system parameters, parameter-dependent SOS-based stability conditions are
obtained to determine the system stability and polynomial feedback gains. A feasible solution to the stability conditions can
be obtained numerically using the third-party Matlab toolbox SOSTOOLS. Unlike the membership functions, as the system
parameters are not single signed functions, the traditional analysis approach obtaining the stability conditions cannot be
applied. Instead, the membership functions and system parameters are considered as symbolic variables for the construction
of the parameter-dependent SOS-based stability conditions. A simulation example is given to demonstrate the merits of the
proposed approach.

1 Introduction Other fuzzy control schemes such as adaptive, switching

and sliding-mode control can also be found in the literature.
Since the introduction of the Takagi–Sugeno (T-S) fuzzy These control methodologies are more suitable to deal with
model [1, 2], fuzzy-model-based (FMB) control [3] has non-linear system subject to parameter uncertainties [24–26].
drawn the attention of fuzzy control researchers for the past By representing the uncertain but bounded non-linear terms
decades because of its superiority dealing with complex and with a fuzzy logic system, adaptive rules were obtained
ill-defined non-linear plants. With the T-S fuzzy model, a with the consideration of system stability to update the
non-linear plant can be represented in a favourable form parameters of fuzzy logic system in an online manner for
to facilitate the stability analysis and control synthesis estimation of non-linear terms [27, 28]. When the values of
with the extension of the linear state-feedback control the parameters of fuzzy logic system converge, it implies
theory. The most popular fuzzy controller is the state- that the values of the uncertain parameters are optimally
feedback type [4, 5], which is referred to as simply estimated. Feedback compensation can be achieved by
fuzzy controller hereafter. Basic stability conditions [4, the adaptive fuzzy controller using the estimated values
5] in the form of linear matrix inequalities (LMIs) [6] of uncertain parameters. By using the switching [29–36]
were obtained based on the Lyapunov stability theorem to or sliding-mode [37–39] control scheme, high-frequency
determine the system stability. The FMB control system switching components are employed to estimate the values
is guaranteed to be asymptotically stable if there exists of uncertain parameters. Similar to the adaptive control
a common solution to a set of LMIs. With convex approach, the estimated parameters are employed by the
programming techniques, a feasible solution to the stability fuzzy controller for the control process. Compared with the
conditions can be found numerically. Under the parallel state-feedback-type fuzzy controller, the adaptive, switching
distributed compensation design concept, various relaxed and sliding-mode fuzzy controllers are more complicated in
LMI-based stability conditions were obtained [7–12]. By terms of structure and analysis.
expanding the degree of fuzzy summations, the work Recently, the T-S fuzzy model has been extended to the
in [12] generalised the stability analysis results with polynomial fuzzy model [40, 41]. It allows polynomials
Pólya’s theorem. It was revealed in [13–23] that the of system states in the consequent part of fuzzy rules
information of membership functions plays an important and represents a non-linear plant as an average weighted
role in further relaxing the stability conditions. With sum of local polynomial systems. A general approach [42]
the consideration of the information of membership using the sector non-linearity concept [4, 5] and the Taylor
functions, more relaxed stability analysis results can be series was proposed to construct the polynomial fuzzy
obtained. model for a given non-linear plant. As polynomials exist

IET Control Theory Appl., 2012, Vol. 6, Iss. 11, pp. 1645–1653 1645
doi: 10.1049/iet-cta.2011.0310 © The Institution of Engineering and Technology 2012
www.ietdl.org
in the local models, the traditional LMI-based stability positive, semi-positive, negative and semi-negative definite
analysis technique cannot be applied directly. Instead, the matrix M, respectively.
sum-of-squares (SOS)-based stability analysis approach was
employed [40, 41, 43] and basic stability conditions in terms 2.2 Parameter-dependent polynomial fuzzy model
of SOS were obtained. A feasible solution to the SOS-
based stability conditions can be found numerically using the The dynamics of a parameter-dependent non-linear plant is
third-party Matlab toolbox SOSTOOLS [44]. Some relaxed described by a parameter-dependent polynomial fuzzy model
SOS-based stability conditions can be found in [42] with the with p rules of the following format
consideration of the information of system parameters and
membership functions. Rule i: IF f1 (x(t)) is M1i AND · · · AND f (x(t)) is Mi
In this paper, we extend the polynomial fuzzy model THEN ẋ(t) = Ai (x(t), h(t))x̂(x(t))
to the parameter-dependent polynomial fuzzy model of
which the local polynomial depends on some changing + Bi (x(t), h(t))u(t) (1)
system parameters, which makes the stability analysis
more complicated and difficult. To deal with the system where Mαi are fuzzy terms of rule i corresponding to
parameters under the SOS-based stability analysis approach, the functions fα (x(t)), α = 1, 2, . . . , ; i = 1, 2, . . . , p; 
it is possible to consider them as augmented state variables. is a positive integer; x(t) ∈ n is the system state vector;
However, the form of augmented system plays an important Ai (x(t), h(t)) ∈ n×N , Bi (x(t), h(t)) ∈ n×m are the known
role to obtain a stable control design and it is still an parameter-dependent non-linear system and input matrices,
open question to obtain a favourable form of polynomial respectively; h(t) ∈ s is the parameter vector; u(t) ∈ m is
model for the stability analysis. To investigate the the input vector; and x̂(x(t)) ∈ N is a vector of monomials
stability of parameter-dependent polynomial-fuzzy-model- in x(t). It is assumed that x̂(x(t)) = 0 iff x(t) = 0.
based (PDPFMB) control systems, a parameter-dependent Denoting h(t) = [h1 (t), h2 (t), . . . , hs (t)] where hk (t),
polynomial Lyapunov function candidate is considered. The k = 1, 2, . . . , s, are parameter-dependent functions, the
information of system parameters is considered in order to system and input matrices Ai (x(t), h(t)) and Bi (x(t), h(t))
construct a valid Lyapunov function. Parameter-dependent are defined as follows
SOS-based stability conditions are developed to guarantee
the system stability based on the Lyapunov stability theorem. 
s
Ai (x(t), h(t)) = hk (t)Aik (x(t)), ∀i (2)
To facilitate the solution searching using SOSTOOLS,
k=1
both the system parameters and membership functions are
denoted as symbolic variables. Some slack polynomial 
s

matrices carrying the information of system parameters and Bi (x(t), h(t)) = hk (t)Bik (x(t)), ∀i (3)
k=1
membership functions are introduced to relax the stability
analysis results.
where Aik (x(t)) ∈ n×N and Bik (x(t)) ∈ n×m , i = 1,
The organisation of this paper is as follows. In
2, . . . , p, k = 1, 2, . . . , s, are the polynomial system and input
Section 2, basic notations used in this paper are
matrices.
introduced. The parameter-dependent polynomial fuzzy
The system dynamics of the parameter-dependent
model representing the non-linear plant is presented.
non-linear plant is described by
A parameter-dependent polynomial fuzzy controller is
proposed for the control process. In Section 3, the stability 
p
of the PDPFMB control systems is investigated based ẋ(t) = wi (x(t))(Ai (x(t), h(t))x̂(x(t))
on a parameter-dependent polynomial Lyapunov function i=1
candidate. Parameter-dependent SOS-based stability cond-
itions are obtained based on the Lyapunov stability + Bi (x(t), h(t))u(t))
theorem to guarantee the system stability and synthesise 
p

s

the polynomial fuzzy controller. In Section 4, simulation = wi (x(t))hk (t)(Aik (x(t))x̂(x(t))

examples are given to illustrate the merits of the i=1 k=1
proposed control scheme. In Section 5, a conclusion is + Bik (x(t))u(t)) (4)
drawn.
where

2 Notations and preliminaries μMli (fl (x(t)))
l=1
wi (x(t)) = p  ≥ 0, ∀i (5)
2.1 Notations k=1 l=1 μMlk (fl (x(t)))


p
Throughout the paper, the following notations are adop- wi (x(t)) = 1 (6)
ted [43]. A monomial in x(t) = [x1 (t), x2 (t), . . . , xn (t)] is i=1
a function of the form x1d1 (t)x2d2 (t) · · · xndn (t), where di ,
i = 1, 2, . . . , n, are non-negative
 integers. The degree of a
wi (x(t)) are the normalised grades of membership,
monomial is defined as d = ni=1 di . A polynomial p(x(t)) μMαi (fα (x(t))), α = 1, 2, . . . , , are the grades of member-
is defined as a finite linear combination of monomials with ship corresponding to the fuzzy term Mαi .
real coefficients. A polynomial  p(x(t)) is an SOS if it
can be written as p(x(t)) = mj=1 qj (x(t))2 where qj (x(t)) Assumption 1: It is assumed that the parameter functions
is a polynomial and m > 0 denotes the number of terms. hk (t) are bounded functions such that hk ≤ hk (t) ≤ h̄k and
Hence, it can be seen that p(x(t)) ≥ 0 if it is an SOS. The qk ≤ ḣk (t) ≤ q̄k for all k where hk , h̄k , qk and q̄k are constant
expressions of M > 0, M ≥ 0, M < 0 and M ≤ 0 denote a scalars.

1646 IET Control Theory Appl., 2012, Vol. 6, Iss. 11, pp. 1645–1653
© The Institution of Engineering and Technology 2012 doi: 10.1049/iet-cta.2011.0310
 www.ietdl.org
2.3 Polynomial fuzzy controller theorem. In the following analysis, for brevity, the time t
associated with variables is dropped for situation without
A polynomial fuzzy controller is proposed to stabilise the ambiguity, for example hk (t), x(t), x̃(t) and x̂(x(t)) are
parameter-dependent non-linear plant represented by the denoted as hk , x, x̃ and x̂(x), respectively. Furthermore,
parameter-dependent polynomial fuzzy model (4), that is wi (x(t)) is denoted as wi .
x(t) → 0 as time t → ∞. The polynomial fuzzy controller We consider the following parameter-dependent
is described by the following p rules. Lyapunov function candidate to investigate the stability of
j
Rule j: IF f1 (x(t)) is M1 AND · · · AND f (x(t)) is M
j the PDFMB control system (10)
THEN u(t) = Gj (x(t), h(t))x̂(x(t)) (7) V = x̂(x)T P−1 (x̃, h)x̂(x) (11)
where s
where 0 < P(x̃, h) = P(x̃, h)T ∈ N ×N , P(x̃, h) = h k Pk

s
(x̃), Pk (x̃) = Pk (x̃)T ∈ N ×N for all k.
k=1
−1
Gj (x(t), h(t)) = hk (t)Njk (x(t))P (x̃(t), h(t)), ∀ j (8)
k=1
Remark 1: According to the Lyapunov stability theorem [45],
Njk (x(t), h(t)) ∈ m×N are polynomial matrices and P(x̃(t), if there exists a Lyapunov function candidate such that V >
h(t)) ∈ N ×N is a polynomial matrix to be deter- 0 and V̇ ≤ 0 for x = 0, the PDPFMB control system (10) is
mined; x̃(t) = (xk 1 (t), xk 2 (t), . . . , xk q (t)); K = {k1 , k2 , . . . , kq } guaranteed to be asymptotically stable, that is x(t) → 0 as
denotes the set of row number that the entries of the entire time t → ∞.
row of Bik (x) for all k are all zero [40, 43].
The polynomial fuzzy controller is defined as In the following, parameter-dependent SOS-based
p conditions are developed to guarantee V > 0, which is
u(t) = wj (x(t))Gj (x(t), h(t))x̂(x(t)) implied by P(x̃, h) > 0. Introducing slack polynomial matri-
j=1 ces 0 ≥ Rk (x̃) = Rk (x̃)T ∈ N ×N , with the consideration of
Assumption 1, we have hk Rk (x̃) ≥ h̄k Rk (x̃). It follows that

p

s
= wj (x(t))hk (t)Njk (x(t))P−1 (x̃(t), h(t))x̂(x(t)) 
s
j=1 k=1 P(x̃, h) = hk (Pk (x̃) + Rk (x̃) − Rk (x̃))
(9) k=1


s 
s
2.4 PDPFMB control system ≥ h̄k Rk (x̃) + hk (Pk (x̃) − Rk (x̃))
k=1 k=1
A PDPFMB control system is formed by connecting
the parameter-dependent polynomial fuzzy model (4) and 
s

polynomial fuzzy = (h̄k Rk (x̃) + hk (Pk (x̃) − Rk (x̃)))

p controller (9) pin aclosed loop. With the
property of w
i=1 i (x(t)) = i=1
p
w
j=1 i (x(t))w j (x(t)) =
k=1

1, from (4) and (9), we have 

s
 + (hk − hk )(Pk (x̃) − Rk (x̃)) (12)

p
 s
ẋ(t) = wi (x(t))hk (t) Aik (x(t))x̂(x(t)) k=1

i=1 k=1
Remark 2: It can be seen from (12) that P(x̃,  h) > 0 which

p

s
can be achieved by Rk (x̃) ≤ 0 for all k, sk=1 (h̄k Rk (x) +
+ Bik (x(t)) wj (x(t))hl (t)Njl (x(t))P−1 hk (Pk (x̃) − Rk (x̃)) > 0 and Pk (x̃) − Rk (x̃) > 0 for all k.
j=1 l=1
 These conditions imply that hk = 0 for all k resulting in
P(x̃, h) = 0 is not be possible to happen. Considering the
× (x̃(t), h(t))x̂(x(t)) conditions Rk (x̃) ≤ 0 and Pk (x̃) − Rk (x̃) > 0 for all k, if hk
 and h̄k are of opposite sign, that is hk < 0 and h̄k > 0 for

p

p

s
all k, the condition sk=1 (h̄k Rk (x̃) + hk (Pk (x̃) − Rk (x̃)) > 0
= wi (x(t))wj (x(t))hk (t) Aik (x(t)) will never be satisfied. Otherwise, at least a pair of hk and h̄k
i=1 j=1 k=1 for any k is of the same sign, which rules out the possibility

s of hk = 0 for all k at any instant leading to P(x̃, h) = 0.
+ Bik (x(t)) hl (t)Njl (x(t))P−1

l=1 Remark 3: Referring to (11), it is difficult to obtain a
condition to guarantee the polynomial matrix P(x̃, h) >
× (x̃(t), h(t)) x̂(x(t)) (10) 0 because the parameter functions hk are not single
signed. With the slack matrices Rk (x̃), the information of
membership functions, that is hk and h̄k , is brought to the
The control objective is to determine the polynomial feed-
conditions guaranteeing P(x̃, h) > 0.
back gains, Njl (x(t)) and the polynomial matrix P(x̃(t), h(t)),
To show the effectiveness of the information of parameter
such that the fuzzy controller (9) drives the system states
functions, we consider an example that all hk > 0 resulting
towards the origin, that is x(t) → 0 as time t → ∞.
in h̄k ≥ hk > 0. It is straightforward to see that P(x̃, h) > 0
can be achieved if Pk (x̃) > 0 for all k are achieved. In
3 Stability analysis this case, theres must exist polynomial matrices Rk (x̃) ≤
0 such that k=1 (h̄k Rk (x) + hk (Pk (x̃) − Rk (x̃))) > 0 and
In this section, the stability of the PDPFMB control Pk (x̃) − Rk (x̃) > 0 for all k. Let Rk (x̃) = −εI ≤ 0 for all
system (10) is investigated using the Lyapunov stability k where ε is a positive scalar and I is the identity matrix

IET Control Theory Appl., 2012, Vol. 6, Iss. 11, pp. 1645–1653 1647
doi: 10.1049/iet-cta.2011.0310 © The Institution of Engineering and Technology 2012
www.ietdl.org
of
s compatible dimension. These  two inequalities become From (13) and (14), we have
k=1 (−h̄k εI + hk (Pk (x̃) + εI)) =
s
k=1 (−(h̄k − hk )εI + hk 
Pk (x̃)) > 0 and Pk (x̃) + εI > 0, which can be achieved with 
p

p

s 
s 
Pk (x̃) > 0 for all k and a sufficiently small value of ε. V̇ ≤ z T
wi w j hk hl Ãik (x)Pl (x̃)
However, it is not true on the other way round. i=1 j=1 k=1 l=1
T
+ Pl (x̃)Ãik (x) + B̃ik (x)Njl (x) + Njl (x)T B̃ik (x)T
To proceed further with the stability analysis, parameter- 
dependent SOS-based stability conditions will be developed  ∂Pl (x̃) κ
to guarantee V̇ < 0 for all x = 0, which implies the − Ãik (x)x̂(x)
κ∈K
∂xκ
asymptotic stability of the PDPFMB control system (10).
s 
 
Denote z = P−1 (x̃, h)x̂(x), Ãik (x) = L(x)Aik (x) and
B̃ik (x) = L(x)Bik (x), where L(x) ∈ N ×n is a polynomial − q̄k Sk (x̃) + qk (Pk (x̃) − Sk (x̃))) z
matrix of which the (α, β)-entry is defined as Lαβ (x) = k=1

∂ x̂α (x)/∂xβ , α = 1, 2, . . ., N , β = 1, 2, . . ., n. From (10) 

s

and (11), it follows that − zT (ḣk − qk )(Pk (x̃) − Sk (x̃))z

k=1

˙ T P−1 (x̃, h)x̂(x)
V̇ = x̂(x) 
p

p

s 
s
=z T
wi wj hk hl ijkl (x, h)
˙
+ x̂(x)T P−1 (x̃, h)x̂(x)
−1
+ x̂(x)T Ṗ (x̃, h)x̂(x) i=1 j=1 k=1 l=1
 s 

p
p
s  
= T
wi wj hk z Ãik (x)P(x̃, h) − q̄k Sk (x̃) + qk (Pk (x̃) − Sk (x̃)) z
i=1 j=1 k=1 k=1


s 
s
T
+ P(x̃, h)Ãik (x) + hl (B̃ik (x)Njl (x) − zT (ḣk − qk )(Pk (x̃) − Sk (x̃))z (15)
l=1 k=1

+ Njl (x) B̃ik (x) ) z − zT Ṗ(x̃, h)z
T T where


p

p

s 
s  T ijkl (x, h) = Ãik (x)Pl (x̃) + Pl (x̃)Ãik (x)
T

= wi wj hk hl z T
Ãik (x)Pl (x̃) + Pl (x̃)Ãik (x)
i=1 j=1 k=1 l=1
 + B̃ik (x)Njl (x) + Njl (x)T B̃ik (x)T
+ B̃ik (x)Njl (x) + Njl (x) B̃ik (x) zT T 
 ∂Pl (x̃) κ
 s  − Ãik (x)x̂(x) (16)
  s  ∂Pk (x̃) ∂xκ
− zT ḣk Pk (x̃) + hk ẋκ z κ∈K

k=1 k=1 κ∈K

∂xκ    
 Define (x, h, w) = pi=1 pj=1 wi wj ( sk=1 sl=1 hk hl ijkl
p
 p
s  s
T (x) − sk=1 (q̄k Sk (x̃) + qk (Pk (x̃) − Sk (x̃)))) where w =
= wi wj hk hl z Ãik (x)Pl (x̃) + Pl (x̃)Ãik (x)
T

i=1 j=1 k=1 l=1

[w1 , w2 , . . . , wp ]. It can be seen from (15) that V̇ < 0 for
x = 0 can be achieved, which implies the asymptotic
+ B̃ik (x)Njl (x) + Njl (x)T B̃ik (x)T stability of the PDFMB control system (10) based on the
 Lyapunov stability theorem, if (x, h, w) < 0 and Pk (x̃) −
 ∂Pl (x̃) κ s
− Ãik (x)x̂(x) z − zT ḣk Pk (x̃)z (13) Sk (x̃) > 0 for all k are satisfied.
κ∈K
∂xκ k=1
Remark 4: By using the stability analysis approach in [40],
where Aκik (x)
denotes the κth row vector of Aik (x). SOS-based stability conditions for the PDFMB control
To deal with the last term in (13), we introduce the slack system (10)  canbe obtained sas follows.
s From (15), we have
polynomial matrices 0 ≥ Sk (x̃) = Sk (x̃)T ∈ N ×N for all k. V̇ ≤ 12 zT pi=1 pj=1 wi wj hk hl (ijkl (x, h) +
 k=1 l=1
With Assumption 1, we have ḣk Sk (x̃) ≥ q̄k Sk (x̃) for all k.  (x, h)) − 2 sk=1 (q̄k Sk (x̃) + qk (Pk (x̃) − Sk (x̃))) z − zT
jikl
s
k=1 (ḣk − qk )(Pk
(x̃) − Sk (x̃))z. V̇ < 0 for x = 0 can
Considering the last term in (13), it follows that
s s
beachieved if k=1 l=1 ηk ηl (ijkl (x, h) + jikl (x, h)) −

s 
s
2 sk=1 (q̄k Sk (x̃) + qk (Pk (x̃) − Sk (x̃))) and Pk (x̃) − Sk (x̃) are
ḣk Pk (x̃) = ḣk (Pk (x̃) + Sk (x̃) − Sk (x̃)) SOS for all i, j and all combinations of ηk and ηl which
k=1 k=1
denote the lower or upper bound of hk and hl , respectively.

s 
s
In addition, it is also required that V > 0 for x = 0.
≥ q̄k Sk (x̃) + ḣk (Pk (x̃) − Sk (x̃))
k=1 k=1
Referring to (15), the solution to the second set

s
of inequalities can be found numerically using the
= (q̄k Sk (x̃) + qk (Pk (x̃) − Sk (x̃))) SOSTOOLS [44]. To deal with the first inequality, we
k=1 consider wi , wj , hk and hl as symbolic variables. In

s
order to incorporate the information of system parameters
+ (ḣk − qk )(Pk (x̃) − Sk (x̃)) (14) and membership functions into the parameter-dependent
k=1 SOS-based conditions, we consider the properties of

1648 IET Control Theory Appl., 2012, Vol. 6, Iss. 11, pp. 1645–1653
© The Institution of Engineering and Technology 2012 doi: 10.1049/iet-cta.2011.0310
 www.ietdl.org
p
i=1 wi − 1 = 0 and wi ≤ wi ≤ wi for all i, where ν T (Tij (x, h, w) − ε3 (x, h, w)I)ν are SOS ∀ i, j
0 ≤ wi ≤ wi ≤ 1 for all i and hk ≤ hk ≤ h̄k for all k
ν T (Ukl (x, h, w) − ε4 (x, h, w)I)ν are SOS ∀ k, l
are pre-defined scalars. Thereby, introducing the slack
polynomial matrices M(x, h, w) = M(x, h, w)T ∈ N ×N , 0 ≤ ν T (Pk (x̃) − Rk (x̃) − ε5 (x̃)I)ν are SOS ∀ k
Tij (x, h, w) = Tij (x, h, w)T ∈ N ×N and 0 ≤ Ukl (x, h, w) =
ν T (Pk (x̃) − Sk (x̃) − ε6 (x̃)I)ν are SOS ∀ k
Ukl (x, h, w)T ∈ N ×N , we have  s 

 p  νT (h̄k Rk (x) + hk (Pk (x̃) − Rk (x̃))) − ε7 (x̃)I ν is SOS

wi − 1 M(x, h, w) = 0 (17) k=1

i=1 −ν T ((x, h, w) + ε8 (x, h, w))ν is SOS


p p

(wi − wi )(wj − wj )Tij (x, h, w) ≥ 0 (18)

i=1 j=1
where ν ∈ N is an arbitrary vector independent of x;
ε1 (x̃), ε2 (x̃), ε3 (x, h, w), ε4 (x, h, w), ε5 (x̃), ε6 (x̃), ε7 (x̃)

s 
s
and ε8 (x, h, w) are predefined scalar polynomials being
(hk − hk )(h̄l − hl )Ukl (x, h, w) ≥ 0 (19)
SOS; wi , wi , hk , h̄k , qk and q̄k are predefined scalars
k=1 l=1
such that wi ≤ wi ≤ wi for all i, hk ≤ hk ≤ h̄k and
qk ≤ q̇k ≤ q̄k for all k; the polynomial feedback gains
Adding (17)–(19) to (x, h, w), we have 
Gj (x(t), h(t)) are defined in (8); P(x̃, h) = sk=1 hk Pk (x̃);
 the polynomial matrix (x, h, w) is defined in (20) which

p

p

s 
s
(x, h, w) = wi wj hk hl ijkl (x) includes the polynomial matrix ijkl (x, h) in (16) containing
i=1 j=1
Njl (x).
k=1 l=1
s 

 
− q̄k Sk (x̃) + qk (Pk (x̃) − Sk (x̃)) Remark 6: SOSTOOLS [44] is employed to find a feasible
k=1 solution of the parameter-dependent SOS-based stability
 

p conditions in Theorem 1. The variables wi and hk are
+ wi − 1 M(x, h, w) considered as symbolic variables. To slightly reduce the
i=1 computational demand,
 the property of the membership

p

p functions, wp = 1 − p−1
i=1 wi , can be used. However, the
+ (wi − wi )(wj − wj )Tij (x, h, w) polynomial matrix M(x, h, w) in (17) will be vanished.
i=1 j=1


s 
s
+ (hk − hk )(h̄l − hl )Ukl (x, h, w) (20) 4 Simulation example
k=1 l=1
A two-rule parameter-dependent polynomial fuzzy model in
the form of (4) is considered. The system and input matrices
Remark 5: The main purpose of the slack polynomial
are chosen as
matrices, M(x, h), Sk (x̃), Tij (x, h), Ukl (x, h), is to bring
the information of system parameters and membership
functions, that is wi , wi , hk and h̄k , to the stability analysis 1.59 − 0.25x22 −7.29 + 0.83x2
A11 (x2 ) =
for relaxing the stability conditions. 0.12 −0.1
0.88 + 3.32x2 3.62 − 7.56x2
Obviously, as (x, h, w) ≤ (x, h, w), (x, h) < 0 is A12 (x2 ) =
0 0
implied by (x, h, w) < 0. Based on the Lyapunov stability
theorem, the parameter-dependent Lyapunov function 4.41 − 6.32x2 −0.9 + 2.56x2
A13 (x2 ) =
candidate (11) satisfying V > 0 and V̇ < 0 implies the 0 0
asymptotic stability of PDPFMB control system (10). The 2 − 0.62x22 −4.33 + 2.16x2
stability analysis result for the PDFMB control system (10) A21 (x2 ) =
0.55 −0.05
is summarised in the following theorem.
2.61 + 1.32x2 0.15 + 6.16x2
A22 (x2 ) =
Theorem 1: The PDPFMB control system (10), formed 0 0
by a non-linear plant [represented by the parameter- −0.26 3.52
dependent polynomial fuzzy model (4)] and the polynomial A33 (x2 ) =
0 0
fuzzy controller (9) connected in a closed loop, is
guaranteed to be asymptotically stable if there exist 1 + 8.52x22
B11 (x2 ) =
polynomial matrices M(x, h) = M(x, h)T ∈ N ×N , Njl (x) ∈ 0
m×N , Pk (x̃) = Pk (x̃)T ∈ N ×N , Rk (x̃) = Rk (x̃)T ∈ N ×N ,
0.2
Sk (x̃) = Sk (x̃)T ∈ N ×N , Tij (x, h) = Tij (x, h)T ∈ n×n , Ukl B12 (x2 ) =
0
(x, h) = Ukl (x, h)T ∈ n×n such that the following SOS-
based conditions are satisfied −0.3
B13 (x2 ) =
0
−ν T (Rk (x̃) + ε1 (x̃)I)ν are SOS ∀ k 4 + 10.75x22
B21 (x2 ) =
−ν (Sk (x̃) + ε2 (x̃)I)ν are SOS ∀ k
T 0

IET Control Theory Appl., 2012, Vol. 6, Iss. 11, pp. 1645–1653 1649
doi: 10.1049/iet-cta.2011.0310 © The Institution of Engineering and Technology 2012
www.ietdl.org
−0.5
B22 (x2 ) =
0
0.3
B23 (x2 ) =
0

The membership functions are chosen as w1 (x2 ) =

μM11 (x2 ) = 1 − 1/1 + e−(x2 /1.5) and w2 (x2 ) = μM12 (x2 ) = 1 −
w1 (x2 ) such that 0 ≤ wi (x2 ) ≤ 1, i = 1, 2. The parameter-
dependent functions are chosen as h1 (t) = 1, h2 (t) = sin(t)
and h3 (t) = (1/2) cos(t) such that h1 = h̄1 = 1, −h2 = h̄2 =
1 and −h3 = h̄3 = 0.5; q1 = q̄1 = 0; −q2 = q̄2 = 1 and
−q3 = q̄3 = 0.5.
A two-rule polynomial fuzzy controller in the form
of (9) is proposed to control the parameter-dependent non-
linear plant. The parameter-dependent SOS-based stability
conditions in Theorem 1 are employed to determine the
a
system stability and feedback gains. Choosing εr = 0.001,
r = 1, 2, . . . , 8, the polynomial matrices Ukl (x2 , h, w) for
k, l = 1, 2 as polynomials of degree 2 in x2 where h =
[h1 h2 h3 ] and w = [w1 w2 ], the matrix P(x2 ) as a
polynomial of degree 2 in x2 and other matrices as
polynomials of degree 0, it is found that

Pk(11) (x2 ) Pk(12) (x2 )

Pk (x2 ) =
Pk(21) (x2 ) Pk(22) (x2 )

for k = 1, 2, 3, where

P1(11) (x2 ) = 0.7174 × 10−2 − 0.3000x2 × 10−12

+ 0.4655x22 × 10−10
P1(12) (x2 ) = P1(21) (x2 ) = 0.2693 × 10−3
+ 0.1307x2 × 10−11 + 0.3605x22 × 10−11 b
P1(22) (x2 ) = 0.1338 × 10−2 + 0.2013x2 × 10−11
+ 0.1516x22 × 10−10
P2(11) (x2 ) = −0.4108 × 10−3 − 0.6436x2 × 10−12
− 0.7265x22 × 10−10

Fig. 2 System responses of x(t) and control signal u(t) for the
PDPFMB control system with x= [10 10] corresponding to the
polynomial fuzzy controller with P(x2 ) as a polynomial of degree
2 in x2 and Njl as polynomials of degree 0
a Response of x1 (t)
b Response of x2 (t)
c Control signal u(t)
Fig. 1 Phase portrait of the PDPFMB control system with
various initial conditions indicated by ‘◦’
Polynomial fuzzy controller is with P(x2 ) as a polynomial of degree 2 in
x2 and Njl as polynomials of degree 0

1650 IET Control Theory Appl., 2012, Vol. 6, Iss. 11, pp. 1645–1653
© The Institution of Engineering and Technology 2012 doi: 10.1049/iet-cta.2011.0310
 www.ietdl.org
P2(12) (x2 ) = P2(21) (x2 ) = 0.7787 × 10−4
+ 0.1051x2 × 10−11 + 0.8663x22 × 10−10
P2(22) (x2 ) = 0.1458 × 10−5 − 0.8013x2 × 10−11
+ 0.6661x22 × 10−10
P3(11) (x2 ) = −0.3818 × 10−5 + 0.1102x2 × 10−7
+ 0.9711x22 × 10−7
P3(12) (x2 ) = P3(21) (x2 ) = −0.7115 × 10−7
− 0.1195x2 × 10−7 − 0.4737x22 × 10−11
X3(22) (x2 ) = 0.6144 × 10−9 − 0.1303x2 × 10−7
+ 0.9277x22 × 10−7
N11 = [−0.9024 0.3020 × 10−3 ]
N12 = [0.2866 × 10−2 −0.2589 × 10−3 ]
N13 = [−0.1735 × 10−3 0.1293 × 10−3 ]
N21 = [−1.1372 −0.1890 × 10−3 ]
N22 = [0.2866 × 10−2 −0.2589 × 10−3 ]
N23 = [−0.672 × 10−3 0.1232 × 10−3 ]

The polynomial fuzzy controller is employed to control the

non-linear plant subject to various initial conditions. The
phase portrait is shown in Fig. 1. It can be seen from
the figure that the non-linear plant can be stabilised by
the proposed polynomial fuzzy controller. Fig. 2 shows the
system responses and control signal of the PDPFMB control
system with x(0) = [10 10].
For demonstration purposes, considering the matrix P(x2 )
as a polynomial of degree 0, the matrices Njl (x2 ) as
polynomials of degree 2 in x2 and the degrees of the
rest polynomials keep unchanged, it is obtained with
the parameter-dependent SOS-based stability conditions in
Theorem 1 that

0.2264 × 10−1 0.4598 × 10−3

P1 =
0.4598 × 10−3 0.2361 × 10−2

Fig. 4 System responses of x(t) and control signal u(t) for the
PDPFMB control system with x= [10 10] corresponding to the
polynomial fuzzy controller with P as a polynomial of degree 0
and Njl (x2 ) as polynomials of degree 2 in x2
a Response of x1 (t)
b Response of x2 (t)
c Control signal u(t)

Fig. 3 Phase portrait of the PDPFMB control system with

various initial conditions indicated by ‘◦’
Polynomial fuzzy controller is with P as a polynomial of degree 0 and
Njl (x2 ) as polynomials of degree 2 in x2

IET Control Theory Appl., 2012, Vol. 6, Iss. 11, pp. 1645–1653 1651
doi: 10.1049/iet-cta.2011.0310 © The Institution of Engineering and Technology 2012
www.ietdl.org
−0.4139 × 10−2 0.1537 × 10−3 non-linear plant, which is controlled by a parameter-
P2 =
0.1537 × 10−3 0.3732 × 10−5 dependent polynomial fuzzy controller. The stability of
PDPFMB control systems has been investigated based on
−0.1122 × 10−4 0.1520 × 10−7 the parameter-dependent Lyapunov function candidate. With
P3 =
0.1520 × 10−7 −0.3196 × 10−8 the consideration of the information of system parameters
and membership functions, parameter-dependent SOS-based
Njl (x2 ) = [Njl(1) (x2 ) Njl(2) (x2 )]
stability conditions have been obtained to determine the
system stability and synthesise the polynomial fuzzy
for k, l = 1, 2, 3, where
controller. A feasible solution to the parameter-dependent
SOS-based stability conditions can be found numerically
(1)
N11 (x2 ) = −1.2660 + 0.7010x2 × 10−2 − 0.1254x22
by using the third-party Matlab toolbox SOSTOOLS. A
(2)
N11 (x2 ) = 0.1333 × 10−2 − 0.1687x2 × 10−4 simulation example has been given to demonstrate the merits
of the proposed approach.
− 0.3903x22 × 10−4
(1)
N12 (x2 ) = 0.1629 × 10−1 + 0.3565x2 × 10−2
6 Acknowledgment
− 0.3884x22 × 10−4
(2)
N12 (x2 ) = 0.1117 × 10−2 − 0.1207x2 × 10−3 The work described in this paper was supported by King’s
College London.
− 0.2868x22 × 10−4
(1)
N13 (x2 ) = 0.7236 × 10−3 − 0.8948x2 × 10−2
7 References
− 0.2836x22 × 10−4 1 Takagi, T., Sugeno, M.: ‘Fuzzy identification of systems and its
−3 −4 applications to modelling and control’, IEEE Trans. Syst. Man.
(2)
N13 (x2 ) = 0.1752 × 10 + 0.4057x2 × 10 Cybern., 1985, smc-15, (1), pp. 116–132
+ 0.1773x22 × 10−5 2 Sugeno, M., Kang, G.T.: ‘Structure identification of fuzzy model’,
Fuzzy Sets Syst., 1988, 28, (1), pp. 15–33
(1)
N21 (x2 ) = −1.2255 + 0.8820x2 × 10−2 − 0.1583x22 3 Feng, G., ‘A survey on analysis and design of model-based
fuzzy control systems’, IEEE Trans. Fuzzy Syst., 2006, 14, (5),
(2)
N21 (x2 ) = 0.9550 × 10−3 − 0.9775x2 × 10−5 pp. 676–697
4 Wang, H.O., Tanaka, K., Griffin, M.F.: ‘An approach to fuzzy control
− 0.4165x22 × 10−4 of nonlinear systems: stability and design issues’, IEEE Trans. Fuzzy
Syst., 1996, 4, (1), pp. 14–23
(1)
N22 (x2 ) = 0.1201 × 10−1 + 0.3588x2 × 10−2 5 Tanaka, K., Ikeda, T., Wang, H.O.: ‘Fuzzy regulators and fuzzy
−4
observers: relaxed stability conditions and LMI-based designs’, IEEE
− 0.2836x22 × 10 Trans. Fuzzy Syst., 1998, 6, (2), pp. 250–265
6 Boyd, S.P.: ‘Linear matrix inequalities in system and control
(2)
N22 (x2 ) = −0.6178 × 10−3 + 0.1482x2 × 10−3 theory’, Society for Industrial and Applied Mathematics (SIAM),
1994
+ 0.2063x22 × 10−4 7 Kim, E., Lee, H.: ‘New approaches to relaxed quadratic stability
−2 −2 condition of fuzzy control systems’, IEEE Trans. Fuzzy Syst., 2000,
(1)
N23 (x2 ) = 0.8079 × 10 − 0.8871x2 × 10 8, (5), pp. 523–534
− 0.4174x22 × 10−4 8 Teixeira, M.C.M., Assuncao, E., Avellar, R.G.: ‘On relaxed LMI-based
designs for fuzzy regulators and fuzzy observers’, IEEE Trans. Fuzzy
(2)
N23 (x2 ) = −0.1428 × 10−3 − 0.2644x2 × 10−5 9
Syst., 2003, 11, (5), pp. 613–623
Liu, X., Zhang, Q.: ‘New approaches to H∞ controller designs based
+ 0.6176x22 × 10−5 on fuzzy observers for Takagi–Sugeno fuzzy systems via LMI’,
Automatica, 2003, 39, (9), pp. 1571–1582
10 Liu, X., Zhang, Q.: ‘Approaches to quadratic stability conditions and
Fig. 3 shows the phase portrait of the PDPFMB control H∞ control designs for Takagi–Sugeno fuzzy systems’, IEEE Trans.
system subject to various initial conditions. Fig. 4 shows Fuzzy Syst., 2003, 11, (6), pp. 830–839
the system responses and control signal of the PDPFMB 11 Fang, C.H., Liu, Y.S., Kau, S.W., Hong, L., Lee, C.H.: ‘A new
LMI-based approach to relaxed quadratic stabilization of Takagi–
control system with x(0) = [10 10]. It can be seen from Sugeno fuzzy control systems’, IEEE Trans. Fuzzy Syst., 2006, 14,
Figs. 1–4 that both polynomial fuzzy controllers are able to (3), pp. 386–397
stabilise the non-linear plant and offer more or less the same 12 Sala, A., Ariño, C.: ‘Asymptotically necessary and sufficient
system performance. conditions for stability and performance in fuzzy control: applications
To show the effectiveness of the proposed stability of Polya’s theorem’, Fuzzy Sets Syst., 2007, 158, (24), pp. 2671–2686
13 Lam, H.K., Leung, F.H.F.: ‘Stability analysis of fuzzy control systems
analysis approach with the consideration of the information subject to uncertain grades of membership’, IEEE Trans. Syst. Man
of system parameters and membership functions, and Cybern. B Cybern., 2005, 35, (6), pp. 1322–1325
treating them as symbolic variables, we compared ours 14 Ariño, C., Sala, A.: ‘Extensions to stability analysis of fuzzy control
with [40]. The SOS-based stability conditions, which are systems subject to uncertain grades of membership’, IEEE Trans. Syst.
Man Cybern. Cybern., 2008, 38, (2), pp. 558–563
obtained based on the stability analysis technique in [40] as 15 Lam, H.K., Narimani, M.: ‘Stability analysis and performance
shown in Remark 4, are employed to check the stability design for fuzzy-model-based control system under imperfect premise
of the PDPFMB control system. However, using the matching’, IEEE Trans. Fuzzy Syst., 2009, 17, (4), pp. 949–961
SOSTOOLS, no feasible solution can be found. 16 Sala, A., Ariño, C.: ‘Relaxed stability and performance conditions
for Takagi–Sugeno fuzzy systems with knowledge on membership
function overlap’, IEEE Trans. Syst. Man Cybern. B Cybern., 2007,
37, (3), pp. 727–732
5 Conclusion 17 Sala, A., Ariño, C.: ‘Relaxed stability and performance LMI conditions
for Takagi–Sugeno fuzzy systems with polynomial constraints on
A parameter-dependent polynomial fuzzy model has been membership function shapes’, IEEE Trans. Fuzzy Syst., 2008, 16, (5),
proposed to represent the dynamics of parameter-dependent pp. 1328–1336

1652 IET Control Theory Appl., 2012, Vol. 6, Iss. 11, pp. 1645–1653
© The Institution of Engineering and Technology 2012 doi: 10.1049/iet-cta.2011.0310
 www.ietdl.org
18 Kruszewski, A., Sala, A., Guerra, T., Arino, C.: ‘A triangulation 32 Lam, H.K., Leung, F.H.F., Lee, Y.S.: ‘Design of a switching controller
approach to asymptotically exact conditions for fuzzy summations’, for nonlinear systems with unknown parameters based on a fuzzy logic
IEEE Trans. Fuzzy Syst., 2009, 17, (5), pp. 985–994 approach’, IEEE Trans. Syst. Man Cybern. B Cybern., 2004, 34, (2),
19 Narimani, M., Lam, H.K.: ‘Relaxed LMI-based stability conditions pp. 1068–1074
for Takagi–Sugeno fuzzy control systems using regional-membership- 33 Lam, H.K., Leung, F.H.F.: ‘Fuzzy rule-based combination of linear
function-shape-dependent analysis approach’, IEEE Trans. Fuzzy Syst., and switching state-feedback controllers’, Fuzzy Sets Syst., 2005, 156,
2009, 17, (5), pp. 1221–1228 (2), pp. 153–184
20 Narimani, M., Lam, H.K., Dilmaghani, R., Wolfe, C.: ‘LMI- 34 Lam, H.K., Leung, F.H.F.: ‘Fuzzy combination of fuzzy and switching
based stability analysis of fuzzy-model-based control systems using state-feedback controllers for nonlinear systems subject to parameter
approximated polynomial membership functions’, IEEE Trans. Syst. uncertainties’, IEEE Trans. Syst. Man Cybern. B Cybern., 2005, 35,
Man Cybern. B Cybern., 2011, 41, (3), pp. 713–724 (2), pp. 269–281
21 Lam, H.K., Narimani, M.: ‘Quadratic stability analysis of fuzzy-model- 35 Ohtake, H., Tanaka, K., Wang, H.O.: ‘Switching fuzzy controller
based control systems using staircase membership functions’, IEEE design based on switching lyapunov function for a class of nonlinear
Trans. Fuzzy Syst., 2010, 18, (1), pp. 125–137 systems’, IEEE Trans. Syst. Man Cybern. B Cybern., 2006, 36, (1),
22 Lam, H.K.: ‘Polynomial fuzzy-model-based control systems: stability pp. 13–23
analysis via piecewise-linear membership functions’, IEEE Trans. 36 Lam, H.K.: ‘Stability analysis of sampled-data fuzzy controller for
Fuzzy Syst., 2011, 19, (3), pp. 588–593 nonlinear systems based on switching T-S fuzzy model’, Nonlinear
23 Lam, H.K.: ‘LMI-based stability analysis for fuzzy-model-based Anal. Hybrid Syst., 2009, 3, (4), pp. 418–432
control systems using artificial T–S fuzzy model’, IEEE Trans. Fuzzy 37 Lo, J.C., Kuo, Y.H.: ‘Decoupled fuzzy sliding-mode control’, IEEE
Syst., 2011, 19, (3), pp. 505–513 Trans. Fuzzy Syst., 1998, 6, (3), pp. 426–435
24 Lam, H.K., Leung, F.H.F., Tam, P.K.S.: ‘A linear matrix inequality 38 Lin, F.J., Wang, D.H., Huang, P.K.: ‘FPGA-based fuzzy sliding-mode
approach for the control of uncertain fuzzy systems’, IEEE Control control for a linear induction motor drive’, IEE Proc. Electr. Power
Syst. Mag., 2002, 22, (4), pp. 20–25 Appl., 2005, 152, (5), pp. 1137–1148
25 Leung, F.H.F., Lam, H.K., Ling, S.H., Tam, P.K.S.: ‘Optimal and 39 Yagiz, N., Hacioglu, Y., Taskin, Y.: ‘Fuzzy sliding-mode control
stable fuzzy controllers for nonlinear systems based on an improved of active suspensions’, IEEE Trans. Ind. Electron., 2008, 55, (11),
genetic algorithm’, IEEE Trans. Ind. Electron., 2004, 51, (1), p. 38833890
pp. 172–182 40 Tanaka, K., Yoshida, H., Ohtake, H., Wang, H.O.: ‘A sum of squares
26 Leung, F.H.F., Lam, H.K., Tam, P.K.S.: ‘Design of fuzzy controllers approach to modeling and control of nonlinear dynamical systems with
for uncertain nonlinear systems using stability and robustness polynomial fuzzy systems’, IEEE Trans. Fuzzy Syst., 2009, 17, (4),
analyses’, Syst. Control Lett., 1998, 35, (4), pp. 237–243 pp. 911–922
27 Labiod, S., Guerra, T.M.: ‘Adaptive fuzzy control of a class of 41 Tanaka, K., Ohtake, H., Wang, H.O.: ‘Guaranteed cost control
SISO nonaffine nonlinear systems’, Fuzzy Sets Syst., 2007, 158, (10), of polynomial fuzzy systems via a sum of squares approach’,
pp. 1126–1137 IEEE Trans. Syst. Man Cybern. B Cybern., 2009, 39, (2),
28 Li, T.S., Tong, S.C., Feng, G.: ‘A novel robust adaptive-fuzzy-tracking pp. 561–567
control for a class of nonlinear multi-input/multi-output systems’, 42 Sala, A., Ariño, C.: ‘Polynomial fuzzy models for nonlinear control:
IEEE Trans. Fuzzy Syst., 2010, 18, (1), pp. 150–160 a Taylor-series approach’, IEEE Trans. Fuzzy Syst., 2009, 17, (6),
29 Tanaka, K., Iwasaki, M., Wang, H.: ‘Switching control of an R/C pp. 284–295
hovercraft: stabilization and smooth switching’, IEEE Trans. Syst. Man 43 Prajna, S., Papachristodoulou, A., Parrilo, P.A.: ‘Nonlinear control
Cybern. B Cybern., 2001, 31, (6), pp. 853–863 synthesis by sum-of-squares optimization: a Lyapunov-based
30 Lam, H.K., Leung, F.H.F., Tam, P.K.S.: ‘Fuzzy control of a class approach’. Proc. Asian Control Conf. (ASCC), Melbourne, Australia,
of multivariable nonlinear systems subject to parameter uncertainties: February 2004, vol. 1, pp. 157–165
model reference approach’, Int. J. Approx. Reason., 2001, 26, (2), 44 Papachristodoulou, A., Prajna, S.: ‘A tutorial on sum of squares
pp. 129–144 techniques for system analysis’. Proc. American Control Conf.
31 Lam, H.K., Leung, F.H.F., Tam, P.K.S.: ‘A switching controller for (ASCC), Portland, OR, 2005, pp. 2686–2700
uncertain nonlinear systems’, IEEE Control Syst. Mag., 2002, 22, (1), 45 Khalil, H.K., Grizzle, J.W.: ’Nonlinear systems’ (Prentice-Hall,
pp. 7–14 Englewood Cliffs, NJ, 1996)

IET Control Theory Appl., 2012, Vol. 6, Iss. 11, pp. 1645–1653 1653
doi: 10.1049/iet-cta.2011.0310 © The Institution of Engineering and Technology 2012