Xie 2008 Multi-Objective H-Infinity Alpha-Stability Controller Synthesis of LTI Systems

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Multi-objective H1/a-stability controller synthesis of

LTI systems
W. Xie

Abstract: A kind of new linear matrix inequality (LMI)-based formulation is presented for multi-
objective H1/a-stability controller synthesis of linear time invariant (LTI) systems. New
LMI-based conditions enable to parameterise controllers without involving the Lyapunov variables
in the formulation. With this feature, multi-objective controllers can be designed, with non-
common Lyapunov variables and two additional adjustable scalars. Furthermore, to find the
minimum possible cost, LMI-based conditions can be solved by the grid of the space which is com-
posed of these scalars. When compared with previous results based on a common Lyapunov func-
tion, this provides less conservative results. Finally, two numerical examples are included to
illustrate the effectiveness of the proposals.

1 Introduction cannot extend the results from H2 performance synthesis


to H1 performance. To the author’s best knowledge, multi-
H1 performance is interpreted as a criterion to evaluate dis- objective H1/a-stability synthesis problem has not been
turbance rejection performance or robust stability and so on. explored yet with non-common Lyapunov variables.
In contrast, pole placement constraints are often taken into In this article, a kind of new LMI-based formulation is
account in the design of control systems, such as a-stability provided for multi-objective H1/a-stability controller syn-
region, circular region and so on. Thus, it is very important thesis of LTI systems. First, by introducing a slack matrix
to study the multi-objective H1/pole placement controller and two additional adjustable scalars, new LMI-based con-
synthesis problem. There are some literatures concerning ditions are presented for H1 performance and a-stability
multi-objective control problems [1– 6]. Especially, in [3], analysis, respectively. In these conditions, the Lyapunov
an efficient method has been proposed for multi-objective matrix variable has no multiplication relations with the
H1/pole placement problems with a common Lyapunov state-space matrix of the controlled plant. Then, using this
function in terms of linear matrix inequalities technology, feature, multi-objective H1/a-stability controllers with
which is a numerically tractable way. However, owing to non-common Lyapunov variables and two additional adjus-
the common Lyapunov matrix variable, this still guarantees table scalars are designed. Furthermore, to find the
the desired specifications at the expense of conservatism. minimum possible cost, the LMI-based conditions can be
Oliveira et al. [7] opened a kind of horizon for LMI-based solved by the grid of the space which is composed of
controller synthesis under the discrete-time setting. Just these scalars. Consequently, it can reduce the conservatism
introduced in [5, 7, 8], the new horizon has a dilated LMI that occurred before in mixed H1/a-stability controller
expression, which has two useful properties. That is, the synthesis problem with a fixed Lyapunov function.
Lyapunov variable has no multiplication relations with Finally, two numerical examples are included to illustrate
the state-space matrix of the controlled plant. Namely, the the effectiveness of the proposals.
new controller parameterisation is independent of the The following notations are used. For a matrix A [ <nn,
Lyapunov variables. By virtue of this new parameterisation, d(A) is the set of the eigenvalues of A. Z denotes the
multi-objective controller synthesis with non-common complex plane.
Lyapunov variables and robust controller synthesis with
parameter-dependent Lyapunov variables have been
established, both are promising for less conservative 2 Preliminary
results [7, 8]. In contrast, the dilated LMIs can reduce
easily to the original one, as this ensures that Consider the linear time-invariant system as
dilated-LMI-based synthesis approaches always encompass
the corresponding original LMI-based ones. In the x_ (t) ¼ Ax(t) þ Bu(t)
continuous-time system case [6, 9 – 11], Ebihara and y(t) ¼ Cx(t) þ Du(t) (1)
Hagiwara presented a dilated LMI formulation for H2 and
D-stability synthesis problems; however, the method where the state vector x(t) [ <n and all other vectors and
matrices have appropriate dimensions. Equation (1) also
# The Institution of Engineering and Technology 2008 has the transfer function formulation as T (s) ¼
doi:10.1049/iet-cta:20060501 C(SI  A)1 B þ D:
Paper first received 28th November 2006 and in revised form 15th April 2007
The author is with the College of Automation Science and Technology, South
Definition 1: a-stability region [3]: Let a matrix A [ <nn
China University of Technology, Guangzhou 510641, People’s Republic of be a matrix. a-stability region denotes that matrix A-
China satisfies d(A) , n(a), where n(a) :¼ {l [ Z:Re(l) ,
E-mail: weixie@scut.edu.cn a} (a . 0):
IET Control Theory Appl., 2008, 2, (1), pp. 51 –55 51

thorized licensed use limited to: CENTRO FED DE EDUCACAO TECNOLOGICA DE MINAS GERAIS. Downloaded on September 23,2020 at 18:47:48 UTC from IEEE Xplore. Restrictions app
Definition 2: H1 performance [4]: Let kT k1 denote the H1 region constraint problems such as cones and circles
norm of T(s), that is, its largest gain across frequency in the region and so on.
singular value norm. The H1 norm measures the system
input – output gain for finite energy signals. The constraint Theorem 2: (H1 performance): Let us consider the
kT k1 , g can be interpreted as a criterion to evaluate the continuous-time system (1), then the following three con-
disturbance rejection performance or robust stability and ditions are equivalent.
so on.
1. Matrix A is stable and the H1 performance is bounded
3 Multi-objective controller synthesis using new by g1 . 0. Namely, kTk1 , g1 .
LMI formulation 2. There exists a symmetric positive-definite matrix P1 to
satisfy
In this section, after new LMI-based consitions for
2 3
a-stability region and H1 performance are derived, control- AP1 þ P1 AT P1 C T B
ler parameterisations will be shown for new LMI-based 6 7
controller synthesis. These new parameterisations enable 4 CP1 g1 I D 5,0 (4)
to design multi-objective H1/a-stability controllers with BT DT g 1 I
non-common Lyapunov variables and two additional adjus-
table scalars. It also turns out that this new LMI-based
method always encompasses the corresponding Lyapunov 3. There exist a symmetric positive-definite matrix P1 and
shaping paradigm [4] as a special case. general matrix F satisfying
2 3
3.1 New LMIs formulation for a-stability and H1 AF þ F T AT P1  F T þ r1 AF F T C T B
performance 6 P  F þ r F T AT T
r1 (F þ F ) r1 F C T T
0 7
6 1 1 7
6 7
4 CF r1 CF g 1 I D 5
In this section, two theorems show new LMI-based con-
ditions for a-stability region and H1 performance. BT 0 DT g1 I
,0 ð5Þ
Theorem 1: (a-stability region): Let A [ <nn be a Matrix.
Then, the following three conditions are equivalent. for a sufficiently small scalar r1 . 0:

1. Matrix A satisfies d(A) , n(a), where Moreover, every solution P1 of (4) is also a solution of (5).
n(a) :¼ {l [ Z:Re(l) , a} (a . 0): Conversely, every solution P1 of (5) is also a solution of (4).
2. There exists a symmetric positive-definite matrix Pa to
satisfy Proof: The equivalence of 1. and 2. is well known in
[12 – 14]. The proof of the equivalence between 2. and 3.
APa þ Pa AT þ 2aPa , 0 (2) is given.
When a symmetric positive-definite matrix P1 satisfying
(4) exists, a positive scalar r1 . 0 as r1 , 2l1 =l2 can be
3. There exist a symmetric positive-definite matrix Pa and found where
general matrix F satisfying 0 0 11
" # AP1 þ P1 AT P1 C T B
AF þ F T AT þ 2aPa Pa  F T þ ra AF B B
l1 ¼ lmin @@ CP1 g 1 I D AA
CC
,0 (3)
Pa  F þ ra F T AT ra (F þ F T ) BT DT g1 I
00 11
for a sufficiently small scalar ra . 0: AP1 AT AP1 C 0T
BB CC
Moreover, every solution Pa of (2) is also a solution of and l2 ¼ lmax @@ CP1 AT CP1 C T 0 AA
(3). Conversely, every solution Pa of (3) is also a solution 0 0 0
of (2).
Then, applying Schur’s complement with respect to (5)
Proof: The equivalence of 1. and 2. is well known in [3]. by choosing F ¼ P1 yields
The proof of the equivalence between 2. and 3. is given.
When a symmetric positive-definite matrix Pa satisfying (2) 0 1
AP1 þ P1 AT P1 C T B
exists, a positive scalar ra . 0 as ra , 2l1 =l2 , where l1 ¼ B C r1
lmin ((APa þ Pa AT þ 2aPa )) and l2 ¼ lmax (APa AT ): @ CP1 g 1 I D Aþ
2
Then, applying Schur’s complement with respect to (3) BT DT g 1 I
by choosing F ¼ Pa yields 0 1
AP1 AT AP1 C T 0
r   B C
(APa þ Pa AT þ 2aPa ) þ a APa AT , 0  @ CP1 AT CP1 C T 0 A , 0 (6)
2
0 0 0
Scalar ra makes (3) always satisfy.
When a positive symmetric matrix Pa , matrix F and a posi- Scalar r1 makes (5) always satisfy.
tive scalar ra . 0 satisfying (3) exist, (3) is multiplied with When a positive symmetric matrix P1 , matrix F and a
T ¼ [ I A ] on the left and T T on the right and (2) can be
2 r1 . 0 satisfying
positive scalar 3 (5) exist, (5) is multiplied
obtained directly. A I A 0 0
Remark 1: The results concerning a-stability region of the with T ¼ 4 0 C I 0 5 on the left and T T on the right,
LMI-based pole placement formulation are discussed. The 0 0 0 I
method proposed here can be easily extended to other we can get (4) directly. A
52 IET Control Theory Appl., Vol. 2, No. 1, January 2008

thorized licensed use limited to: CENTRO FED DE EDUCACAO TECNOLOGICA DE MINAS GERAIS. Downloaded on September 23,2020 at 18:47:48 UTC from IEEE Xplore. Restrictions app
3.2 Output – feedback synthesis with constraints only involve the following terms
change-of-variables " #
T
AX þ Bu L A
In this section, output-feedback synthesis is considered for q Acl Fa ¼
Q YA þ GCy
linear continuous-time systems with the change-of-variables
method. " #
Bw
Let us consider the continuous-time LTI plant described T
q Bcl ¼
by YBw þ GDyw ð14Þ
 
Ccl F q ¼ Cz X þ Dzu L Cz
x_ ¼ Ax þ Bw w þ Bu u
 
z ¼ Cz x þ Dzu u (7) X I
qT F q ¼
y ¼ Cy x þ Dyw w S Y

where x(t) [ <n is the system state vector, w(t) [ <q the The above terms are affine with respect to V, X, Y, Q, G, L
exogenous disturbance signal, z(t) [ <m the objective func- and U. Thus, the matrix inequalities (3) and (5) turn out to
tion signal including state combination and u [ <r the LMIs with respect to the variables V, X, Y, Q, G, L, U and
control input. g1 . Once the variables V, X, Y, Q, G, L and U satisfying
A full-order output-feedback controller is of the form the constraint conditions have been found, the output-
feedback controller matrix will be given as
x_ k ¼ Ac xk þ Bc y
  " #
u ¼ C c xk (8) Ac Bc V 1 V 1 YBu
K¼ :¼
Cc 0 0 I
where xk [ <nk is the controller state. A gain matrix con-
taining all the unknown controller parameters is defined as   " #
Q  YAX G U 1 0
  (15)
Ac Bc L 0 Cy XU 1 I
K :¼ (9)
Cc 0
where U and V are non-singular matrices and
Thus, the closed-loop state matrices are obtained as S ¼ YX þ UV .
" # " #
A Bu Cc Bw
Acl ¼ , Bcl ¼ 3.3 Multi-objective H1/a-stability controller
Bc Cy Ac Bc Dyw ð10Þ synthesis via new LMI formulation
 
Ccl ¼ Cz Dzu Cc , Dcl ¼ 0
New LMI-based conditions are applied to the multi-
These matrices Acl , Bcl , C and Dcl are substituted into objective H1/a-stability controller synthesis. For plant
matrix inequality-based conditions (3) and (5). (7), the problem is to find an output-feedback controller,
How to linearise the bilinear matrix inequality (3) or (5) which minimises g1 subject to the a-stability constraint.
is shown in the rest of this section. With a suitable nonlinear Tracking this problem, the standard and new LMIs lead to
transformation (change-of-variables) in [4, 5, 13– 17], the two different approaches.
above inequality (3) or (5) will be transformed into a
linear matrix inequality for dynamic output-feedback syn- 1. Standard LMI-based approach (4): Minimise g1 subject
thesis problem. to (2) by a common Lyapunov variable P ¼ Pa ¼ P1 .
Let us partition the matrices F and F 1 as 2. New LMIs-based approach (5): Minimise g1 subject to
   T  (3) by a common multiplier F with different Lyapunov vari-
X ? 1 Y ? ables Pa and P1 and two adjustable scalars r1 and ra .
F :¼ and F :¼ (11)
U ? VT ?
With approach 1. a common Lyapunov matrix variable is
where matrices X [ <nn and Y [ <nn , and ‘?’ denotes enforced to convexify the synthesis problem via the
block in these matrices with no importance for the deri- Lyapunov shaping paradigm [3, 4]. The restriction inher-
vations to be presented in the sequel. ently brings conservatism into the design. With approach
In addition, a transformation matrix has been introduced 2., a common multiplier is to used construct a controller
  according to different Lyapunov matrix variables for each
I YT control performance. When compared with approach 1.,
q :¼ (12)
0 VT this method provides less conservative results as non-
common Lyapunov variables and two different adjustable
which has already been used in [4, 5]. Then, the nonlinear parameters r1 and ra are used for each design specification.
transformation is shown as
    " #   Theorem 3: For multi-objective H1/a-stability control per-
Q G V YBu U 0 Y   formance synthesis problem, suppose that the standard
:¼ K þ A X 0
L 0 0 I Cy X I 0 LMI-based approach with a common Lyapunov variable
  achieves an upper bound g1C of the cost functional. Then,
r J the new LMI-based approach with a common multiplier
: ¼ qT Pcl q and two adjustable positive scalars, but with non-common
JT H
Lyapunov variables, always achieves an upper bound g1G
S : ¼ YX þ VU (13) that is lower than or equal g1C . Namely, g1G  g1C .
Then, appropriate congruence transformations with Proof: The assertion follows immediately from Theorems 1
matrix q are applied to (3) and (5) so that the resulting and 2. A
IET Control Theory Appl., Vol. 2, No. 1, January 2008 53

thorized licensed use limited to: CENTRO FED DE EDUCACAO TECNOLOGICA DE MINAS GERAIS. Downloaded on September 23,2020 at 18:47:48 UTC from IEEE Xplore. Restrictions app
Remark 2: As to the multi-objective synthesis based on LMI If the existence of solutions is affirmative, the multi-
conditions (3) and (5), the choosing of these adjustable objective robust state feedback controller matrix guarantee-
scalars r1 and ra is very important to converge to a pre- ing the minimum value of H1 performance can be given as
scribed feasible value. To obtain the minimum possible K ¼ MF 1 .
g1 , solving 2. by iterating over these two adjustable par-
ameters r1 or ra , can be considered. That is, LMI-based
method 2. can be solved with the grid of the space which 4 Numerical example
is composed of scalars r1 and ra to find the minimum poss-
ible g1 . Although computational complexity is increased, In this section, the approaches described earlier are illus-
less conservative results can be obtained. trated by some numerical examples. All LMI-related com-
putations were performed with the LMI Toolbox of
Remark 3: Compared with the existing results in the Matlab [18].
continuous-time setting [8], just introduced in Section 3.4
of [6], the H1 controller synthesis cannot be handled Example 1: Consider the LTI plant described as
with the multiplier approach. However, using new slack
variable directly, the H infinity synthesis problem can be 0 1 0 1 0 1
0 10 2 1 0
solved in this article. Furthermore, the proposed method B C B C B C
can also be easily extended to H2 synthesis problem. x_ ¼ @ 1 1 0 Ax þ @ 0 Aw þ @ 1 Au
Thus, the approach proposed here is beneficial to multi- 0 2 5 1 0
objective synthesis problem in the continuous-time setting.  
z1 ¼ 1 0 0 xþu
 
Remark 4: Compared with the results in the discrete-time y¼ 1 0 0 x þ 2w
setting, the usefulness of the proposed method has been
shown to design multi-objective H1/a-stability controller where u and y denote the control and measurement
synthesis of continuous-time systems. The related results respectively.
can also be extended easily to other multi-objective syn- The goal is to design a full-order output-feedback con-
thesis problems. In addition, the results in Theorem 3 can troller K that minimises kTz1 w k1 , g subject to a-stability
be viewed as partial counterparts of those obtained in the region, where a ¼ 2.5. The following results are obtained
discrete-time setting [5]. by solving this problem using the two approaches.
Remark 5: As to robust control problem of LTI systems a. With the standard LMI-based approach [3, 4], the
with polytopic-type uncertainties, multi-objective robust minimum H1 cost guaranteed level is found to be
state feedback synthesis can be considered by these new for- g ¼ 30.19. The standard LMI-based approach yields the
mulations. When the system state matrices in (7) are controller
assumed as
66:83s2  1482s  5815
(A(a), Bw (a), Bu (a), Cz (a), Dzu (a)) [ @ ¼: K¼
8 9 s3 þ 32:71s2 þ 360:7s þ 1112
>
>(A(a), Bw (a), Bu (a), Cz (a), Dzu (a)) >
>
>
> >
>
<¼ Pa (A , B , B , C , D ), a  0, i ¼ 1, . . . , N , >
> N b. With the proposed new LMI-based approach, (20) can be
=
i i w,i u,i z,i zu,i i solved iterating over these two adjustable parameters or r1
i¼1
>
> P >
> or ra . The relation between control performance and two
>
> N >
>
>
: ai ¼ 1 >
; parameters r1 and ra is shown in Fig. 1.
i¼1 The minimum H1 cost-guaranteed level is found to be
g ¼ 19.06. In this case r1 ¼ ra ¼ 0:15 in (3) and (5). The
multi-objective robust state feedback synthesis can be for-
proposed method yields the controller
mulated as: minimise g1 in (16) subject to (17) by a
common multiplier F with different Lyapunov variables
Pa and P1 and two adjustable scalars r1 and ra 65:15s2  1:321e9s  1:036e10

2 s3 þ 17:31s2 þ 4:294e8s þ 3:367e9
(Ai F þ Bu,i M) þ F T ATi P1  F T
6 þM T BTu,i þr1 (Ai F þ Bu,i M)
6
6
6 P  F þ r (F A þ M B )
T T T T
r1 (F þ F T )
6 1 1 i u,i
6
4 Ci F þ Dzu,i M r1 (Ci F þ Dzu,i M)
BTw,i 0
3
F T CiT þ M T DTzu,i Bw,i
r1 (F T
CiT þM T
DTzu,i ) 0 7
7
7,0 (16)
g1 I 0 5
0 g 1 I
2 3
(Ai F þ Bu,i M) þ F T ATi Pa  F T
6 7
4 þM T BTu,i þ 2aPa þra (Ai F þ Bu,i M) 5 , 0
Pa  F þ ra (F T ATi þ M T BTu,i ) ra (F þ F T )
Fig. 1 Relation between performance and two parameters r1
(17) and ra

54 IET Control Theory Appl., Vol. 2, No. 1, January 2008

thorized licensed use limited to: CENTRO FED DE EDUCACAO TECNOLOGICA DE MINAS GERAIS. Downloaded on September 23,2020 at 18:47:48 UTC from IEEE Xplore. Restrictions app
performance for linear continuous-time systems are pre-
sented. By introducing a multiplier and two additional
adjustable scalars, the new LMIs enable us to parameterise
controllers without involving the Lyapunov variables in the
formulation. Thus, the design of controller with multi-
objectives can be addressed without employing a common
Lyapuonv variable to satisfy all objectives. When compared
with a common Lyapunov function-based method, it can
provide less conservative results.

6 Acknowledgment

This work is supported by the National Natural Science


Foundation of China (Grant no. 60704022) and the
Guangdong Natural Science Foundation (Grant no.
07006470).
Fig. 2 Relation between performance and two parameters r1
and ra
7 References
Example 2: Consider the LTI plant introduced in [6].
1 Bernstein, D.S., and Haddad, W.M.: ‘LQG control with an H1
0 1 0 1 0 1 performance bound: a Riccati equation approach’, IEEE Trans.
0:32 0:04 0:01 1 0:42
Autom. Control, 1989, 34, pp. 293– 305
B C B C B C
x_ ¼ @ 0:45 0:99 0:64 Ax þ @ 0 Aw þ @ 0:74 Au 2 Khargonekar, P.P., and Rotea, M.A.: ‘Mixed H2/ H1 control:
a convex optimization approach’, IEEE Trans. Autom. Control,
0:76 0:37 0:40 1 0:31 1991, 39, pp. 824 –837
2 3 2 3 3 Chilali, M., and Gahinet, P.: ‘H1 design with pole placement
0 2 0 0 constraints: an LMI approach’, IEEE Trans. Autom. Control, 1996,
6 7 6 7 41, (3), pp. 358– 363
z1 ¼ 4 0 0 1 5 x þ 4 0 5 u
4 Scherer, C., Gahinet, P., and Chilali, M.: ‘Multiobjective
0 0 0 5 output-feedback control via LMI optimization’, IEEE Trans. Autom.
  Control, 1997, 42, (7), pp. 896 –911
y ¼ 0:72 0:85 0:88 x þ 2w 5 Oliveira, M.C., Geromel, J.C., and Bernussou, J.: ‘Extended H2 and
H1 norm characterizations and controller parameterizations for
where u and y denote the control and measurement. discrete-time systems’, Int. J. Control, 2002, 75, pp. 666 –679
Just like example 1, a full-order output-feedback control- 6 Ebihara, Y., and Hagiwara, T.: ‘New dilated LMI characterizations for
continuous-time multiobjective controller synthesis’, Automatica,
ler K is designed to minimise kTz1 w k1 , g subject to 2004, 40, pp. 2003–2009
a-stability region, where a ¼ 1. Solving this problem by 7 Oliveira, M.C., Bernussou, J., and Geromel, J.C.: ‘A new
the two approaches discussed in the section above, the fol- discrete-time robust stability condition’, Syst. Control Lett., 1999,
lowing results are obtained. 37, pp. 261– 265
8 Oliveira, M.C., and Skelton, R.E.: ‘Stability tests for constrained
1. With the standard LMI-based approach [3, 4], the linear systems’, ‘Lecture notes in control and information sciences’,
minimum H1 cost guaranteed level is found to be ‘Perspectives in robust control’ (Springer, Berlin, 2001, vol. 268)
g ¼ 692.52. The standard LMI-based approach yields the 9 Ebihara, Y., and Hagiwara, T.: ‘New dilated LMI characterizations for
controller continuous-time control design and robust multiobjective control’.
Proc. 2002 ACC, pp. 47–52
1:42e7s2  6:08e11s þ 5:05e14 10 Ebihara, Y., and Hagiwara, T.: ‘Robust controller synthesis with
K¼ parameter-dependent Lyapunov variables: a dilated LMI approach’.
s3 þ 2:251e6s2  2:399e13s  2:97e13 Proc. 41st CDC, 2002, pp. 4179–4184
11 Gahinet, P., and Apkarian, P.: ‘A linear matrix inequality approach to
2. With the proposed new LMI-based approach, 2. can be H1 control’, Int. J. Robust Nonlinear Control, 1994, 4, pp. 421 –448
solved by iterating over these two adjustable parameters 12 Masubuchi, I., Ohara, A., and Suda, N.: ‘LMI-based controller
r1 or ra , respectively. The relation between control per- synthesis: a unified formulation and solution’, Int. J. Robust
formance and two parameters r1 and ra is shown in Fig. 2. Nonlinear Control, 1998, 8, pp. 669– 686
13 Apkarian, P., Tuan, H.D., and Bernussou, J.: ‘Continuous-time
The minimum H1 cost-guaranteed level is g ¼ 392.53. analysis and H2 multi-channel synthesis with enhanced LMI
In this case, r1 ¼ 0:25 and ra ¼ 0:2 in (3) and (5), respect- characterizations’, IEEE Trans. Autom. Control, 2001, 46, (12),
ively. The proposed method yields the controller pp. 1941–1946
14 Boyd, S.P., Ghaoui, L.E., Feron, E., and Balakrishnan, V.: ‘Linear
1176s2  882:7s  532 matrix inequalities in system and control theory’ (SIAM,
K¼ Philadelphia, PA, 1994)
s3 þ 11:42s2 þ 1472s  617:7 15 Iwasaki, T., and Skelton, R.E.: ‘All controllers for the general H1
control problem: LMI existence conditions and state space
From the above results, it is obvious that the proposed formulas’, Automatica, 1994, 30, (8), pp. 1307– 1317
method could provide a less conservative result than that 16 Shaked, U.: ‘Improved LMI representations for the analysis and the
provided by thestandard LMI-based approach. design of continuous-time systems with polytopic type uncertainty’,
IEEE Trans. Autom Control, 2001, 46, (4), pp. 52– 656
17 Skelton, R.E., Iwasaki, T., and Grigoriadis, K.: ‘A unified algebraic
5 Conclusions approach to linear control design’ (Taylor & Francis, London, 1997)
18 Gahinet, P., Nemirovski, A., Laub, A.J., and Chilali, M.: ‘LMI control
In this article, new results on new LMI-based controller syn- toolbox for use with matlab, user’s guide’ (The Math Works Inc.,
thesis of multi-objective H1/a-stability control Natick, MA, USA, 1995)

IET Control Theory Appl., Vol. 2, No. 1, January 2008 55

thorized licensed use limited to: CENTRO FED DE EDUCACAO TECNOLOGICA DE MINAS GERAIS. Downloaded on September 23,2020 at 18:47:48 UTC from IEEE Xplore. Restrictions app

You might also like