Académique Documents
Professionnel Documents
Culture Documents
L1* = (W1* )−1 R1* L 2* = (W2* ) −1 R *2 (29) where F∞1 and F∞1 result of the partition of the state
feedback gain F∞ of the centralized controller. In this
condition, the parametrization (13) is reduced again to
5. H∞ decentralized controller design
(19). To find the unique unknowns L1 , L 2 , Ranjit et al
In this section, we consider the case of H∞ decentralized [2] propose a state space approach and an iterative
control design. As in H2 case, H∞ decentralized control procedure to solve numerically the problem. The idea is
design also requires two stages. Given centralized based on the resolution of a Riccati equation that cannot
controller (12) achieving Twz (C) = Twz ≺ γ , the be solved directly because the decentralization constraint
∞ ∞
L =diag ( L1 , L 2 ), but an algorithm can be designed to
second stage of the design process is to solve for
iteratively solve it. To avoid this difficulty, we propose as
decentralizing parameter having the propriety:
previously an LMI-based method.
Qd ≺γ (30)
∞ According to “Bound Real Lemma” [16], the transfer
In [2] is given the following theorem: function Qd is stable and Q d ≺ γ if, and only if, there
∞
exists a symmetric matrix Z > 0 such as:
Theorem 3: (Suboptimality of decentralized controllers)
A TL Z+ Z A L Z (L- ˆ L) FT
1. The set of all concurrent decentralized stabilizing
observer controllers (6) is given by Qd(p) (13) where (L- ˆ L)T Z -γ I 0 =
−2 T
 = A + γ B1B1 X ∞ , F=F∞ and L=L∞ are the
F 0 -γ I
suboptimal centralized feedback and observer gains
(34)
(12), X∞ is given by (11), F1 , F2 are the feedback A Z+ Z A − Z L C− C L Z Z L-
T T T ˆ Z L FT
gains, and L1 , L 2 are the decentralized observer gains L̂T Z - LT Z -γ I 0 ≺0
from (6), F 0 -γ I
2. If F and L are chosen such that the resulting
parameter Q*d (p) achieves: This inequality is bilinear with respect to optimization
variables because the presence of Z L term. We can make
Qd ≺γ (31)
∞ the same change of variable S = Z L used in the previous
then the controller: section to transform (34) to an LMI form. Finally, as in H2
decentralized control, we also must choose Z and S block-
Aˆ − B2 Fˆ − L1C 21 0 L1 0 diagonal form in order to obtain L block-diagonal form.
ˆ − B Fˆ − L C So we have by using a similar notation to (28):
0 A 0 L2
C∞ ( p ) ↔ (32)
2 2 22
− F1 0 0 0 L1* = ( Z1* )−1 S1* L 2* = ( Z 2* ) −1 S*2 (35)
0 − F2 0 0
6. Conclusion
is a suboptimal decentralized observer controller
In this work, we have reconsidered the methodology
satisfying
d
proposed in [2] which deals with an H2 and H∞
Twz ≺γ ■ decentralized controllers design. We have reduced on the
∞
The pre-specified γ does not have necessarily the same one hand the number of parameters to be optimized and
value in centralized and decentralized design. The cost of on the other hand, we have proposed a method more
decentralization control can be found as the difference adapted to determine these parameters. Indeed, the
optimization problem initially consists to determine the
feedback gains and observer gains of a decentralized H2 Networks and Systems, Notre Dame, Indiana, August,
and H∞ controllers via optimization of a parametrization 2002, 12 -16.
transfer function of controllers. By using an appropriate [5] L.Vandenberghe, V. Balakrishnan, Algorithms and
choose for the decentralized feedback gains, we have software tools for LMI problems in control: an overview,
proposed a method based on LMI approach to optimize in Proceedings of the 1996 IEEE International
only the observer gains. Although we have considered a Symposium on Computer-Aided Control System Design,
development in two channels, this approach is easily Dearborn, Michigan, 1996, Invited session.
adapted for a greater number of channels. In addition, the
LMIs techniques can take into account the structural [6] J.K. Shiau, J.H. Chow, Structurally Constrained H∞
constraints, namely the optimization variables must have a Suboptimal Control Design Using an Iterative Linear
block-diagonal structure which is the case in decentralized Matrix Inequality Algorithm Based on a Dual Design
control. Finally, the proposed method has the advantage Formulation, Tamkang Journal of Science and
that it can be implemented numerically very efficiently Engineering, vol.1, No 2, 1998, 133 -143.
using recently developed algorithms for solving LMIs [7] C. Yong-Yang, S. You-Xian, M. Wei-Jie, Output
(LMI solvers). feedback decentralized stabilization: ILMI approach,
Systems & Control Letters, 35, 1998, 183-194.
Appendix [8] D. Biahcgnopdrejna, J. How, Parametric robust H2
control design with generalized multipliers via LMI
The closed-loop spectrum of the system (1) controlled by synthesis, Int.J.Control, Vol.70, N°3, 1998, 481-503.
the decentralized controller (6) are given by the
eigenvalues of the state space matrix: [9] F. Claveau, P. Chevrel, M. Yagoubi, D. Knittel,
Decentralized control for winding systems: Which
A - B21F1 - B22 F2
incidence on reachable performances, 11th IEEE
ˆ
A cl = L1C21 A- B2 F- L1C21 0 Mediterranean Conference on Control and Automation,
ˆ Rhodes, Juin, 2003,T7-099.
L C
2 22 0 A- B 2 F- L C
2 22
[10] G. Zhai, M. Ikeda, Y. Fujisaki, Decentralized H∞
Performing a similarity transformation with: controller design: A matrix inequality approach using a
I 0 0 I 0 0 homotopy method, Automatica, vol. 37, 2001, 565 -572.
Tc = - I I 0 and Tc = I I 0
-1 [11] N. Chen, M. Ikeda, W. Gui, Design of Robust H∞
Control for Interconnected Systems: A Homotopy
- I 0 I I 0 I Method, International Journal of Control, Automation,
we obtain : and Systems, vol. 3, N°2, June, 2005, 143 -151.
A F - B21F1 - B22 F2 [12] K. Zhou, J.C. Doyle, K. Glover, Robust and Optimal
Tc A cl Tc-1 = Control (Prentice-Hall, Upper Saddle River, NJ, 1995).
0 AL
[13] N. Bennis, M. Haloua, H. Youlal, An H2
The spectrum is σ(Tc A cl Tc−1 ) = σ(A F ) ∪ σ(A L ) . If we decentralized control, Proceedings of European Control
let F1 = F21 and F2 = F22 , the first spectrum is stable by Conference 1999 (ECC'99), Karlsruhe, Germany, 31
propriety of the Riccati equation associated to August - 3 September, 1999.
Hamiltonian matrix H 2 . The second is also stable by [14] J. Beseler, J.H. Chow, K.D. Minto, A feedback
conception of decentralized controller. descent method for solving constrained LQG control
problems, Proc. ACC, 1992, 1014 -1048.
[15] S. Boyd, L. El Ghaoui, E. Feron, V. Balakrishnan,
Linear Matrix Inequalities in System and Control Theory,
References In Proc. Annual Allerton Conf. on Communication,
[1] J.C. Doyle, K. Glover, P. P., Khargonekar, B.A. Control and Computing, Allerton House, Monticello,
Francis, State space solution to standard H2 and H∞ Illinois, 1993, 237-246.
control problem, IEEE Transactions on Automatic [16] P. Gahinet, P. Apkarian, Linear matrix inequality
Control, 34 (8) , August, 1989, 831-847. approach to H∞ control, International Journal of Robust
[2] A.D. Ranjit, J.H. Chow, A parametrization approach to and Nonlinear Control, vol 4, 1994, 421- 448.
optimal H2 and H∞ decentralized control problems,
Automatica, 29, 1993, 457- 463.
[3] J.H. Seo, H.J. Choel, L. Sang-Hyuk, Decentralized H∞
Controller design, Automatica, Vol 35, May, 1999, 865-
876.
[4] V. Balakrishnan, Linear Matrix Inequalities in Robust
Control: A Brief Survey, Proceeding of Fifteenth
International Symposium on Mathematical Theory of