Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives SDH-04-06-2012 Observer based stabilization of linear system under sparse measurement J-P Barbot J-P Barbot Observer based stabilization under sparse measurement
41
Embed
New SDH-04-06-2012 *0.5cmObserver based stabilization of linear … · 2012. 6. 14. · Introduction Stability results Observerbased control Simulation results Strange attractors
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
SDH-04-06-2012
Observer based stabilization of linear system under sparse
measurement
J-P Barbot
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Road Map
◮ Introduction
◮ A Stability Results
◮ Stable unmeasured subspace
◮ Unstable unmeasured subspace
◮ Example
◮ Example of base
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Questions
(a)
(b)
Figure : (a) The traditional paradigm to capture information ; (b) TheCS paradigm to capture information, where Φ is a random sensing matrix.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Questions
How use Compressive Sensing in Control System theory ?This question generates several important questions :
◮ a- How to pass from signal to system ?
◮ b- What is the appropriate base for system ?
◮ c- How to verify the RIP condition for system ?
◮ d- How to bypass the optimization algorithm in order to dealwith real time algorithms ?
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Answers
Answer to question a :
◮ Remember Kalman-Bucy filter
◮ Signal is generated by a system
◮ Cluster, Bayesian approach,... can be see as the first step of amodelization
Answer to question b :
◮ Bases in CS are Frequency, Wavelet, Curvelet,...
◮ Oscillations are generated by system
◮ Some normal forms may be use as base ?
◮ Some other assumption than s < m << N are generally donein CS (approach signal processing). In system theory forexample diagnostic, only few systems are selected.
An example of base will be given at the end of this presentation.J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Answers
Answer to question c :In Compressive Sensing
y = Φv
with y ∈ Rm and v ∈ Rn. The RIP condition is : There existsδ ∈]0, 1[ such that
(1− δ)||v ||22 < ||Φv ||22 < (1 + δ)||v ||22
With respect to linear system undersampling this condition isclosed to a random sampling in order to preserve the systemobservability.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Answers
Answer to question d :In this presentation an answer to question d is given, but adding astability study of the observer based control.In the next, it is firstly recalled and introduced some stabilityresults before to present our main results.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
References
• Some references in Signal Processing
◮ E. Candes, The restricted isometry property and itsimplications for compressed sensing, Compte Rendus del’Academie des Sciences, Series I, vol. 346, pp. 910, 2008.
◮ E. Candes and M. B. Walkin, An introduction to compressivesampling, IEEE Signal Processing Magazine, vol. 21, no. 2,pp. 2130, 2008.
◮ D. Donoho, I. Drori, V. Stoden, Y. Tsaig, and M. Shahram,Sparselab, http ://sparselab.stanford.edu.
◮ J. Starck and J. L. Bodin, Astronomical data analysis andsparcity : from wavelelets to compressed sensing, Processingof the IEEE, vol. 98, pp. 10211030, 2010.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
References
• Some references in Signal Processing
◮ A. Woiselle, J. Starck, and M. Fadili, 3d data denoising andinpainting with the fast curvelet transform, J. of MathematicalImaging and Vision (JMIV), vol. 39, no. 2, pp. 121139, 2011.
◮ L. Berec, A multi-model method to fault detection anddiagnosis : Bayesian solution. an introductory treatise,International Journal of Adaptive Control and SignalProcessing, vol. 12, p. 8192, 1998.
◮ L. Yu, H. Sun, J-P. Barbot, and G. Zeng, Bayesiancompressive sensing for cluster structured sparse signals,Signal Processing, vol. 92, no. 1, pp. 259269, 2012.
◮ ...
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
References
• Some references in Control Systems Theory
◮ B. Anderson, T. Brinsmead, F. De Bruyne, J. Hespanha, D.Liberzon, and A. Morse, Multiple model adaptive control...,George Zames Special Issue, International Journal of Robustand Non- linear Control, vol. 10, pp. 909929, 2000.
◮ S. Bhattachara and T. Basar, Sparcity based feedback design :a new paradigm in opportunistic sensing, IEEE ACC, 2011.
◮ Y. Khaled, J.-P. Barbot, D. Benmerzouk, and K. Busawon, Anew type of impulsive observer for hyper- chaotic system, inIFAC 3th Conference on analysis and control of chaoticsystems, Cancum, Mxico, June 20 th- 22th 2012 2012.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
References
• Some references in Control Systems Theory
◮ H. Nijmeijer and I. Mareels, An observer looks atsynchronization, IEEE Transactions on Circuits andSystems-1 : Fundamental theory and Applications, vol. 44, no.10, pp. 882891, 1997.
◮ W. Kang and A. J. Krener, Extended quadratic controllernormal form and dynamic state feedback linearization ofnonlinear systems, SIAM J. Control and Optimization, Vol 30,pp 1319-1337, 1992.
◮ A. Yang, M. Gastpar, R. Bajcsy, and S. Sastry, Distributedsensor perception via sparse representation, Proceddings ofIEEE, vol. 98, no. 6, pp. 10771088, 2010.
◮ ...
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Nonlinear case
Let us consider the folowing system :
x1(t) = f1(x1(t), x2(t)); t 6= tk
x2(t) = f2(x2(t)); t 6= tk
x1(t+k ) = Rx1(tk)
x2(t+k ) = x2(tk)
(1)
where x1(t) ∈ Rp, x2(t) ∈ R
n−p, f1 : Rn → R
p andf2 : R
n−p → Rn−p.
Assumption
f1 is at least locally lipschitz, where l1 and l2 are respectively theLipschitz constants with respect to x1 and x2.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Nonlinear case
Assumption
The sampling sequence tk ∈ T = {ti : i ∈ N} ⊂ R verifies thatthere exists τmax and τmin with τmax > τmin > 0 such that ∀i > 0
ti+1 ≥ ti + τmin and ti+1 ≤ ti + τmax (2)
Moreover we define :
θk , tk+1 − tk
x(t+k ) , limh→0
x(tk + h)
x(t−k ) , limh→0
x(tk − h) = x(tk)
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Nonlinear case
TheoremIf the system (1) verifies the previous assumptions and satisfies thefollowing conditions :
1. there exists a strictly positive definite functionV2 : R
n−p 7→ R+ V2 ∈ C1, with V2(0) = 0 such that ∀x2 6= 0
V2(x2(t)) < 0 and V2(0) = 0 .
2. ‖R‖1el1θk < 1 for all k > 0.
then, ∀ǫ∗ > 0, the ball 1 B2ǫ∗ is globally asymptotically stableconsidering the sequence of time t+k (i.e. the time just after thereset instants).
1. with respect to norm one
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Nonlinear case
Proposition
Assume that the conditions of the theorem hold for system (1),then for any ǫ∗ > 0 (of theorem 1) the ball Bβ is globallyasymptotically stable with β = ǫ∗
‖R‖1+ ǫ∗, if ‖R‖1 6= 0, and the
system (1) converges globally asymptotically to zero if ‖R‖1 = 0.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Linear case
Let us consider the following class of linear system :
x1(t) = A11x1(t) + A12x2(t); t 6= tk
x2(t) = A22x2(t)
x1(t+k ) = Rx1(t)
x2(t+k ) = x2(t)
(3)
Where x(t) ∈ Rn, A11 ∈ R
p×p, A12 ∈ Rp×(n−p),
A22 ∈ R(n−p)×(n−p) and R ∈ R
p×p.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Linear case
TheoremIf the system (3) verifies the following condition :
1. A22 is Hurwitz continuous
2. ‖ReA11θk‖2 < 1 ∀k > 0
Then, ∀ǫ > 0, the state of system (3) converges to a ball of radiusepsilon.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Stable unmeasured subspace
Now, we consider the following class of system :
{
x(t) = Ax(t) + Bu(t)
y(tk) = Cx(tk)(4)
Where x(t) ∈ Rn is the state, u(t) ∈ R
m is the input andy(tk) ∈ R
p is the output. The matrixes A, B and C are constantand of appropriate dimension. Moreover the system is assumeddetectable and stabilizable.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Stable unmeasured subspace
Assumption
There exist a regular matrix T which transform the system 4 intothe system 5 described below (with x = Tx).
˙x1(t) = A11x1(t) + A12x2(t) + B1u(t)
˙x2(t) = A22x2(t) + B2u(t)
y(tk) = x1(tk)
(5)
Where x(t) = (xT1 , xT2 )T with A11 ∈ Rp×p, A12 ∈ R
p×(n−p),A22 ∈ R
(n−p)×(n−p), B1 ∈ Rp×m, B2 ∈ R
(n−p)×m and A22 Hurwitzcontinuous.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Stable unmeasured subspace
For the system (5) the proposed observer base control is :a- The control
Where K ∈ Rm×n is a pole placement matrix and R ∈ R
m×n is thereset matrix.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Stable unmeasured subspace
TheoremIf the system (4) verifies the previous assumptions, then it ispossible, for considered sampling sequence (2), to find an observerbased control (6)-(7) which stabilize practically asymptotically thesystem.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Unstable unmeasured subspace
We consider again the system 4, but with :
Assumption
There exist a regular matrix T which transform the system 4 intothe system 8 described below (with x = Tx).
˙x1(t) = A11x1(t) + A12x2(t) + B1u(t)
˙x2(t) = A22x2(t) + B2u(t)
y(tk) = x1(tk)
(8)
Where x(t) = (xT1 , xT2 )T with A11 ∈ Rp×p, A12 ∈ R
p×(n−p),A22 ∈ R
(n−p)×(n−p), B1 ∈ Rp×m, B2 ∈ R
(n−p)×m and A22 isunstable.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Unstable unmeasured subspace
The observer based control for system (8) is quite different in itsobserver part than the one proposed in the previous section andhas the following form :a- Control Law :
u(t) = −Kx , (K1,K2)
(
x1x2
)
(9)
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
m×n, with Rn1 = diag{r1, .., rn1} and −1 < ri < 1, fori ∈ [1, n1]
⋂
N and R , W , M, L1 et L2 are matrixes of appropriatedimensions.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Unstable unmeasured subspace
Figure : The block diagram of state feedback with generalizedimpulsive observer
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Unstable unmeasured subspace
TheoremThe system (4), which verified previous assumptions, is practicallystabilized, under the sampling 2 by the observer based control(9)-(10), if the following conditions are verified :
1. ‖ReA11θk‖ < 1.
2. the matrixe
A22 −M 0 M
0 A11 − L1 A12
0 −L2 A22
is Hurwitz.
3. The matrix
(
A11 − B1K1 A12 − B1K2
−B2K1 A22 − B2K2
)
is Hurwitz.
4. ‖x(0)‖ < KM , ‖x(0)‖ < KM , ‖x(0)‖ < KM and ∀k > 0θk < θm(KM)
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Simulation results
Consider the following linear triangular system :
x1 = 2x1 + x2 + 3x3 + u
x2 = x2 + x3
x3 = −x3 − u
y(tk) = x1(tk)
(11)
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Simulation results
The goal is to make the system (11) stable. For this, let synthesizethe proposed observer base control (10). By choosing K1 = 48,
K2 = (30, 55), R = 0, M =
(
5 00 0
)
, L1 = 6 and L2 =
(
60
)
.
The parameters of observer control satisfy the conditions oftheorem 3, with corresponding matrix of observer is
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Simulation results
and the corresponding matrix of control is
−48 31 581 1 0
−30 −56 −48
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Simulation results
0 5 10 15 20 25 30−50
0
50
0 5 10 15 20 25 30−50
0
50
0 5 10 15 20 25 30−50
0
50
TIME(S)
x1
x1obs
x2
x2obs
x3
x3obs
Figure : controlled state
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Simulation results
Now, consider the following Chaotic system :
x1(t) = a(x2(t)− x1(t))
x2(t) = bx1(t) + cx2(t)− x1(t)x3(t) + x4(t)
x3(t) = −dx3(t) + x1(t)x2(t)
x4(t) = −kx1(t) + x4(t)
y(tk) = x2(tk)
(12)
Avec : a = 35, b = 7, c = 12, d = 3, k = 5 et(x10, x20, x30, x40) = (0.05, 0.01, 0.05, 0.5). Les exposants deLyapunov sont : λ1 = 16.472, λ2 = 1.2729, λ3 = −3 etλ4 = −39.7443. The system (12) is hyperchaotic.
J-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives
Simulation results
−15
−10
−5
0
5
10
15
−15
−10
−5
0
5
10
1510
15
20
25
30
35
40
x1
x2
x 3
Figure : Phase portraitJ-P Barbot Observer based stabilization under sparse measurement
Introduction Stability results Observer based control Simulation results Strange attractors Conclusion & Perspectives