Stabilization of a Chain of Exponential Integrators Using a Strict Lyapunov Function

Transcription

1 Stabilization of a Chain of Exponential Integrators Using a Strict Lyapunov Function Michael Malisoff Miroslav Krstic

2 Chain of Exponential Integrators { Ẋ = (Y Y )X Ẏ = (D D)Y, (X, Y ) (0, ) 2 X = pest density, Y = predator density, D = control Constant D > 0 and Y > 0

3 Chain of Exponential Integrators { Ẋ = (Y Y )X Ẏ = (D D)Y, (X, Y ) (0, ) 2 X = pest density, Y = predator density, D = control Constant D > 0 and Y > 0 D = D X r Y Y X

4 Chain of Exponential Integrators { Ẋ = (Y Y )X Ẏ = (D D)Y, (X, Y ) (0, ) 2 X = pest density, Y = predator density, D = control Constant D > 0 and Y > 0 D = D X r Y Y X ( ) X V (X, Y ) = ln + X r X r X 1 + Y D ( Y ( )) Y Y 1 ln Y

5 Chain of Exponential Integrators { Ẋ = (Y Y )X Ẏ = (D D)Y, (X, Y ) (0, ) 2 X = pest density, Y = predator density, D = control Constant D > 0 and Y > 0 ( ) X V (X, Y ) = ln + X r X r D = D X r Y Y X X 1 + Y D V = Y X r X ( Y Y 1 ( Y Y 1 ln ) 2 ( )) Y Y

6 Chain of Exponential Integrators { Ẋ = (Y Y )X Ẏ = (D D)Y, (X, Y ) (0, ) 2 X = pest density, Y = predator density, D = control Constant D > 0 and Y > 0 ( ) X V (X, Y ) = ln + X r X r D = D X r Y Y X X 1 + Y D V = Y X r X ( Y Y 1 ( Y Y 1 ln ) 2 ( )) Y Y Theorem: The closed loop system is GAS to (X r, Y ) on (0, ) 2.

7 Chain of Exponential Integrators Fig. 1: X (Left), Y (Right), D = 5Y /X (Bottom). Y = X r = 1. D = 5.

8 Chain of Exponential Integrators Fig. 2: (X(t), Y (t)) Converging to (1, 1) through Level Curves of V.

9 Importance of Strict Lyapunov Functions

10 Importance of Strict Lyapunov Functions Indirect ways to prove stability properties for nonlinear systems

11 Importance of Strict Lyapunov Functions Indirect ways to prove stability properties for nonlinear systems Strictness means V < 0 along all solutions outside equilibrium

12 Importance of Strict Lyapunov Functions Indirect ways to prove stability properties for nonlinear systems Strictness means V < 0 along all solutions outside equilibrium Useful for proving robustness, e.g., input-to-state stability (ISS)

13 Importance of Strict Lyapunov Functions Indirect ways to prove stability properties for nonlinear systems Strictness means V < 0 along all solutions outside equilibrium Useful for proving robustness, e.g., input-to-state stability (ISS) Facilitate ISSification through LgV feedback control redesigns

14 Importance of Strict Lyapunov Functions Indirect ways to prove stability properties for nonlinear systems Strictness means V < 0 along all solutions outside equilibrium Useful for proving robustness, e.g., input-to-state stability (ISS) Facilitate ISSification through LgV feedback control redesigns Strict Lyapunov functions for subsystems facilitate backstepping

15 Importance of Strict Lyapunov Functions Indirect ways to prove stability properties for nonlinear systems Strictness means V < 0 along all solutions outside equilibrium Useful for proving robustness, e.g., input-to-state stability (ISS) Facilitate ISSification through LgV feedback control redesigns Strict Lyapunov functions for subsystems facilitate backstepping Strictification turns nonstrict Lyapunov functions into strict ones

16 Importance of Strict Lyapunov Functions Indirect ways to prove stability properties for nonlinear systems Strictness means V < 0 along all solutions outside equilibrium Useful for proving robustness, e.g., input-to-state stability (ISS) Facilitate ISSification through LgV feedback control redesigns Strict Lyapunov functions for subsystems facilitate backstepping Strictification turns nonstrict Lyapunov functions into strict ones Malisoff, M., and F. Mazenc, Constructions of Strict Lyapunov Functions, Communications and Control Engineering Series, Springer-Verlag London Ltd., London, UK, 2009.

17 Strictification

18 Strictification New variables x = ln(x/x r ), y = ln(y /Y ), and u = ln(d/d )

19 Strictification New variables x = ln(x/x r ), y = ln(y /Y ), and u = ln(d/d ) { ẋ = Y (1 e y ) ẏ = D (1 e y x ), (x, y) R 2

20 Strictification New variables x = ln(x/x r ), y = ln(y /Y ), and u = ln(d/d ) { ẋ = Y (1 e y ) ẏ = D (1 e y x ), (x, y) R 2 Theorem: We can build a function R : [0, ) [1, ) such that V 3 (x, y) = V 1 (x,y) 0 R(s)ds xy, where V 1 (x, y) = x + e x 1 + Y D (e y y 1) satisfies V 3 (x, y) 1 2 V 1(x, y) for all (x, y) R 2 and V 3 (x, y) 1 { 2 R(V1 (x, y))e x Y (1 e y ) 2 + D x(1 e x ) } along all solutions (x(t), y(t)) so V 3 is a strict Lyapunov function.

21 Extensions and Applications

22 Extensions and Applications Other changes of coordinates cover { Ẋ = G(t, X, Y )X Ẏ = H(t, X, Y, D)Y, (X, Y ) (0, ) 2 for many choices of G and H.

23 Extensions and Applications Other changes of coordinates cover { Ẋ = G(t, X, Y )X Ẏ = H(t, X, Y, D)Y, (X, Y ) (0, ) 2 for many choices of G and H. Predator-prey model for lynxes and hares: { ( Ḣ = rh 1 H L = abhl c+h DL k ) ahl c+h

24 Extensions and Applications Other changes of coordinates cover { Ẋ = G(t, X, Y )X Ẏ = H(t, X, Y, D)Y, (X, Y ) (0, ) 2 for many choices of G and H. Predator-prey model for lynxes and hares: { ( Ḣ = rh 1 H k ) ahl c+h L = abhl c+h DL ISS under uncertainty, delays in gestation or in interconnection,..

25 Extensions and Applications Other changes of coordinates cover { Ẋ = G(t, X, Y )X Ẏ = H(t, X, Y, D)Y, (X, Y ) (0, ) 2 for many choices of G and H. Predator-prey model for lynxes and hares: { ( Ḣ = rh 1 H k ) ahl c+h L = abhl c+h DL ISS under uncertainty, delays in gestation or in interconnection,.. Malisoff, M., and M. Krstic, Stabilization and robustness analysis for a chain of exponential integrators using strict Lyapunov functions," Automatica, to appear.

26 Conclusions

27 Conclusions Nonstrict Lyapunov functions and LaSalle invariance are useful.

28 Conclusions Nonstrict Lyapunov functions and LaSalle invariance are useful. Strictification often helps certify robustness under delays.

29 Conclusions Nonstrict Lyapunov functions and LaSalle invariance are useful. Strictification often helps certify robustness under delays. Finding auxiliary functions in Matrosov s approach can be hard.

30 Conclusions Nonstrict Lyapunov functions and LaSalle invariance are useful. Strictification often helps certify robustness under delays. Finding auxiliary functions in Matrosov s approach can be hard. Our strict Lyapunov constructions enable feedback redesigns.

31 Conclusions Nonstrict Lyapunov functions and LaSalle invariance are useful. Strictification often helps certify robustness under delays. Finding auxiliary functions in Matrosov s approach can be hard. Our strict Lyapunov constructions enable feedback redesigns. Radial unboundedness of our V 3 helped prove robustness.

32 Conclusions Nonstrict Lyapunov functions and LaSalle invariance are useful. Strictification often helps certify robustness under delays. Finding auxiliary functions in Matrosov s approach can be hard. Our strict Lyapunov constructions enable feedback redesigns. Radial unboundedness of our V 3 helped prove robustness. Our strictification covered interesting biology inspired systems.

33 Conclusions Nonstrict Lyapunov functions and LaSalle invariance are useful. Strictification often helps certify robustness under delays. Finding auxiliary functions in Matrosov s approach can be hard. Our strict Lyapunov constructions enable feedback redesigns. Radial unboundedness of our V 3 helped prove robustness. Our strictification covered interesting biology inspired systems. Thank you for your attention!