Generalized Output Regulation for a Class of Nonlinear ... - CiteSeerX

1 downloads 0 Views 229KB Size Report
Pachuca-Tulancingo, Km. 4.5 C.P. 42084, Pachuca, Hgo. MEXICO. †Institute of Information Theory and Automation. Academy of Sciences of the Czech Republic ...
Generalized Output Regulation for a Class of Nonlinear Systems via the Robust Control Approach ∗ † ∗ †   L. E. RAMOS-VELASCO , S. CELIKOVSKÝ , V. LÓPEZ-MORALES AND V. KUCERA ∗ Centro de Investigación en Tecnologías de Información y Sistemas

Universidad Autónoma del Estado de Hidalgo Carr. Pachuca-Tulancingo, Km. 4.5 C.P. 42084, Pachuca, Hgo. MEXICO †

Institute of Information Theory and Automation Academy of Sciences of the Czech Republic. P.O. Box 18, 182 08 Prague CZECH REPUBLIC

Abstract: - We address the problem of generalized output regulation for nonlinear systems in the presence of unknown parameters in the full information case. We generalize the classical output regulation problem in order to expand the class of reference or disturbance signals. Under appropriate sufcient conditions, a state feedback regulator is built for a class of nonlinear systems where the term including the unknown parameter is assumed to satisfy a matching condition. Our study of the above problem, further referred to as the so-called generalized output regulation problem, combines the approach based on the well-known notion of the regulator equation with the classical concept of the invariant distribution and on the Lyapunov theory. Key-Words: - Nonlinear systems, output regulation, center manifold theory, Lyapunov function, disturbance rejection.

1 1

Introduction

systems have been presented in [8, 10] using “full

A central problem in control theory and applica-

information" which includes the measurements of

tions is to design a control law to achieve asymptotic

exogenous signals as well as of the system state.

tracking with disturbance rejection in a nonlinear

The necessary and sufcient conditions for the exis-

system. When a class of reference inputs and dis-

tence of a local full information solution of the clas-

turbances are generated by an autonomous differen-

sical output regulation problem are given in [10, 8];

tial equation, this problem is called nonlinear output

they basically mean that the linearized system is sta-

regulation problem, or alternatively, nonlinear ser-

bilizable and there exists a certain invariant mani-

vomechanism problem [10]. The corresponding au-

fold. The classical output regulation via error feed-

tonomous differential equation is usually called as

back has been solved in [1, 9] by application of sys-

the exogeneous system and is supposed to be neu-

tem immersion technique.

traly stable. In the sequel, the above setting will be

parametrized by unknown constant parameters is

referred to as the classical output regulation prob-

treated as a special case of exogenous signals and

lem.

In other words, the classical output regula-

the solution, extended from the error feedback reg-

tion problem treats a possible unknown reference

ulation, is referred to as the structurally stable reg-

signal and/or disturbances generated by the known

ulation in [1]. Some of the recent results in the ro-

neutrally stable autonomous exosystem with possi-

bust control eld [13, 14, 18] are based on the center

ble unknown initial states.

manifold theory and the related nonlinear regulator

For linear systems the classical output regulation

The plant uncertainty

theory; in the adaptive control eld. However, the

For nonlin-

main limitation of the classical regulation scheme

ear systems, the problem was rst studied in [7],

is that a precise model of the system that generates

and solutions to the output regulation of nonlinear

all exogenous inputs must be available, to be repli-

was extensively studied in

[2, 3, 4].

cated in the control law. 1

Supported

103.5/03/1130

by

PROMEP

under

research

program

This limitation becomes

immediately evident in the problem of rejecting a sinusoidal disturbances, not only of unknown am-

plitude and phase, but also of unknown frequency.

we outline the generalized nonlinear output regula-

Therefore, an alternative formulation would be to

tion problem which, as anticipated above, is based

require asymptotic tracking of known reference tra-

on the introduction of a driving signal to the exosys-

jectories in spite of unmodelled disturbances acting

tem. In Section 3, we state the assumptions neces-

on the exosystem. To unify this alternative formu-

sary to the well-posedness of the problem. In Sec-

lation with the classical output regulation concept,

tion 4 an adaptive controller solving the state feed-

the so-called generalized output regulation may

back generalized output regulation problem for a

be considered.

specic class of nonlinear systems including an un-

The generalized output regulation was rst posed

known vector parameter, is determined. We show

and solved in [17] for linear systems both continu-

how this specic problem is related to the standard

ous and discrete time in terms of necessary and suf-

problems of disturbance decoupling for nonlinear

cient geometric conditions involving the classical

systems which have been recently studied in [9, 12].

notions of disturbance decoupling. The correspond-

Finally, Section 5 draws conclusions and outlines

ing design procedure presented in [17] handles the

some future research.

unmodelled bounded disturbances generated by the known nonautonomous linear system driven by an unknown bounded reference signal. The question arises how the results established by [17] can be extended to the general case in which the plant is described by a nonlinear equation. The generalized output regulation problem resembles somehow the asymptotic model matching

2

Problem formulation

Following the linear concept of the generalized output regulation problem [17], the generalized output regulation problem for nonlinear systems can be introduced via the conguration provided by the master-slave block diagram of Figure 1. The task

problem (AMM) considered in [19], nevertheless,

       !     '

the crucial difference is that the input of the reference model in AMM is supposed to be known and

                         

is then used in the corresponding feedback compensator. Therefore, AMM solution cannot be directly

     $  !  4 . 2 0 /        3

applied to the generalized output regulation problem, considering the exogeneous system driven by

  "    #  y( ) *,+- . /10    / 76    2 %  & 5

an unknown signal as the analogue of the reference model in AMM problem. In [15], we were able to characterize the solvability of the state feedback generalized output regulation problem for nonlinear systems in terms of

Figure 1:

 

Conguration of the output regulation

r 6≡ 0, generalized output r ≡ 0, classical output regulation.

schemes. (a) (b)

regulation.

the solvability of the regulator equation with the classical concept of the invariant distribution. The

of the controller is to generate u so that the tracking

state feedback generalized almost output regulation

error e is converging to zero for all initial conditions

problem for a class of nonlinear systems is solved

of both plant and exosystem and all external signals

in [16].

r(t)

from a suitable functional class.

The plants

The purpose of this paper is to point out that

(slaves) we consider in this paper are afne multi-

by combining the approach to classical output reg-

input multi-output (MIMO) nonlinear systems, de-

ulation presented in [9] with the design method for

scribed by the equations of the form

generalized output regulation via state feedback presented in [15], it is easy to address the problem of



generalized output regulation for a signicant class We establish a link between the two

approaches; as a matter of fact, in this work, an adaptive controller, including a particular Lyapunov function is determined by exploiting concepts of the classical nonlinear regulation theory. The paper is organized as follows. In Section 2

f (x) +

gi (x)ui +

i=1

of nonlinear systems, in the presence of unknown parameters.

=

m X

yi

l X

pi (x)wi

(1)

i=1

:= f (x) + g(x)u + p(x)w, = hi (x), i = 1, . . . , p

(2)

where (1) describes the plant with state x, dened

X of the p output y ∈ R ,

Rn ,

on a neighborhood

origin of

u ∈

subjected to the ef-

Rm and

input

fect of a disturbance represented by the vector elds

pi (x), i = 1, . . . , l. tor eld f (·), and the

It is assumed that the vec-

p(·) are smooth vector elds, while each component of hi (·) is a smooth function, with f (0) = 0, g(0) = 0 and h(0) = 0. We only consider reference outputs to be columns of

g(·)

RF I There exists a neighborhood U

the solution of the closed loop system

x˙ = f (x) + g(x)γ(x, w) + p(x)w w˙ = s(w) + d(w)r

tracked and perturbations to be rejected which both are generated by an unknown exosystem as follows.

lowing nonautonomous system with output

lim e(t) = 0.

(3)

s(·) and the columns of d(·) are smooth vector elds with s(0) = 0 and q(0) = 0, q, q(0) = 0, is smooth function, while r(t) is unknown external driving signal. Further, it is assumed that r(t) is Rρ limited to a functional subclass of L∞ where it holds for all solutions of (3) and some class-K functions α, β that

In [15] the following result is proved: Theorem 1 Consider the system given in (1)-(2). Let Assumption 1 be satised. Then, the generalized output regulation problem via state feedback regulator is solvable if

C k (k ≥ 2) mappings x = π(w), u = c(w), π : Rl → Rn , c : Rl → Rm , ρ : Rl → Rρ , dened locally in a neighborhood of 0 l the origin W ⊂ R , with π(0) = 0, c(0) = 0, ρ(0) = 0 satisfying the so-called generalized

(a) there exist

kw(t)k Rl ≤ α(kw(0)kRl ) + β(kr(t)kLRρ ). L∞ ∞

regulation equation

∂π(w) (s(w) + d(w)ρ(w)) = f (π(w)) ∂w +g(π(w))c(w) + p(π(w))w (7) 0 = h(π(w)) − q(w) (8)

Roughly speaking, the Assumption 1 restricts the class of exogenous inputs to those signals which do not decay to zero and do not tend to innity as time

r.

The controller is

to be designed so that the slave obeys the master,

C k (k ≥ 2) mapping u ˜(x, w), u ˜: × → Rm , u ˜(0, 0) = 0, dened locally 0 n l in a neighborhood of the origin U ⊂ R × R n and a regular involutive distribution ∆ in T R

(b) there exist a

Rn

namely the error signal

e(t) = y(t) − q(w(t)) converges asymptotically to zero as t

(6)

t→∞

Here,

goes to innity for any signal

(5)

and error (4) satises

Assumption 1 (Exosystem): Exosystem is the fol-

w˙ = s(w) + d(w)r(t), w ∈ R yref = q(w), r ∈ Rρ , d := [d1 | . . . |dρ ].

of

(0, 0) such that, for each initial condition on U and for any signal r (piecewise continuous),

and

l

⊂ Rn × Rl

(4)

→ ∞.

In the following statement we give a precise formulation of the control problem under consideration.

Rl

such that (b1 ) the linear approximation of

[f + g˜ u(x, 0)]

is

Hurwitz,

(b2 ) u ˜(π(w), w) = c(w), (b3 ) for all w , [f (x) + g(x)˜ u(x, w) (b4 )

+ p(x)w, ∆] ⊂ ∆ ⊂ ker dh, dπ(∆w ) ⊂ ∆ ∀w, ∆w = span [d1 (w), . . . , dρ (w)] ⊂ T Rl .

Denition 1 (State Feedback Generalized Output Regulation Problem (SFGORP)): Given the reference output yref generated by an exosystem (3), the SFGORP consists in nding a state feedback con-

u = γ(x, w) where γ(·, ·) is a C k (k ≥ 2) mapping, with γ(0, 0) = 0 such that:

troller

SF I The equilibrium x

3

Standing assumptions

In this section we consider a plant modelled by equations of the form

x˙ = f (x) +

gi (x)ui +

i=1

= 0 of +

x˙ = f (x) + g(x)γ(x, 0)

m X

l X

pi (x)wi

i=1

vj (x)θj

j=1

is asymptotically stable in the rst approximation.

m X

yi

= f (x) + g(x)u + p(x)w + v(x)θ, = hi (x), i = 1, . . . , p

(9) (10)

where

θ

is a vector of unknown parameters and the

same considerations presented in the previous sec-

parameter

θ, the state behavior.

tion with regard to system (1)-(2), apply.

u=u ˜(x, w) − k(x)θˆ

Our goal is to solve the SFGORP for the system (9)-(10) without knowing the parameter

θ.

In the

ˆ −θ φ(t) = θ(t)

(11)

ˆ and θ represent the current estimate and θ(t)

the exact value of the unknown parameter, respectively. The main assumptions needed to solve the problem are, for convenience, listed in the following and then briey justied:

The closed loop system (9)-(10) with (13), taking

m functions k(x) such that v(x) = g(x)k(x)

By the Assumption 3, there exists a regular involu-

(12)

T Rn such that

(b3 ) and (b4 ) hold.

u ˜(x, w)

with

u ˜(0, 0) = 0,

satisfying (b1 )

and (b2 ).

Let

exist by virtue of the Frobenius Theorem [10], such

∆ = span{ ∂ζn∂ +1 , · · · , ∂ζ∂n }. 1

(ζ 1 , ζ 2 )T with ζ 1

pact notation ζ and

= ζ 2 = (ζn1 +1 , . . . , ζn ),

c(w), ρ(w),

with

= π(w), u = π(0) = 0, c(0) = 0, ρ(0) = 0

satisfying (7)-(8).

ζ˙ 1

=

f˜1 (ζ) + g˜1 (ζ)˜ u(Φ−1 (ζ), w) +˜ p1 (ζ)w − v˜1 (ζ)φ

by (b3)

p¯(ζ 1 , w) − v˜1 (ζ 1 )φ f˜2 (ζ) + g˜2 (ζ)˜ u(Φ−1 (ζ), w)

:=

ζ˙ 2

=

(17)

yi

˜ i (ζ 1 ) i = 1, . . . , p. h

=

Further, let

tions show that



π ˜ (w) =

ular form. Assumption 3, namely the existence of a regular involutive distribution and, in turn, the exis-

set to zero, is established in Assumption 4. Finally Assumption 5 is standard in classical output regulation of nonlinear systems [10].

4

Straightforward computa-

π ˜ 1 (w) π ˜ 2 (w)

 = Φ(π(w)),

c˜(w) = u ˜(Φ−1 (π(w)), w),

tence of the internal triangular decomposition (16)regulation problem when the unknown parameter is

with dim

π ˜ 1 = n1 , dim π ˜ 2 = n − n1

are solutions

to the following equations

∂π ˜ 1 (w) s(w) ∂w

=

f˜1 (˜ π (w)) + g˜1 (˜ π (w))˜ c(w) +˜ p1 (˜ π (w))w − v˜1 (˜ π (w))φ

by (b3)

:=

Regulator via state feedback

p¯(˜ π 1 (w), w) − v˜1 (˜ π 1 (w))φ

Although the main goal of paper is the design of an state feedback regulator, this section is devoted to briey discuss the solution when the state (x, w) is available. This preliminary discussion will make

(19)

∂π ˜ 2 (w) ∂w

s(w)

=

f˜2 (˜ π (w)) + g˜2 (˜ π (w))˜ c(w) +˜ p2 (˜ π (w))w − v˜2 (˜ π (w))φ

the presentation of the general solution more mean-

(20)

ingful. The rst step consists of choosing a control law to force, despite the presence of the unknown

(18)

π(w), c(w) be a solution of the reg-

parameter whose values enter to the plant in a partic-

(18). The state feedback controller, which solves the

(16)

+˜ p2 (ζ)w − v˜2 (ζ)φ

ulator equation (7)-(8). As specied in Assumption 2 we consider unknown

= (ζ1 , . . . , ζn1 ) ζ-

we have that in the

form

Assumption 5 (Solution of the regulator equation): There exist two smooth mappings x

Using the com-

coordinates the system (14)-(15) takes the following

Assumption 4 (Nominal regulator): There exists a mapping

∆ with dim∆ = n − n1 for some ζ = Φ(x) be new coordinates, which

tive distribution

that

Assumption 3 (Local decomposition): There exists a regular involutive distribution ∆ in

x˙ = f (x) + g(x)˜ u(x, w) − g(x)k(x)θˆ +p(x)w + v(x)θ (14) = f (x) + g(x)˜ u(x, w) + p(x)w − v(x)φ yi = hi (x), i = 1, . . . , p. (15)

n1 < n .

Assumption 2 (Matching condition): There exists a vector of

(13)

into account the Assumptions 2, has the form

following, we set

where

In view of this, we

consider the static feedback regulator

0

=

˜ π 1 (w)) − q(w). h(˜

(21)

Moreover, by Assumption 3

is stable. Then, the theory of stability [6, 11] assures the existence of a Lyapunov function V

∂π ˜ 1 (w) d(w) ≡ 0. ∂w

(22)

ζ˜ = ζ − π ˜ (w)

(23)

V (0, 0) = 0 (33) ˜ ˜ V (ζ, w) > 0 ∀ (ζ, w) ∈ U − (0, 0) (34) ˜ w) = ∂V F (ζ, ˜ w) + ∂V (s(w) V˙ (ζ, ∂w ∂ ζ˜ ˜ w) ∈ U (35) +d(w)r) ≤ 0 ∀ (ζ,

where π ˜ (w) is a solution of the system (19)-(20) and

ζ˜ = (ζ˜1 , ζ˜2 )T with ζ˜1 = (ζ˜1 , . . . , ζ˜n1 ) and 2 ˜ ζ = (ζ˜n1 +1 , . . . , ζ˜n ). Then the system (16)-(17) can be rewritten in the form



  1  F 1 (ζ˜1 , w) G (ζ˜1 , w) + ˜ w) ˜ w) φ F 2 (ζ, G2 (ζ, ˜ w) + G(ζ, ˜ w)φ = F (ζ, (24) w˙ = s(w) + d(w)r (25) ˙ ζ˜ =

˜ w) = 0, R be the set of all points where V˙ (ζ, and M be the largest invariant set (with respect to the motion of the system (24)-(25)) in R. Let

Assumption 6 (Largest invariant): Let P be the following set

˜ w) : ζ˜ = 0} P = {(ζ, M ⊆ P.

F 1 (ζ˜1 , w) = p¯(ζ˜1 + π ˜ 1 (w), w) − p¯(˜ π 1 (w), w) (26)

˜2

˜ w) = f (ζ˜ + π F (ζ, ˜ (w)) + g˜ (ζ˜ + π ˜ (w)) · −1 ˜ u ˜(Φ (ζ + π ˜ (w)), w) + 2

2

+˜ p2 (ζ˜ + π ˜ (w))w − f˜2 (˜ π (w)) 2 2 −˜ g (˜ π (w))˜ c(w) − p˜ (˜ π (w))w

Now, we present the following theorem on the generalized output regulation problem via state feedback regulator. Theorem 2 Consider the system given in (9)-(10). Assume Assumptions 1-6. Then, the controller

u=u ˜(x, w) − k(x)θˆ

(27)

∂V ˜ w) φ˙ = − G(ζ, ∂ ζ˜

G (ζ˜1 , w) = −˜ v 1 (ζ˜1 + π ˜ 1 (w)) + v˜1 (˜ π 1 (w)) (28)

˜ w) = −˜ G (ζ, v 2 (ζ˜ + π ˜ (w)) + v˜2 (˜ π (w)) 2

u ˜(x, w)

(29)

solves the State Feedback General-

ized Output Regulator Problem for the system (9)-

˜=0 (10) considered with θ = 0, the equilibrium ζ of the system

" ˙ #   ζ˜1 F 1 (ζ˜1 , w) = ˙ ˜ w) F 2 (ζ, ζ˜2   p¯(ζ˜1 , w) = ˜ + g˜2 (ζ)˜ ˜ u(Φ−1 (ζ, ˜ 0) f˜2 (ζ)

where

˜ w) G(ζ,

and

˜ w) V (ζ,

(38)

are the previously de-

ned functions, solves the State Feedback Generalized Output Regulation Problem. Proof: Assuming Assumptions 1-6 hold, we aim to show that Denition 1 is satised with

u ˜(x, w) − k(x)θˆ.

γ(x, w) =

Obviously, by Assumption 4 the

condition SF I of Denition 1 holds. To prove the condition RF I consider the Lyapunov function

˙ ζ˜ =

˜ φ, w) = V (ζ, ˜ w) + 1 φ2 W (ζ, 2 (30)

(39)

Taking into account (24)-(25)) and (38), the derivative of the function (39) is such that

is exponentially stable. Moreover, the equilibrium

w = 0 of the exosystem (25) is stable by Assump-

W (0, 0, 0) = 0 (40) ˜ φ, w) > 0 ∀ (ζ, ˜ w) ∈ U − (0, 0), ∀ φ W (ζ,

tion 1. The above properties imply [11] that the equilib-

˜ w) = (0, 0) of the system (ζ,

˙ ˜ w) ζ˜ = F (ζ, w˙ = s(w) + d(w)r

(37)

together with the update law

1

rium

(36)

with

where

Since

of a

neighborhood of the origin U such that

Further, let

denote

˜ w) (ζ,

(31) (32)

(41)

˜ φ, w) = ˙ (ζ, W

∂V ˜ w) + ∂V (s(w) F (ζ, ˜ ∂w ∂ζ ˜ w) ∈ U, ∀ φ +d(w)r) ≤ 0∀ (ζ, (42)

From (40)-(42), it can be deduced that the origin

[8] J. Huang and W.J. Rugh, Stabilization on zero-

of the closed loop system is stable. Moreover, by

error manifolds and the nonlinear servomech-

applying the invariance principle of LaSalle's The-

anism problem, IEEE Transactions on Auto-

orem [6, 11] it is possible to claim that the mo-

matic Control, Vol. 37, 1992, pp. 1009-1013.

tion of the system (24)-(25) originated in a point U , asymptotically converge to the largest invariant sub-

˜ φ, w) ˙ (ζ, set characterized by W

= 0, that is, due to

[9] A. Isidori, Nonlinear Control Systems, 3rd ed., New York: Springer-Verlag, 1995.

ζ˜ = 0. From (23), this implies, that every motion ζ(t), originated in U , asymptotically converges to the center manifold π ˜ (w), i.e. we can see from (4) and (8) that error e(t) tends to zero

[10] A. Isidori and C. I. Byrnes, Output regula-

as time tends to innity. In other words, the condi-

[11] H.K. Khalil, Nonlinear Systems, 2nd ed. New

Assumption 6, by

tion (RF I ) of Denition 1 holds.

/

tion of nonlinear systems, IEEE Transactions on Automatic Control, Vol. 35, 1990, pp. 131140.

York: MacMillan, 1996. [12] R. Marino and P. Tomei, Nonlinear Control

5

Conclusions

The solution of the problem of generalized regu-

Design-Geometric, Adaptive and Robust.London, U.K.: Prentice-Hall, 1995.

lation with nonautonomous exosystem for a large

[13] J.B. Pomet and L. Praly, Adaptive nonlinear

class of nonlinear systems characterized by the pres-

regulation: Equation error from the Lyapunov

ence of an unknown parameter, has been presented.

function, 28th Conference on Decision and

The output regulation is achieved by linking concepts of the robust control theory (center manifold), with concepts of the adaptive control theory (Lyapunov function).

Control, Tampa, 1989. [14] F.D. Priscoli, Adaptive control of a class of nonlinear systems via the robust control approach, IFAC NOLCOS92, Vol. 1, 1992, pp. 320-323.

References: [1] C. I. Byrnes, F. Delli Priscoli, A. Isidori and W. Kang, Structurally stable output regulation of nonlinear systems, Automatica, Vol. 33, 1997, pp. 369-285. [2] E.J. Davison, The output control of linear time-invariant multi-variable systems with unmeasured arbitrary disturbances, IEEE Transactions on Automatic Control, Vol. 17, 1972, pp. 621-630. [3] B.A. Francis and W. Murray Wonham, The internal model principle of control theory, Automatica, Vol. 12, 1976, pp. 457-465. [4] B.A. Francis, The linear multivariable regulator problem, SIAM Journal on Control and Optimization, Vol. 15, 1977, pp. 486-505. [5] R.A. Freeman and P.V. Kokotovi c, Tracking controllers for systems linear in unmeasured states, Automatica, Vol. 32, 1996, pp. 735-746. [6] W. Hahn, Stability of motion, Springer-Verlag, 1967. [7] J.S.A. Hepburn and W.M. Wonham, Error feedback and internal models on differentiable manifolds, IEEE Transactions on Automatic Control, Vol. 29, 1981, pp. 397-403.

 and V. Ku cera, [15] L.E. Ramos, S. Celikovský “Generalized Output Regulation Problem for a Class of Nonlinear Systems with Nonautonomous Exosystem," IEEE Transactions on Automatic Control, accepted.  [16] L.E. Ramos, S. Celikovský, V. Ku cera and J. Ruíz, “Almost Output Regulation of A Class of Nonlinear Systems with Nonautonomous Exosystem," Latin American Control Conference. Guadalajara, México, 2002. [17] A. Saberi, A.A. Stoorvogel and P. Sannuti, On output regulation for linear systems, International Journal of Control, Vol. 74, 2001, pp. 783-810. [18] A.R Tell, Robust and adaptive nonlinear output regulation, European Control Conference, Grenoble, 1991. [19] M. Yokomichi and M. Shima, Another approach to asymptotic model matching problem for nonlinear systems, International Journal of Robust and Nonlinear Control, Vol. 8, 1998, pp. 1119-1131.