Professional Documents
Culture Documents
10.1007/s11768 006 6053 8
10.1007/s11768 006 6053 8
1007/s11768-006-6053-8
Abstract: The problem of observer-based robust H-infinity control is addressed for a class of linear discrete-time
switched systems with time-varying norm-bounded uncertainties by using switched Lyapunov function method. None of
the individual subsystems is assumed to be robustly H-infinity solvable. A novel switched Lypunov function matrix with
diagonal-block form is devised to overcome the difficulties in designing switching laws. For robust H-infinity stability
analysis, two linear-matrix-inequality-based sufficient conditions are derived by only using the smallest region function
strategy if some parameters are preselected. Then, the robust H-infinity control synthesis is studied using a switching
state feedback and an observer-based switching dynamical output feedback. All the switching laws are simultaneously
constructively designed. Finally, a simulation example is given to illustrate the validity of the results.
Keywords: Linear discrete-time switched system; Robust H-infinity control; Switched Lyapunov function; Linear
matrix inequality (LMI)
nally, a simulation example is given to illustrate the validity analysis and control synthesis for system (1) via an ap-
of the results. propriately designed switching law. The robust H∞ perfor-
Notations In the sequel, the Euclidean norm is used mance analysis looks for the condition under which the un-
for vectors. μmax (W ) denotes the maximum singular value forced system (1) has robust H∞ performance. The robust
of any square matrix W . The notation W > (, <, )0 H∞ control synthesis designs a switching feedback con-
is used to denote a symmetric positive-definite (positive- troller such that the resulting closed-loop system of system
semidefinite, negative-definite, negative-semidefinite, re- (1) has robust H∞ performance, in this case, system (1) is
spectively) matrix W , I is the identity matrix of appropriate said to be robust H∞ solvable.
dimension. N denotes the set of nonnegative integer. The In this paper, we consider using the switching state feed-
space l2 [0, ∞) consists of quadratically integrable vectors back controller
sequences over [0, ∞). The symbol ∗ is used to induce a uk = Kσ xk , (3)
symmetric structure of a matrix, that is, if given matrices
and the observer-based switching dynamical output feed-
L = LT and R = RT of appropriate dimensions, then
back controller
L ∗ L ST uk = Kσ x̂k ,
= . (4)
S R S R x̂k+1 = Aσ x̂k + Bσ uk + Lσ (yk − Fσ x̂k ),
For simplicity, the system with uk ≡ 0 (or wk ≡ 0) is where x̂k is the observer state, Kσ and Lσ are the controller
referred to as the unforced system ( or the disturbance-free gain and observer gain to be designed, respectively.
system). Remark 1 The problem of robust H∞ control synthe-
sis for system (1) is studied when none of the individual
2 System description subsystems is robust H∞ solvable. This is because if there
We consider a switched system of the form exists a subsystem in system (1), say, j-th subsystem, which
⎧
⎪ is robust H∞ solvable, then the problem will be trivial.
⎨ xk+1 = Âσ xk + B̂σ uk + H1σ wk ,
⎪
yk = Fσ xk , (1)
⎪
⎪ 3 Robust H∞ performance analysis
⎩z = C x + D u + H w ,
k σ k σ k 2σ k In this section, we first consider the unforced system (1).
where k ∈ N, xk ∈ Rn is the state, uk ∈ Rl is the con- We introduce
trol input, wk ∈ Rq is the disturbance input which belongs 1, i = σ(k),
to l2 [0, ∞), yk ∈ Rr is the measurement output, zk ∈ Rp αi (k) = i ∈ M̄ . (5)
0, i = σ(k),
is the controlled output, σ(k) : N → M̄ = {1, 2, · · · , m}
Then, the unforced system (1) is described as follows,
is a piecewise constant switching signal. For ∀i ∈ M̄ , Ci , ⎧
⎪
m
m
Di , Fi , H1i and H2i are known real constant matrices; Âi ⎪
⎨ xk+1 = αi (k)Âi xk + αi (k)H1i wk ,
i=1 i=1
and B̂i are unknown matrices with time-varying parameter
m
m (6)
⎪
⎪
uncertainties, and are assumed to be of the form ⎩ zk = αi (k)Ci xk + αi (k)H2i wk .
i=1 i=1
[Âi , B̂i ] = [Ai + ΔAi , Bi + ΔBi ] Choose the switched Lyapunov function for system (6) as
= [Ai , Bi ] + Ei Γ i [Fi1 , Fi2 ], (2)
m
V (k, xk ) = xT
k( αi (k)Pi )xk , (7)
where Γi is an unknown time-varying matrix function sat- i=1
isfying μmax (Γi ) 1, Ai , Bi , Ei , Fi1 and Fi2 are known where Pi > 0, i ∈ M̄ .
real constant matrices. Proposition 1 Given γ > 0. The system (6) has robust
Definition 1 Given γ > 0. The unforced system (1) H∞ performance γ, if there exist a set of scalars βij 0 and
is said to have robust H∞ performance γ, if there exists a matrices Pi > 0, i, j ∈ M̄ , such that the following matrix
switching signal σ(k), such that for all admissible uncer- inequalities hold,
⎡
tainties, the following conditions are satisfied: T −1 T
m
⎢ Âi Pi Âi − Pi + γ Ci Ci + βij (Pj − Pi )
i) The unforced disturbance-free system (1) is stable; ⎣ j=1
T
ii) Under the zero initial condition, the controlled output H1i Pi Âi + γ −1 H2i
T
Ci
zk satisfies ∗
∞
2
∞
2 < 0. (8)
|zk | < γ 2 |wk | T −1 T
H1i Pi H1i + γ H2i H2i − γI
k=0 k=0
Proof Choose a switching law as follows,
for any nonzero wk ∈ l2 .
In this paper, we investigate the robust H∞ performance σ(k) = arg min{xT
k Pi xk }. (9)
i∈M̄
280 Z. SONG et al. / Journal of Control Theory and Applications 2007 5 (3) 278–284
By the Schur complements, we get (16). Hence this propo- i, j ∈ M̄ , such that the following matrix inequalities hold,
sition holds. ⎡ ⎤
−λi I ∗ ∗ ∗
By Proposition 1, A), B) and C) in Proposition 2 are suf- ⎢ T T ⎥
⎢ Gi Fi1 ηi (Qi − Gi − GT i ) ∗ ∗⎥
ficient conditions for system (6) to have robust H∞ perfor- ⎢ ⎥ < 0. (19)
⎢ ⎥
mance γ. But Proposition 2 is not applicable directly be- ⎣ 0 Ji Gi X̄i ∗ ⎦
cause of the uncertainty involved in matrix Âi . For its ap- 0 β̄i Gi 0 Q̄
plication, we need the following Lemma. Proof It follows from C) in Proposition 2 and the proof
Lemma 1 [9] Let Y, M, N be given matrices of ap- of Theorem 1.
propriate dimensions, then for any matrix Γ satisfying Remark 2 Unlike the positive definite matrix Qi in
μmax (Γ ) 1, (18). Gi in (19) is a general matrix, not even necessar-
Y + M Γ N + N T Γ T M T < 0, ily symmetric. This feature means that inequalities (19) are
much easier to solve than inequalities (18).
if and only if there exists a constant λ > 0 such that
1 T 4 Robust H∞ control synthesis
Y + λM M T + N N < 0.
λ In this section, we design the controller (3) and the con-
Theorem 1 Given γ > 0. The system (6) has robust troller (4), such that the resulting closed-loop system of sys-
H∞ performance γ under the switching law (9), if there ex- tem (1) has robust H∞ performance.
ist matrices Qi > 0, scalars βij 0 and λi > 0, i, j ∈ M̄ , 4.1 Switching state feedback
such that the following matrix inequalities hold,
When state variables are completely available, applying
⎡ ⎤
−λi I ∗ ∗ the state feedback controller (3) to system (1) results in the
⎢ ⎥
⎢ QT F T −η Q + β Q Q−1 Q ∗ ⎥
m closed-loop system
⎢ i i1 i i ij i j i ⎥ < 0, (18) ⎧
⎣ j=1 ⎦ ⎪ m m
⎪
⎨ xk+1 = αi (k)(Âi + B̂i Ki )xk + αi (k)H1i wk ,
0 Ji Qi X̄i i=1 i=1 (20)
⎪
m
m
⎪
⎩ kz = α i (k)(Ci + D K )x
i i k + αi (k)H w
2i k .
where
i=1 i=1
JiT = [ AT
i CiT 0 ], X̄i = Xi + λi Ēi ĒiT , From Proposition 2 and Lemma 1, we can obtain the the-
orems below.
ĒiT = [ EiT 0 0 ].
Theorem 3 Given γ > 0. System (1) is robust H∞
Proof From (16) and (2), we have solvable under the switching law (9), if there exist matri-
Zi + Mi Γi Ni + NiT ΓiT MiT < 0, ces Qi > 0, Yi , scalars βij 0 and λi > 0, i, j ∈ M̄ , such
that the following matrix inequalities hold,
where ⎡ ⎤
⎡ ⎤ −λi I ∗ ∗
m ⎢ ⎥
⎢ QT F T + Y T F T −η Q + β Q Q−1 Q ∗ ⎥
−1 m
⎢ −ηi Qi + βij Qi Q Qi ∗ ⎥ 0
Zi = ⎣ j=1
j
⎦ , Mi = , ⎢ i i1 i i2 i i ij i j i ⎥
⎣ j=1 ⎦
Ji Qi Xi Ē i
0 Ji Qi + Jui Yi X̄i
Ni = [ Fi1 Qi 0 ]. < 0, (21)
where we get
T
Jˆui = [ B̂iT DiT 0 ]. σ(k) = arg min{x̂T −1
k Qi x̂k }. (28)
i∈M̄
Therefore, system (1) is robust H∞ solvable.
Let Yi = Ki Qi , then By Proposition 2, system (26) has robust H∞ perfor-
⎡ ⎤ mance γ, if there exist matrices Q̃i > 0 and scalars βij 0,
m
−1
⎢ −ηi Qi + βij Qi Qj Qi ∗ ⎥ i, j ∈ M̄ , such that the following matrix inequalities hold,
⎣ j=1 ⎦ < 0. ⎡ ⎤
Jˆi Qi + Jˆui Yi Xi m
−1
⎢ −η i Q̃i + β ij Q̃i Q̃j Q̃i ∗ ∗ ∗ ⎥
⎢ j=1 ⎥
From Lemma 1 and the proof of Theorem 1, we find that ⎢ ⎥
⎢ Ãi Q̃i −Q̃i ∗ ∗ ⎥ < 0. (29)
(21) is equivalent to (23). Hence this theorem holds. ⎢ ⎥
⎢ C̃i Q̃i 0 − γI ∗ ⎥
Theorem 4 Given γ > 0. System (1) is robust H∞ ⎣ ⎦
T T
solvable under the switching law (9), if there exist matri- 0 H̃1i H̃2i − γI
ces Qi > 0, Gi , Yi , scalars βij 0 and λi > 0, i, j ∈ M̄ ,
By Lemma 1, we get the following theorems.
such that the following matrix inequalities hold,
⎡ ⎤ Theorem 5 Given γ > 0. System (26) has robust H∞
−λi I ∗ ∗ ∗ performance γ under the switching law (28), if there exist
⎢ T T ⎥
⎢ Gi Fi1 + YiT Fi2 T
ηi (Qi − Gi − GT i ) ∗ ∗⎥ matrices Qi > 0, Q > 0, Ui , Vi and Yi , scalars βij 0 and
⎢ ⎥ < 0. (24)
⎢ ⎥ λi > 0, i, j ∈ M̄ , such that the following matrix inequali-
⎣ 0 Ji Gi + Jui Yi X̄i ∗ ⎦
ties hold,
0 β̄i Gi 0 Q̄ ⎡
In this case, a switching state feedback controller is given −λi I ∗
⎢ m
⎢
by ⎢ −Qi Fi1T
− YiT Fi2
T
− ηi Qi + βij Qi Q−1
j Qi
m ⎢ j=1
⎢
uk = αi (k)Ki xk , Ki = Yi G−1
i . (25) ⎢ QFi1 T
0
i=1 ⎣
Proof It is easy to obtain from (C) in Proposition 2 and 0 J1i Qi + J˜u1i Yi
˜
the proof of Theorem 3. ⎤
∗ ∗
⎥
4.2 Observer-based switching dynamic output feed- ∗ ∗ ⎥
⎥ < 0, (30)
back ⎥
−Q ∗ ⎦
When state variables are not completely available, apply-
J˜2i Q − Ui J˜u2i X̃i + λi Ẽi ẼiT
ing the controller (4) to system (1) results in the following
closed-loop system: where
⎧
⎪
m m T ˜T
⎪
⎨ x̃k+1 = αi (k)Ãi x̃k + αi (k)H̃1i wk , J˜1i = [ AT T T T
i 0 Ci 0 ], Ju1i = [ Bi 0 Di 0 ],
i=1 i=1 T ˜T
⎪
m
m (26) J˜2i = [ 0 AT T T T
i −Ci 0 ], Ju2i = [ Fi Fi 0 0 ],
⎪
⎩ zk = αi (k)C̃i x̃k + αi (k)H̃2i wk , ⎡ ⎤ ⎡ ⎤
i=1 i=1 0 −Qi ∗ ∗ ∗
⎢ ⎥ ⎢ ⎥
⎢ Ei ⎥ ⎢ ∗ ⎥
⎥ , X̃i = ⎢ 0 −Q ∗
where
Ẽi = ⎢
⎢ ⎥ ⎢
⎥,
⎥
x̂k Ai + Bi Ki −Li Fi ⎣ 0 ⎦ ⎣ 0 0 −γI ∗ ⎦
x̃k = , Ãi = ,
x̂k − xk −(ΔAi +ΔBi Ki ) Âi −Li Fi T T
0 0 −H1i H2i −γI
0
and
H̃1i = , C̃i = Ci + Di Ki −Ci , H̃2i = H2i .
−H1i
Vi Fi = Fi Q. (31)
Choose a switched Lyapunov function candidate
The controller gain Ki and the observer gain Li are given
m
Ṽ (k, x̃k ) = x̃T
k( αi (k)Q̃−1
i )x̃k , by
i=1
5 Simulation example
In the section, a simulation example is given to illustrate
the validity of the results. Fig. 1 The state response of the first subsystem of (37).
284 Z. SONG et al. / Journal of Control Theory and Applications 2007 5 (3) 278–284