Enlarged Finite Time Exact Compensation in Discrete Disturbed Systems
Larbi AFIFI
Laboratory MACS, Faculty of Sciences Ain Chock B.P.5366-Maˆ arif, Casablanca, Morocco
l.afifi@fsac.ac.ma, larbi
−afifi@yahoo.fr Maria HAKAM
Laboratory MACS, Faculty of Sciences Ain Chock B.P.5366-Maˆ arif, Casablanca, Morocco
[email protected] Abdelhakim CHAFIAI
Laboratory MACS, Faculty of Sciences Ain Chock B.P.5366-Maˆ arif, Casablanca, Morocco
a.chafi[email protected] Mehdi BENSSASSI
Laboratory MACS, Faculty of Sciences Ain Chock B.P.5366-Maˆ arif, Casablanca, Morocco
[email protected] Abstract
In this work, we study the notion of enlarged exact compensation for
a class of discrete linear distributed systems. It’s an extension to the dis-
crete and regional cases of previous works on the continuous finite time
remediability problem. It consists to study, with respect to the output
operator (sensors), the existence of an input one ensuring the reduction
of a disturbance effect on the output by bringing the observation in a
given tolerance zone C. We show that under convenient hypothesis and
with a convenient choice of the input operator (actuators), the consid-
ered control problem admits a unique solution and we show how to find
it. Then, we extend this notion as well as the developed approach to
the regional discrete case.
Keywords: Enlarged exact remediability, control, observation, actuators, sensors, region.
1 Considered system and notations
Without loss of generality , we consider a class of linear distributed systems [2, 4, 6, 7, 8] described by the following discrete equation
(S
d)
⎧ ⎨
⎩
z
k+1= φz
k+ Bu
k+ f
k; k = 0, · · · , N − 1 z
0∈ X
(1)
where φ ∈ L (X), B ∈ L (U, X ), z
k, f
kand u
k∈ U are respectively the state, the disturbance, the control at step k and N is an integer, N ≥ 1. X and U are supposed to be Hilbert spaces. The system (1) is augmented by the output equation
(E
d) y = Cz (2)
where C ∈ L (X, Y ), z = (z
1, ...., z
N)
tr, y = (y
1, ...., y
N)
trwith y
k= Cz
kfor k = 1, · · · , N . Y is the observation space, a Hilbert space.
The state of the system (1) at the final step N , noted z
u,f, is given by
z
u,f= φ
Nz
0+
N
−1 i=0φ
N−1−iBu
i+
N−1
i=0
φ
N−1−if
iand the corresponding observation is given by
y
u,f= Cφ
Nz
0+
N−1
i=0
Cφ
N−1−iBu
i+
N−1
i=0
Cφ
N−1−if
iLet H
det H
dbe the operators defined by
H
d: U
N−→ X
u = (u
0, ..., u
N−1)
tr−→ H
du =
N
−1 i=0φ
N−1−iBu
i(3)
and
H
d: X
N−→ X f = (f
0, ..., f
N−1)
tr−→ H
df =
N−1
i=0
φ
N−1−if
i(4) The observation at the final step N becomes
y
u,f= Cφ
Nz
0+ CH
du + CH
df or
y
u,f= Cφ
Nz
0+ K
du + R
df (5) where K
dand R
dare the operators defined by
K
du = CH
du =
N
−1 i=0Cφ
N−1−iBu
i(6)
and
R
df = CH
df =
N−1
i=0
Cφ
N−1−if
i(7)
Let us note that the systems (1) and (2) can be a discrete version of con- tinuous time equations. Indeed, if we consider the system described by the following linear state equation
⎧ ⎨
⎩
z(t) =
·Az(t) + Bu(t) + g(t) ; 0 < t < T
z
0∈ X
(8)
augmented by the following output equation
y(t) = Cz(t) ; 0 < t < T (9) where A generates a strongly continuous semi-group (s.g.f.c) (S(t))
t≥0on the state space X, B ∈ L (U, X ), C ∈ L (X, Y ), u ∈ L
2(0, T ; U) and g ∈ L
2(0, T ; X), then, for τ =
TN
> 0 with N sufficiently large, the corresponding discrete ver-
sion is as follows
⎧ ⎨
⎩
z
k+1= φz
k+ B u
k+ f
k; 0 ≤ k ≤ N − 1 z
0∈ X
(10)
augmented by the output equation
y
k= Cz
k; 0 ≤ k ≤ N − 1 (11)
where z
k= z(kτ ); u
k= u(kτ ) and
φ = S(τ ); B u
k=
τ0
S(τ − s)Bu
k(s)ds ; f
k=
τ0
S(τ − s)g
k(s)ds (12) u
kand g
kare respectively the restrictions of u and g to the time interval [kτ, (k + 1)τ [. If τ is small, u
kand g
kcan be assumed to be constant on the interval [kτ, (k + 1)τ [.
2 Notion of enlarged exact remediability
In the ”normal” case, i.e. without disturbance and without control term, the system (1) becomes
⎧ ⎨
⎩
z
k+1= φz
k; k = 0, · · · , N − 1 z
0∈ X
The state z
kat step k is given by
z
k= φ
kz
0and the corresponding observation y
kis given by
y
k= Cφ
kz
0But in the ”abnormal” case where the system is excited by a known or unknown
disturbance f = (f
0, · · · , f
N−1)
tr, the observation at step k becomes
y
k= Cφ
kz
0+
k−1i=0
Cφ
k−1−if
i; k = 1, · · · , N Generally, we have y
k= Cφ
kz
0; k = 1, · · · , N.
The problem consists to study the existence of an input operator B, with respect to the output one, in order to reduce the effect of this disturbance at the final step N by bringing the observation in a given zone of tolerance C , i.e.
such that:
For any f = (f
0, · · · , f
N−1)
tr∈ X
N, there exists u = (u
0, · · · , u
N−1)
tr∈ U
Nsatisfying
y
u,f= Cφ
Nz
0+
N−1
i=0
Cφ
N−1−iBu
i+
N−1
i=0
Cφ
N−1−if
i∈ C (13) where C is a convex, closed and nonempty subset of Y . This leads to the following definition
Definition 2.1
i) A disturbance f ∈ X
Nis said to be C -remediable if there exists u ∈ U
Nsuch that
y
u,f= Cφ
Nz
0+ K
du + R
df ∈ C where K
dand R
dare respectively defined by (6) and (7).
ii) The system (1) augmented by (2) is said to be C -remediable if any distur- bance f ∈ X
Nis C -remediable.
Remark 2.2 Let us note that:
1) f is C -remediable ⇐⇒ Im(K
d) ∩C
1= ∅ where C
1= C − Cφ
Nz
0− R
df . 2) If C =
Cφ
Nz
0, then we have a problem of exact remediability.
3) If C = B(0, ), then a disturbance f C -remediable is said -remediable, and f is -remediable ⇐⇒ Im(K
d) ∩ B (R
df, ) = ∅
⇐⇒ P
Ker(B∗R∗d)(R
df) <
where P
Fis the orthogonal projection on F, F is a closed subset of Y .
4) If f is -remediable, then for any
> , f is
-remediable. The converse is not true.
5) The cost increases when decreases.
3 Enlarged exact remediability with minimum energy
Let C be a convex, closed and nonempty subset of Y and f ∈ X
N. We consider the following problem (P
N) of enlarged exact remediability with minimum energy
(P
N)
min J(u) with J(u) = u
2UNsubject to y
u,f∈ C (14)
We suppose that the disturbance f is C -remediable, the problem (P
N) is then well defined and has a unique solution v
∗in the set of admissible controls defined by
U
ad=
u ∈ U
N/y
u,f∈ C The solution v
∗of (P
N) is characterized by
J (v
∗)(v − v
∗) ≥ 0 ; ∀ v ∈ U
adi.e.
v
∗, v − v
∗≥ 0 ; ∀ v ∈ U
adThe problem (P
N) is a generalization of the following problem (P
N0) of exact compensation
(P
N0)
min u
2subject to K
du + R
df = 0 (15) It is sufficient to consider C =
Cφ
Nz
0. If u
∗is the solution of (P
N0), then
v
∗≤ u
∗Hence, the optimal cost corresponding to (P
N) is reduced with respect to that corresponding to (P
N0).
The problem (P
N) is also a generalization of the -remediability one, it sufficient to consider C = B (0, ).
If C
1and C
2are two nonempty, closed and convex subsets of Y such that C
1⊂ C
2, then the C
1-remediability implies the C
2-remediability, and the cost is decreasing when C is increasing.
For the resolution of the problem (P
N), we use an extension of the Hilbert Uniqueness Method (HUM) combined with an appropriate penalization method.
For θ ∈ Y
≡ Y , let
θ
FN= (
N
−1 k=0B
∗(φ
∗)
kC
∗θ
2U
)
12(16)
·
FNis a semi-norm. If
1Ker(C
∗) = { 0 } , then ·
FNis a norm on Y if and only if (S
d) + (E
d) is weakly remediable. Under this condition, we consider the space
F
N= Y
·FNF
Nis a Hilbert space with the inner product
θ, σ
FN=
N−1
k=0
B
∗(φ
∗)
kC
∗θ, B
∗(φ
∗)
kC
∗σ
U
and the operator Λ
Ndefined on Y by
Λ
Nθ =
N−1
k=0
Cφ
kBB
∗(φ
∗)
kC
∗θ
admits a unique extension as an isomorphism from F
N−→ F
Nsuch that Λ
Nθ, σ
Y= θ, σ
FN; ∀ θ, σ ∈ F
NThe following result gives a characterization of the optimal control ensuring the C -compensation of a disturbance f and shows how to find it.
1
This hypothesis is true in the usual case where the observation is given by means of
sensors [2, 5, 7].
Theorem 3.1
If C is a convex, closed and nonempty subset of Y and if C ∩ (Cφ
Nz
0+ R
df + F
N) = ∅ then
i- there exists a unique θ
fin F
Nsuch that
Λ
Nθ
f+ Cφ
Nz
0+ R
df ∈ C (17) and
θ
f, y − Λ
Nθ
f− Cφ
Nz
0− R
df
≥ 0 ; ∀ y ∈ C ∩ (Cφ
Nz
0+ R
df + F
N) (18) ii- The control u
θfdefined by
u
θf= K
d∗θ
f(19)
is the unique solution of the problem (P
N). Moreover, u
θfis optimal with u
θf2
UN
= θ
f2FN
Proof:
Unicity: Let θ
fand σ
f∈ F
Nbe such that for any y ∈ C ∩ (Cφ
Nz
0+ R
df + F
N), we have
θ
f, y − Λ
Nθ
f− Cφ
Nz
0− R
df
≥ 0
and
σ
f, y − Λ
Nσ
f− Cφ
Nz
0− R
df
≥ 0
then for y = Λ
Nσ
f+ Cφ
Nz
0+ R
df in the first inequality and y = Λ
Nθ
f+ Cφ
Nz
0+ R
df in the second one, we obtain
θ
f, Λ
Nσ
f− Λ
Nθ
f≥ 0 and
σ
f, Λ
Nθ
f− Λ
Nσ
f≥ 0
Hence
θ
f, Λ
Nσ
f− Λ
Nθ
f+ σ
f, Λ
Nθ
f− Λ
Nσ
f= θ
f− σ
f, Λ
Nσ
f− Λ
Nθ
f= − θ
f− σ
f2FN
≥ 0 consequently, θ
f= σ
f.
Existence: It’s more complex and is done in stages.
For α > 0 , we consider the following criterion with two variables
J
α(y, v) = 1
α y
v,f− y
2+ v
2(20) where y ∈ Y ; v ∈ U
Nand y
v,fis given by (5). We consider the corresponding minimization problem
(P
α)
min J
α(y, v)
y ∈ C et v ∈ U
N(21)
We have the following result concerning the existence of a solution for (P
α).
Proposition 3.2
1- The minimization problem (P
α) admits a solution (y
α, v
α).
2- (y
α, v
α) verifies the inequalities 1
α y
α− y
vα,f, y − y
α≥ 0 ; ∀ y ∈ C (22) and
− 1
α y
α− y
vα,f, y
v,f− y
vα,f+ v
α, v − v
α≥ 0 ; ∀ v ∈ U
N(23) Proof:
1- Let (y
α(k), v
α(k))
k≥0be a minimizing sequence.
J
α(y
α(k), v
(k)α) inf
(y,v)∈C×UN
J
α(y, v) when k −→ + ∞
The sequence k −→ J
α(y
α(k), v
α(k)) is bounded because it is convergent. Conse-
quently, there exists c
1> 0 such that J
α(y
α(k), v
α(k)) ≤ c
1; ∀ k ≥ 0.
Since v
(k)α2
= J
α(y
α(k), v
α(k)) − 1 α
y
v(k)α ,f
− y
α(k)2
≤ J
α(y
(k)α, v
α(k))
then v
α(k)2
≤ c
1; ∀ k ≥ 0, i.e. (v
α(k))
k≥0is bounded in U
N. The map v = (v
0, ..., v
N−1)
tr∈ U
N−→ K
dv =
N−1
i=0
Cφ
N−1−iBv
i∈ Y is linear and continuous, then there exists c
2> 0 such that
K
dv = y
v,f− y
0,f≤ c
2v ; ∀ v ∈ U
NConsequently, (y
vα(k),f
)
k≥0is bounded in Y . On the other hand y
v(k)α ,f
− y
α(k)2
= αJ
α(y
α(k), v
α(k)) − α v
(k)α2
≤ αJ
α(y
α(k), v
α(k)) ≤ αc
1; ∀ k ≥ 0 then (y
α(k))
k≥0is bounded in Y .
Finally, the sequence (y
α(k), v
α(k))
k≥0is bounded in Y × U
N. Hence, one can extract a convergent subsequence which converges to a limit (y
α, v
α).
J
αis continuous on Y × U
N, then J
α(y
α, v
α) ≤ lim inf
k
J
α(y
α(k), v
α(k)) = inf
(y,v)∈C×UN
J
α(y, v) Since C is closed, (y
α, v
α) ∈ C × U
Nand
(y,v)∈C×U
inf
NJ
α(y, v) ≤ J
α(y
α, v
α) Then J
α(y
α, v
α) = inf
(y,v)∈C×UN
J
α(y, v), i.e. (y
α, v
α) is a solution of (P
α).
2- (y
α, v
α) satisfies the following necessary condition
J
α(y
α, v
α)(y − y
α, v − v
α) ≥ 0 ; ∀ (y, v) ∈ C × U
NWe have
J
α(y
α, v
α)(y, v) = 2 α
y
vα,f− y
α, y
v,f− Cφ
Nz
0− R
df
− 2
α y
vα,f− y
α, y
+2 v
α, v
then
J
α(y
α, v
α)(y − y
α, v − v
α) = 2 α
y
vα,f− y
α, y
v,f− y
vα,f− 2
α y
vα,f− y
α, y − y
α+ 2 v
α, v − v
αThe necessary condition becomes
1
α y
vα,f− y
α, y
v,f− y
vα,f− 1
α y
vα,f− y
α, y − y
α+ v
α, v − v
α≥ 0 (24) By replacing respectively in (24), v by v
αand y by y
α, we obtain the inequalities (22) and (23). We deduce the following characterization result.
Proposition 3.3 (y
α, v
α) is characterized by
b
α, y − y
α≥ 0 ; ∀ y ∈ C (25) where b
αis given by
b
α= 1
α (y
α− y
vα,f) (26)
and v
α= (v
α0, · · · , v
αN−1) with
v
kα= B
∗(φ
∗)
N−1−kC
∗b
α; k = 0, ..., N − 1 (27) Proof: The element b
αgiven by (26) appears in inequality (22) of propo- sition 3.2. We have
− 1
α y
α− y
vα,f, y
v,f− y
vα,f+ v
α, v − v
α= −
b
α,
N−1
i=0
Cφ
N−1−iB (v
i− v
αi)
+
N
−1 i=0v
αi, v
i− v
αi= −
N−1i=0
B
∗(φ
∗)
N−1−iC
∗b
α, v
i− v
iα+
N−1
i=0
v
αi, v
i− v
αi=
N
−1 i=0v
αi− B
∗(φ
∗)
N−1−iC
∗b
α, v
i− v
αiBy reporting this relation in (23), we obtain
N−1
i=0
v
αi− B
∗(φ
∗)
N−1−iC
∗b
α, v
i− v
iα≥ 0; ∀ v ∈ U
N(28)
which implies
v
iα= B
∗(φ
∗)
N−1−iC
∗b
α; i = 0, ..., N − 1 (29)
Before proceeding to the limit, we show the following preliminary convergence results.
Proposition 3.4
1- The sequence (y
vα,f− y
α) converges strongly in Y to 0.
2- The sequence J
α(y
α, v
α) is increasing and bounded, then convergent.
3- The sequence (y
α, v
α, b
α) is bounded in Y × U
N× F
N. Proof:
1- Under the C -remediability hypothesis, there exists a control ˆ u = (ˆ u
0, ..., u ˆ
N−1) in U
Nsuch that y
u,fˆ∈ C . By noting ˆ y = y
u,fˆ, we have (ˆ y, u) ˆ ∈ C × U
N, and hence
J
α(y
α, v
α) ≤ J
α(ˆ y, u) = ˆ u ˆ
2(30) then for every α > 0, we have
y
vα,f− y
α2
= αJ
α(y
α, v
α) − α v
α2
≤ αJ
α(y
α, v
α) ≤ α u ˆ
2consequently lim
α−→0
y
vα,f− y
α= 0.
2- i- Using (30), we deduce that the sequence (J
α(y
α, v
α))
α>0is bounded.
ii- For δ ≤ α, we have
J
α(y, v) ≤ J
δ(y, v) ; ∀ (y, v) ∈ C × U
Nconsequently
J
α(y
α, v
α) ≤ J
α(y
δ, v
δ) ≤ J
δ(y
δ, v
δ) then J
α(y
α, v
α) increases when α −→ 0.
3- i- (v
α)
α>0is bounded because
v
α2
≤ J
α(y
α, v
α) ≤ J
α(ˆ y, u) = ˆ u ˆ
2ii- The map v ∈ U
N−→ y
v,f∈ Y is affine and continuous, then there exists c
3> 0 such that
y
v,f− Cφ
Nz
0− R
df ≤ c
3v ; ∀ v ∈ U
NThe sequence (y
vα,f)
α>0is then bounded because (v
α)
α>0is.
This shows that y
α= (y
α− y
vα,f) + y
vα,fis bounded in Y .
iii- Generally, (b
α)
α>0is not bounded in Y but it is bounded in F
N, because v
α2UN
=
N
−1 i=0v
αi2
U
=
N
−1 i=0B
∗(φ
∗)
N−1−iC
∗b
α2
U
=
N
−1 k=0B
∗(φ
∗)
kC
∗b
α2
U
= b
α2FN
To conclude, we show the following result where we give the solution of the initial problem (P
N) and its characterization.
Proposition 3.5 We consider a convergent subsequence of (y
α, v
α, b
α)
α>0, with the same notation. Its limit (y
∗, v
∗, b) is characterized by
1- y
∗= y
v∗,f2- v
∗is the solution of the problem (P
N) given by (14).
3- The sequence J
α(y
α, v
α) is increasing with a limit v
∗2
and the sequence (v
α)
α>0converges strongly to v
∗in U
N.
4- The control v
∗=
v
∗0, ..., v
N∗−1is given by
v
∗k= B
∗(φ
∗)
N−1−kC
∗b ; k = 0, ..., N − 1 (31)
5- The element b is characterized by
b, y − y
v∗,f≥ 0 ; ∀ y ∈ C ∩ (Cφ
Nz
0+ R
df + F
N) (32) Proof:
1- The sequence (v
α)
α>0converges weakly to v
∗and the map v ∈ U
N−→
y
v,f− Cφ
Nz
0− R
df ∈ Y is linear and continuous, then (y
vα,f)
α>0converges weakly to y
v∗,fand hence
y
∗= lim
α−→0
y
α= lim
α−→0
(y
α− y
vα,f) + lim
α−→0
y
vα,f= y
v∗,f2- The set C is closed, then y
v∗,f= lim
α−→0
y
α∈ C . Moreover, if v ∈ U
Nis such that y
v,f∈ C , then
v
α2
= J
α(y
α, v
α) − 1
α y
vα,f− y
α2
≤ J
α(y
α, v
α) ≤ J
α(y
v,f, v) = v
2(33)
i.e. v
α2
≤ v
2, consequently
v
∗2
≤ lim inf
α
v
α2
≤ v
2Then (v
∗) is a solution of (P
N).
3.i- We show in proposition 3.4 that the sequence J
α(y
α, v
α) is increasing when α is decreasing to 0.
ii- Using (33), for any v ∈ U
Nsuch that y
v,f∈ C , we have v
α2
≤ J
α(y
α, v
α) ≤ v
2Particulary for v = v
∗, we obtain
v
α2
≤ J
α(y
α, v
α) ≤ v
∗2
(34)
The weak convergence of (v
α)
α>0to v
∗implies v
∗2
≤ lim inf
α
v
α2
≤ lim
α
J
α(y
α, v
α) ≤ v
∗2
then
α−→0
lim J
α(y
α, v
α) = v
∗2
iii- Using the inequality (34), we have
v
∗2
≤ lim inf
α
v
α2
≤ lim sup
α
v
α2
≤ v
∗2
and then v
α2
converges to v
∗2
. Using the weak convergence of (v
α)
α>0to v
∗, we deduce that (v
α)
α>0converges strongly to v
∗in U
N.
4- Since v
α=
v
α0, ..., v
αN−1with v
αk= B
∗(φ
∗)
N−1−kC
∗b
α; k = 0, ..., N − 1 then for w = (w
0, ..., w
N−1) ∈ U
N, we have
v
α, w =
b
α,
N−1
k=0
Cφ
N−1−kBw
k=
b
α, y
w,f− R
df − Cφ
Nz
0On the other hand y
w,f− R
df − Cφ
Nz
0∈ F
N, then from the weak convergence of (v
α)
α>0to v
∗and that of (b
α)
α>0to b, we deduce that
v
∗, w =
b, y
w,f− R
df − Cφ
Nz
0=
b,
N−1
k=0
Cφ
N−1−kBw
k=
N−1
k=0
B
∗(φ
∗)
N−1−kC
∗b, w
kand hence v
k∗= B
∗(φ
∗)
N−1−kC
∗b ; k = 0, ..., N − 1.
5- The inequality (25) can be written
b
α, y
α≤ b
α, y ; ∀ y ∈ C
i.e.
b
α, y
α− R
df − Cφ
Nz
0≤
b
α, y − R
df − Cφ
Nz
0; ∀ y ∈ C (35) We have
J
α(y
α, v
α) = 1
α y
vα,f− y
α2
+ v
α2
= b
α, y
α− y
vα,f+ v
α2
=
b
α, y
α− R
df − Cφ
Nz
0−
b
α, y
vα,f− R
df − Cφ
Nz
0+ v
α2
=
b
α, y
α− R
df − Cφ
Nz
0−
b
α,
N−1
k=0
Cφ
N−1−kBv
αk+ v
α2
=
b
α, y
α− R
df − Cφ
Nz
0−
N−1
k=0
B
∗(φ
∗)
N−1−kC
∗b
α, v
kα+ v
α2
=
b
α, y
α− R
df − Cφ
Nz
0then
b
α, y
α− R
df − Cφ
Nz
0= J
α(y
α, v
α) Consequently
α−→0
lim
b
α, y
α− R
df − Cφ
Nz
0= lim
α−→0
J
α(y
α, v
α) = v
∗2
=
N
−1 k=0B
∗(φ
∗)
N−1−kC
∗b, v
k∗=
b,
N−1
k=0
Cφ
N−1−kBv
∗k=
b, y
v∗,f− R
df − Cφ
Nz
0For y − R
df − Cφ
Nz
0∈ F
Nand when α −→ 0 in (35) , we obtain b, y
v∗,f− R
df − Cφ
Nz
0≤
b, y − R
df − Cφ
Nz
0; ∀ y ∈ C ∩ (Cφ
Nz
0+ R
df + F
N)
i.e.
b, y − y
v∗,f≥ 0 ; ∀ y ∈ C ∩ (Cφ
Nz
0+ R
df + F
N)
Remark 3.6 From the proof, we deduce that θ
f= b, u
θf= v
∗and y
v∗,f= Λ
Nθ
f+ Cφ
Nz
0+ R
df.
4 Extension to the regional case
4.1 Problem statement
In this part, we study the possibility to reduce the effect of a disturbance in a region ω ⊂ Ω at the final step N . The regional aspect of the considered problem is motivated by the fact that a system can be C -remediable in a region ω ⊂ Ω but not on the whole domain Ω, and even if it’s C -remediable in Ω, the cost is reduced if we are interested only by a subregion ω ⊂ Ω [3, 2]. Moreover, the C -remediability in a region ω ⊂ Ω is a notion more general than the C - remediability in the whole domain Ω which corresponds to ω = Ω [1].
We consider the disturbed system described by the discrete state equation (S
d), defined by (1), with X = L
2(Ω). The system (S
d) is augmented by the following regional output equation
(E
dω) y
ωk= Ci
ωp
ωz
k; k = 1, · · · , N (36) where p
ωis the restriction operator defined by:
p
ω: L
2(Ω) −→ L
2(ω)
z −→ p
ωz = z
|ω(37)
and i
ω= p
∗ωis the adjoint operator of p
ω. It is defined by i
ω: L
2(ω) −→ L
2(Ω)
ϕ −→ i
ωϕ =
ϕ in ω 0 otherwise
(38) Let z
u,fbe the state of (S
d) at the final step N . The corresponding obser- vation is given by
y
u,fω= Ci
ωp
ωφ
Nz
0+
N−1
i=0
Ci
ωp
ωφ
N−1−iBu
i+
N−1
i=0
Ci
ωp
ωφ
N−1−if
i(39)
Obviously, if f = 0 and u = 0, y
0,0ω= Ci
ωp
ωφ
Nz
0, the observation is normal.
But, if f = 0, generally y
0,fω= Ci
ωp
ωφ
Nz
0.
Then we introduce a control term Bu in order to reduce regionally the effect of this disturbance at the final step N by bringing the regional observation in a given zone of tolerance C , i.e. such that
For any f = (f
0, · · · , f
N−1)
tr∈ (L
2(Ω))
N, there exists u = (u
0, · · · , u
N−1)
tr∈ U
Nsatisfying
y
u,fω∈ C
where C is a convex, closed and nonempty subset of Y and y
ωu,fis given by (39). This leads to the following definition.
Definition 4.1
i) A disturbance f ∈ (L
2(Ω))
Nis said C -ω − remediable if there exists u ∈ U
Nsuch that
y
u,fω= Ci
ωp
ωφ
Nz
0+ K
dωu + R
ωdf ∈ C where K
dωand R
ωdare the operators defined respectively by
K
dωu = Ci
ωp
ωH
du =
N−1
i=0
Ci
ωp
ωφ
N−1−iBu
i(40) and
R
ωdf = Ci
ωp
ωH
df =
N−1
i=0
Ci
ωp
ωφ
N−1−if
i(41) ii) The system (1) augmented by (36) is said to be C -ω − remediable if any disturbance f ∈ X
Nis C -ω − remediable.
Let us note that if C =
Ci
ωp
ωφ
Nz
0, then we have a problem of regional exact remediability.
4.2 Cheap regional enlarged exact remediability
Let f ∈ (L
2(Ω))
N, we consider the following problem (P
Nω) of enlarged exact
remediability with minimum energy
where
J(u) = u
2UNand U
adω=
u ∈ U
N/y
u,fω∈ C
We suppose that the disturbance f is C -ω − remediable, then U
adω= ∅ and the problem (P
Nω) is well defined and admits a unique solution v
∗characterized by
v
∗, v − v
∗≥ 0 ; ∀ v ∈ U
adωAs in the previous case, the problem (P
Nω) is resolved using an extension of the HUM approach and a penalization method. For θ ∈ Y
≡ Y , let
θ
FωN
= (
N−1
k=0
B
∗(φ
∗)
ki
ωp
ωC
∗θ
2U
)
12(42)
·
FωN
is a semi-norm. If ker(p
ωC
∗) = 0, then ·
FωN
is a norm on Y if and only if (S
d) + (E
dω) is ω − weakly remediable. Under this condition, we consider the space
F
Nω= Y
·F ωNF
Nωis a Hilbert space with the inner product
θ, σ
FωN
=
N−1
k=0
B
∗(φ
∗)
ki
ωp
ωC
∗θ, B
∗(φ
∗)
ki
ωp
ωC
∗σ
U
and the operator Λ
ωNdefined on Y by
Λ
ωNθ =
N−1
k=0
Ci
ωp
ωφ
kBB
∗(φ
∗)
ki
ωp
ωC
∗θ
has a unique extension as an isomorphism from F
Nω−→ (F
Nω)
such that Λ
ωNθ, σ
Y= θ, σ
FωN