Estimation for Linear Repetitive Processes based on Event-triggering

Mechanism under DoS Attack

Xinye Jiang 1, Dehao Li 1,2, Chengsheng Li 1
1.Jinling Institute of Technology, Nanjing 211169, China. 2.Department of Mathematics, Southeast University, Nanjing 210096, China
E-mail: [email protected]

Abstract: This paper addresses the state estimation of linear repetitive processes affected by DoS attacks. An esti-
mator with H∞ performance is designed based on an event-triggering mechanism. Firstly, the linear repetition process
2024 36th Chinese Control and Decision Conference (CCDC) | 979-8-3503-8778-0/24/$31.00 ©2024 IEEE | DOI: 10.1109/CCDC62350.2024.10588338

is two-dimensional (2-D) system, and a 2-D DoS attack and event triggering mechanism model is established. Then
designed an estimator, on which we built the augmentation error system and specified the desired performance metrics
to ensure that the system would remain robustness in the face of DoS attacks and event triggering mechanisms. By using
Lyapunov theory, the sufficient criterion to ensure the robust stability of augmented error system is provided. The explicit
expressions of the estimator’s parameters are obtained by solving a matrix inequality. Finally, a simulation example is
presented to validate the feasibility of the theoretical results.
Key Words: Linear repetitive processes, DoS attack, event-triggering mechanism, estimation

1 INTRODUCTION State estimation, a method used to gauge the internal state

of dynamic systems based on available measurement data,
Linear repetitive processes constitute a distinctive sub- plays a crucial role in fault diagnosis, fault tolerance con-
set within 2-D systems, holding significant applications trol, monitoring, and related fields. As a key component in
in learning iterative control, longwall coal cutting, and control theory, particularly when integrated with an event-
metal rolling operations [1]. However, the integration of triggered mechanism, has garnered significant attention in
networking in recent years has led to escalated security recent years. In the literature [7], the fusion estimation
risks, encompassing spoofing attacks, Man-in-the-Middle problem for a class of discrete time-varying multi-rate lin-
attacks, and others. The inherent complexity and multi- ear repetitive processes is addressed using the weighted try-
dimensionality of linear repetitive processes heighten their once-discard protocol. Additionally, [8] explores output-
susceptibility to external interference and malicious intru- feedback-based proportional integral derivative control and
sions. Denial-of-Service (DoS) attacks, as a malicious in- filtering problems for the discrete-discrete Roesser system
trusion method, present a substantial threat to the stabil- with disturbances. Notably, current studies primarily fo-
ity and performance of these systems. Despite notable cus on one-dimensional systems, leaving research on 2-D
progress made by researchers, both domestic and interna- systems relatively limited.
tional, in the study of DoS attacks, the dynamic evolution This paper delves into H∞ -based state estimation for
of these attacks within linear repetitive processes has be- linear repetitive processes under DoS attacks, with the ob-
come notably intricate [2, 3, 4]. This evolving complexity jective of addressing the following three challenges: 1) Es-
poses formidable challenges for research, leaving numer- tablishing a linear repetitive process under DoS attack with
ous unresolved issues within this domain. an event-triggered effect. 2) Employing the matrix inequal-
To enhance defense against attacks, an introduced event- ity method to investigate the estimation problem of linear
trigger mechanism operates as an adaptive strategy for data repetitive processes under DoS attack. 3) Designing an ef-
acquisition and control. This mechanism intelligently se- fective estimator such that the estimation error system sat-
lects the optimal time for data transmission while mini- isfies the H∞ performance constraint in the mean square
mizing unnecessary communication overhead. By setting sense.
precise trigger conditions, it effectively reduces the impact
2 PROBLEM FORMULATION
of DoS attacks on the system, thereby improving overall
stability. Research efforts, exemplified in [5] delving into Consider a linear repetitive process of the following
event-triggered control under DoS attack conditions. Ad- form:
ditionally, [6] proposes two distinct event-triggered control ⎧
(ETC) schemes aimed at stabilizing K-valued impulsive ⎪
⎪ x(l + 1, m + 1) = A  1 x(l + 1, m) + B  1 y (l, m)
⎪
⎪
logical control networks (ILCNs) toward predefined state ⎪
⎨  1ω
+D  1 (l + 1, m)
sets. However commendable these strides, there are chal- (1)
⎪
⎪ 
y (l + 1, m) = A2 x(l + 1, m) + B  2 y (l, m)
lenges in this domain necessitate further investigation and ⎪
⎪
⎪
⎩
resolution. +D  ω (l + 1, m)
2 2

979-8-3503-8778-0/24/$31.00 ©2024 IEEE 4558

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on January 16,2025 at 15:44:06 UTC from IEEE Xplore. Restrictions apply.
 where, for l ∈ 0, ∞), and m ∈ 0,  is the pass length. In the transmission channel output to the estimator, the
x(l, m) ∈ Rn is the state vector, y (l, m) ∈ Rm is the pass signal is subjected to DoS attack, and the system output
 1 (l, m) ∈ Rq is the white process noise
profile vector, and ω signal after the attack becomes z̄(l, m). In the DoS attack,
with mean zero and covariance Q(l, m), A 1, A2, B
 1, B
 2, the output signal is disconnected and the data cannot be
 
D1 , D2 is the real constant matrix. The initial boundary transmitted; when the DoS attack is dormant, the output
conditions of system (1) satisfy the following formula: signal is not disturbed and the data is transmitted normally.
After a DoS attack, the signal output changes to:
E{x(l + 1, 0)} = f (l + 1), E{y (0, m)} = g(m) (2)
z̄(l, m) = α(l, m)z(l, m)
(9)
where, f (l + 1) ∈ Rn and g(m) ∈ Rm are known vectors. = ᾱz(l, m) + α̃(l, m)z(l, m)
The output model before transmission through the network
is defined on l ∈ 0, ∞) and m ∈ 0, , as follow: where α̃(l, m) = α(l, m) − ᾱ, and the random vari-
able α(l, m) is Bernoulli’s white noise sequence of
z(l + 1, m) = L  2 y (l, m)
 1 x(l + 1, m) + L (3) values 0 or 1, and has P rob{α(l, m) = 1} = ᾱ,
P rob{α(l, m) = 0} = 1 − ᾱ , where α ∈ [0, 1] is the
where z(l, m) ∈ Rq is the observed output, ω  2 (l, m) ∈ Rq known constant.
is the observed noise, L 1 and L 2 is the real constant matrix.
In addition, this paper also needs a vector to be estimated, Hypothesis 1 The system is not affected by external dis-
let the vector to be estimated is v(l, m). turbance signals except DoS attack signals during channel
transmission.
v(l + 1, m) = L  4 y (l, m)
 3 x(l + 1, m) + L (4) In order to reduce the system damage caused by DoS
attacks, the ETM is triggered. This mechanism is triggered
where v(l, m) ∈ Rq is the output to be estimated, L  3 and
to compensate the system damage caused by DoS attacks

L4 is the real constant matrix. to a certain extent. When DoS attack signals enter the 2-D
To facilitate later analysis, system (1) is to be trans- event trigger, the data that does not exceed the threshold is
formed into a general Fornasini-Machesini (F-M) second filtered. The event trigger sequence available by the defined
state-space model.   ordered sequence can be expressed as:
By defining x(l + 1, m) = xT (l + 1, m) y T (l, m) ,
 T  T
ω(l, m) = ω  2T (l, m) , for l ∈ 0, ∞) and
 1 (l, m) ω (2, 1) = (l, m) < · · · < (ls , ms ) < · · · (10)
m ∈ 0, , we obtain from (1)-(4) that
where, (ls , ms )(s ∈ 1, ∞)) indicates that the fourth event
⎧ s is triggered. Define the 2-D event trigger error as follows:
⎪ x(l + 1, m + 1) = A1 x(l + 1, m) + A2 x(l, m + 1)
⎪
⎪
⎨ + D1 ω(l + 1, m) + D2 ω(l, m + 1) e(l, m) = z̄(ls , ms ) − z̄(l, m) (11)
⎪z(l, m) = Cx(l, m)
⎪
⎪
⎩ where, z(ls , ms ) indicates the information received at the
v(l, m) = Hx(l, m)
(5) last trigger time (ls , ms ), and z(l, m) is the current obser-
where vation signal. Define the event generator as follows:
  eT (l, m)Ωe(l, m) > δz T (l, m)Ωz(l, m)
A1 B1 1
D   (12)
A1 = , D1 = , C= L 1 L 2 ,
0 0 0 where Ω is the undetermined positive definite weighted ma-
 
0 0 0   trix and δ ∈ [0, 1] is the assumed scalar. It should be
A2 =   
A2 B  2 , D2 = D  2 , H = L 3 L4 . noted that although the 2-D ETM can be regarded as the
extension of the 1-D ETM, compared with the 1-D system
with x(l, m) ∈ Rn+m represents the state vector, etm, there are still differences in the transmission state up-
ω(l, m) ∈ R2q is an external disturbance of the system date law of 2-D event triggering. Assuming that the current
belonging to l2 {0, ∞), 0, )}. time is (l, m) and the real transmission signal is zt (l, m),
The initial conditions of a 2-D F-M model satisfy the the transmission state update law is as follows:
following inequality constraints:
zt (l, m) = zt (l − 1, m − 1) if T rans(l, m) = 0
∞ 
2 2 zt (l, m) = z(l, m) if T rans(l, m) = 1
lim E x(l, 0) + x(0, m) < ∞. (6)
K→∞ (13)
l=0 m=0
where T rans(l, m) is the function of whether the current
The boundary conditions of system (6) are: state will be transmitted, which can be written as
 ⎧
x(l, 0) ⎪
⎪ T rans(l, m) = 0
x(l, 0) = , l ∈ 1, ∞) (7) ⎪
⎪
y (l − 1, 0) ⎨ if Ωe(l, m) ≤ δz T (l, m)Ωz(l, m)
 (14)
⎪
⎪ T rans(l, m) = 1
x(1, m) ⎪
⎪
x(1, m) = , m ∈ 0,  (8) ⎩
y (0, m) if eT (l, m)Ωe(l, m) > δz T (l, m)Ωz(l, m)

2024 36th Chinese Control and Decision Conference (CCDC) 4559

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on January 16,2025 at 15:44:06 UTC from IEEE Xplore. Restrictions apply.
 In order to get the state estimation of the Fornasini- If this is true, then the augmented estimation error system
Machesini(F-M) model under event triggering, we con- (16) is said to be robust H∞ stable and has a disturbance
struct an estimator of the following form: decay level γ, where
⎧ 
∞ h
  1 ∞
 
⎪ x̂(l + 1, m + 1)
⎪
⎪ ṽ2E = E ṽ (l, m)2 + E ṽ (l, 0)2
⎪
⎪ = A1 x̂(l + 1, m) + A2 x̂(l, m + 1) 2
⎪
⎪ m=1
l=1 l=1
⎪
⎨
1  
h
+ K1 (z̄(l + 1, m) − C x̂(l + 1, m)) 
(15) + E ṽ (0, m)2 ,
⎪
⎪ + K2 (z̄(l, m + 1) − C x̂(l, m + 1)) 2 m=1
⎪
⎪
⎪
⎪ 
∞ 
1
+ K1 Ce(l + 1, m) + K2 Ce(l, m + 1) h ∞
⎪
⎪
⎩ ω22 = ω (l, m)2 + ω (l, 0)2
v̂(l, m) = H x̂(l, m) 2
l=1 m=1 l=1

where, l ∈ 2, ∞), m ∈ 1,  , x̂(l, m) ∈ R n+m

repre- 1 
h
+ ω (0, m)2 .
sents the state of the estimator, v̂(l, m) ∈ Rnv represents 2 m=1
the estimated signal of the output vector v(l, m) to be esti-
mated. K1 and K2 the estimator parameters to be designed. The purpose of this paper is to design a suitable estimator
Δ for the F-M model (5), so that the augmented estimation
The construction error term is f (l, m) = x(l, m) −
 T T error system is robust H∞ stable when the attenuation level
x̂(l, m). By defining η (l, m) = x (l, m) f T (l, m) , is γ, also known as the H∞ state estimation problem for
 T
ω (l, m) = ω1T (l, m) ω2T (l, m) , the augment error two-dimensional systems.
system of the F-M model can be derived as follows:
3 MAIN RESULTS
⎧
⎪ η(l + 1, m + 1) = (A1 + α̃(l + 1, m)Ā1 )η(l + 1, m)
⎪
⎪ Before introducing the main results, the following lemma
⎪
⎪ + (D1 + α̃(l + 1, m)D̄1 )ω(l + 1, m)
⎪
⎪ is given.
⎪
⎨ + (A2 + α̃(l, m + 1)Ā2 )η(l, m + 1) Lemma 1 [10] Let X, Y , U and V be real matrices of
⎪
⎪ + (D2 + α̃(l, m + 1)D̄2 )ω(l, m + 1) appropriate dimensions with X satisfying X = X T . For
⎪
⎪
⎪
⎪ + B1 e(l + 1, m) + B2 e(l, m + 1) all V T V ≤ I, the inequality X + U V Y + Y T V T U T < 0
⎪
⎪
⎩ holds if and only if there exists a positive scalar ε such that
ṽ(l, m) = H̃η(l, m)
X + εU U T + ε−1 Y T Y < 0.
(16)
where Assume that there is no external signal in the system
  (5), i.e.ω(·, ·) ≡ 0 . Then, the augmented estimation error
A1 0 0 0
A1 = , D̄1 = , system (16) becomes:
(1 − ᾱ)K1 C A1 − K1 C 0 −K1
 
A2 0 0 0 η(l + 1, m + 1) = (A1 + α̃(l + 1, m)Ā1 )η(l + 1, m)
A2 = , D̄2 = ,
(1 − ᾱ)K2 C A2 − K2 C 0 −K2 + (A2 + α̃(l, m + 1)Ā2 )η(l, m + 1)
 
D1 =
D1 0
, D2 =
D2 0
, + B1 e(l + 1, m) + B2 e(l, m + 1).
D1 −ᾱK1 D2 −ᾱK2 (19)
 
0 0 0 0
Ā1 = , Ā2 = , Theorem 1 Under the Hypothesis 1, for the given filter pa-
−K1 C 0 −K2 C 0 rameter K1 and K2 , when ω(·, ·) ≡ 0, if there is a positive
 
0 0   definite matrix P , Q ∈ R2(n+m)×2(n+m) satisfies P > Q
B1 = , B2 = , H̃ = 0 H .
K1 C K2 C and 
χ11 χ12
For the convenience of subsequent proof, we provide the <0 (20)
∗ χ22
following definition.
then the augmented estimation error system (16) is robustly
Definition 1 For any given initial condition (2), at that stable in the mean square sense, where
time ω(·, ·) ≡ 0, if:
X11 = −diag{Q, P − Q, Ω, Ω},
lim E η(l, m)2 = 0. (17) X22 = −diag{P −1 , P −1 , P −1 , Ω −1 , Ω −1 },
l+m→∞
  √ T
G11 G12 A1 α̂Ā1 √ 0
It is said that the augmented error system (16) is robustly X12 = , G11 = ,
G21 0 A2 0 α̂ĀT2
stable in the sense of mean square.    T
δ̃ C̄ T  0 B 0 0
Definition 2 For a given scalar γ > 0, if the following two G12 = , G21 = 1T ,
conditions are true: (1) When ω(·, ·) ≡ 0 , the augmented 0 δ̃ C̄ T B2 0 0
estimation error system (16) is robustly stable in the mean α̂ = ᾱ(1 − ᾱ), δ̃ = δ(α̂ + ᾱ2 ).
square sense; (2) When ω(·, ·) ∈ l2 {0, ∞), 0, } and
the initial boundary is always zero, inequality: Proof: Consider the following class Lyapunov function

ṽ22 ≤ γ 2 ω22 (18) V (l, m) = V1 (l, m) + V2 (l, m) (21)

4560 2024 36th Chinese Control and Decision Conference (CCDC)

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on January 16,2025 at 15:44:06 UTC from IEEE Xplore. Restrictions apply.
 with s14 = (A1 + α̃(l + 1, m)Ā1 )T P B2 ,

V1 (l, m) = η T (l, m)Qη(l, m), s23 = (A2 + α̃(l, m + 1)Ā2 )T P B1 ,

V2 (l, m) = η T (l, m)(P − Q)η(l, m), s24 = (A2 + α̃(l, m + 1)Ā2 )T P B2 ,

s33 = B1T P B1 , s34 = B1T P B2 , s34 = B2T P B2 .
where P , Q are positive definite matrices to be deter-
mined with P − Q > 0. Let ξ(l, m) = [η T (l + Triggered by the event condition (13), can be obtained
1, m), η T (l, m + 1)]T and ℵ(l, m) = {η(l, m + 1), η(l − eT (l, m)Ωe(l, m) > δz̄ T (l, m)Ωz̄(l, m)
1, m + 1), . . . , η(0, m + 1), η(l + 1, m), η(l + 1, m − (24)
1), . . . , η(l + 1, 0)}. Consider the following index: = (α̂ + ᾱ2 )η T (l, m)C̄ T ΩC̄η(l, m)
 
J(l,m)  E{V (l+1, m+1) where C̄ = 0 C , ξ T (l, m)Λξ(l, m) < 0, Λ =
− V1 (l+1, m) − V2 (l, m+1)|ℵ(l, m)} (22) diag{δ̃ C̄ T ΩC̄, δ̃ C̄ T ΩC̄, −Ω, −Ω}. Based on (25) Ξ(l +
  1, m + 1), we can get
= E{ΔV1 (l, m)|ℵ(l, m) + E ΔV2 (l, m)|ℵ(l, m)} ⎡ ⎤
s̄11 s12 s13 s14
where, ΔV1 (l, m) = V1 (l + 1, m + 1) − V1 (l + 1, m), ⎢ 0 s̄22 s23 s24 ⎥
Ξ̄(l + 1, m + 1) = ⎢ ⎣ 0
⎥ (25)
ΔV2 (l, m) = V2 (l + 1, m + 1) − V2 (l, m + 1).Then, along 0 s̄33 s34 ⎦
the solution of the augmented error system (16), we have 0 0 0 s̄44
E{V1 (l + 1, m + 1) + V2 (l + 1, m + 1)|ℵ(l, m)}
where s̄11 = s11 + δ̃ C̄ T ΩC̄, s̄22 = s22 + δ̃ C̄ T ΩC̄,
= E{η T (l + 1, m + 1)P η(l + 1, m + 1)|ℵ(l, m)} s̄33 = s33 − Ω, s̄44 = s44 − Ω. It is easy to verify
= E{η T (l + 1, m)(A1 + α̃(l + 1, m)Ā1 )T that E{α̃(l, m)} = 0, E{α̃2 (l, m)} = ᾱ(1 − ᾱ) = α̂. Ac-
× P (A1 + α̃(l + 1, m)Ā1 )η(l + 1, m) cording to the statistical properties of random variables and
Schur’s complement lemma, it can be seen that if matrix in-
+ 2η T (l + 1, m)(A1 + α̃(l + 1, m)Ā1 )T P B1 e(l + 1, m) equality (20) holds, then Ξ̄(l + 1, m + 1) ≤ 0 holds. More
+ 2η T (l + 1, m)(A1 + α̃(l + 1, m)Ā1 )T P B2 e(l + 1, m) specifically, there is a scalar ρ that causes
+ 2η T (l, m + 1)(A2 + α̃(l, m + 1)Ā2 )T P B1 e(l, m + 1) Ξ̄(l + 1, m + 1)
+ 2η T (l, m + 1)(A2 + α̃(l, m + 1)Ā2 )T P B2 e(l, m + 1)  T  
η(l + 1, m) −ρI 0 η(l + 1, m)
+ eT (l + 1, m)B1T P B1 e(l + 1, m) ≤ (26)
η(l, m + 1) ∗ 0 η(l, m + 1)
+ 2eT (l + 1, m)B1T P B2 e(l, m + 1) = −ρη(l + 1, m)2 .
T T
+ η (l, m + 1)(A2 + α̃2 (l, m + 1)Ā1 )
Find the mathematical expectation on both sides of equa-
× P (A2 + α̃(l, m + 1)Ā2 )η(l, m + 1) tion (26), and you get
+ 2η T (l + 1, m)(A1 + α̃(l + 1, m)Ā1 )T
E{V (l + 1, m + 1) − V1 (l + 1, m) − V2 (l, m + 1)}
× P (A2 + α̃(l, m + 1)Ā2 )η(l + 1, m)
2
≤ −ρE{η(l + 1, m) }.
+ eT (l, m + 1)B2T P B2 e(l, m + 1)|ℵ(l, m)}
(27)
T T
Let ξ(l, m)  [η (l + 1, m) η (l, m + 1) T
e (l + Similar to the proof of theorem 1 in [9], we concludes
1, m) eT (l, m + 1)]T , there we have that system (19) is robustly stable in the mean square when
ω(·, ·) ≡ 0 , and the proof of Theorem 1 is complete.
J(l,m) = E{ξ T (l, m)Ξ(l + 1, m + 1)ξ(l, m)|ℵ(l, m)}
Theorem 2 Under Hypothesis 1, let the parameters K1 ,
(23)
K2 , γ > 0 be given. The augmented system (5) achieves
where the H∞ performance if there exist positive definite matrices
⎡ ⎤ P , Q > 0 such that P > Q and
s11 s12 s13 s14 
⎢ 0 s22 s23 s24 ⎥ Y11 Y12
Ξ(l + 1, m + 1) = ⎢
⎣ 0
⎥
∗ Y22
<0 (28)
0 s33 s34 ⎦
0 0 0 s44 hold uniformly for all l ∈ 1, ∞) and m ∈ 0, , where
with Y11 = −diag{Q, P − Q, Ω, Ω, γ 2 I, γ 2 I},
s11 = (A1 + α̃(l + 1, m)Ā1 )T P Y22 = −diag{P −1 , P −1 , P −1 , Ω −1 , Ω −1 , I, I},
  T
× (A1 + α̃(l + 1, m)Ā1 ) − Q, G11 G12 R13 H̃ 0
Y12 = , R13 = ,
R21 0 0 0 H̃ T
s22 = (A2 + α̃(l, m + 1)Ā2 )T P ⎡ T ⎤
× (A2 + α̃(l, m + 1)Ā2 ) − P + Q, B1 0 0
⎢ B2T
0 ⎥
s12 = (A1 + α̃(l + 1, m)Ā1 )T P (A2 + α̃(l, m + 1)Ā2 ), R21 = ⎢ √0 T ⎥.
⎣D1T α̂D̄1 √ 0 ⎦
s13 = (A1 + α̃(l + 1, m)Ā1 )T P B1 , D2T 0 α̂D̄2T

2024 36th Chinese Control and Decision Conference (CCDC) 4561

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on January 16,2025 at 15:44:06 UTC from IEEE Xplore. Restrictions apply.
 The remaining parameters that are not specified are de- thus have
fined in Theorem 1.
E{η T (l + 1, m + 1)P η(l + 1, m + 1 |ℵ(l, m))}
Proof: To analyze the H∞ performance of system (16), let 
us consider the following index: ≤ E ξ T (l, m) diag {Q, P − Q} ξ (l, m)

T 2 T −v̆ T (l, m)v̆(l, m) + γ 2 ω̄ T (l, m)ω̄ (l, m) |ℵ (l, m) .
I(l,m) E{v̆ (l, m)v̆(l, m) − γ ω̆ (l, m)ω̆(l, m)
+ J(l,m) |ℵ(l, m)} (29) Similar to the proof of theorem 1 in [9], we know that
ṽ22 ≤ γ 2 ω22 is always true for all non-zero sequences
where v̆(l, m)  [ṽ T (l + 1, m) ṽ T (l, m)]T , ω̆(l, m)  ω(l, m), and the theorem is proved.
[ω T (l + 1, m) ω T (l, m + 1)]T , and J (l, m) is defined in Theorem 3 Under Hypothesis 1, let the scalar γ > 0 be
(23). Then, along the solution of system (16), we have given. The augmented system (16) is robustly stable in the
mean square sense and achieves the H∞ performance for
I(l,m) all external disturbance ω(·, ·), estimator parameters K̂1 ,
= E{η T (l + 1, m)HT Hη(l + 1, m) + η T (l, m + 1) K̂2 and positive definite matrices P  diag{P1 , P2 }, Q 
diag{Q1 , Q2 } > 0, such that P > Q and
× HT Hη(l, m + 1) − γ 2 ω T (l + 1, m)ω(l + 1, m) 
− γ 2 ω T (l, m + 1)ω(l, m + 1) + J(l,m) |ℵ(l, m)} χ̃11 χ̃12
< 0. (30)
∗ χ̃22
= E{ξ¯T (l, m)Θσ(l+1,m+1) ξ(l,
¯ m)|ℵ(l, m)}
Therefore, the gain of the estimator is
¯ m)  [η T (l + 1, m) η T (l, m + 1) eT (l +
with ξ(l,   −1 
Ki 0 P 0 K̂i 0
1, m) eT (l, m + 1) ω T (l + 1, m) ω T (l, m + 1)]T , and = 2 , i = 1, 2 (31)
Ĥ 0 0 I Ĥ 0
⎡ ⎤
S11 s12 s13 s14 s15 s16 where
⎢ ∗ S22 s23 s24 s25 s26 ⎥
⎢ ⎥ χ̃11 = −diag{M1 , M2 },
⎢ ∗ ∗ s33 s34 s35 s36 ⎥
⎢
Θσ(l+1,m+1) = ⎢ ⎥,
∗ s44 s45 s46 ⎥ M1 = diag{Q1 , Q2 , P1 − Q1 , P2 − Q2 , ΩI, ΩI},
⎢ ∗ ∗ ⎥
⎣ ∗ ∗ ∗ ∗ s55 s56 ⎦ M2 = diag{γ 2 I, γ 2 I, γ 2 I, γ 2 I},
∗ ∗ ∗ ∗ ∗ s66 χ̃22 = −diag{P1 , P2 , P1 , P2 , P1 , P2 , Ω, Ω, I, I},
⎡ ⎤
with N11 N12 N13
χ̃12 = ⎣N21 0 0 ⎦,
S11 = (A1 + α̃(l + 1, m)Ā1 )T P (A1 + α̃(l + 1, m)Ā1 ) N31 N32 0
⎡ T ⎤
− Q + H̃ T H̃, A1 P1 (1 − ᾱ)C T K̂1
⎢ 0 AT1 P2 − C T K̂1 ⎥
S22 = (A2 + α̃(l, m + 1)Ā2 )T P (A2 + α̃(l, m + 1)Ā2 ) N11 = ⎢ ⎥
⎣AT P1 (1 − ᾱ)C T K̂2 ⎦ ,
2
− (P − Q) + H̃ T H̃,
0 AT2 P2 − C T K̂2
s15 = (A1 + α̃(l + 1, m)Ā1 )T P (D1 + α̃(l + 1, m)D̄1 ), ⎡ T ⎤
D1 P1 D1T P2

s16 = (A1 + α̃(l + 1, m)Ā1 )T P (D2 + α̃(l, m + 1)D̄2 ), 0 C T K̂1 ⎢ 0 −ᾱK̂1 ⎥
N21 = , N31 = ⎢ ⎣
⎥,
s25 = (A2 + α̃(l, m + 1)Ā2 )T P (D1 + α(l + 1, m)D̄1 ),
T
0 C K̂2 D2 P1 D2T P2 ⎦
T

0 −ᾱK̂2
s26 = (A2 + α̃(l, m + 1)Ā2 )T P (D2 + α̃(l, m + 1)D̄2 ), ⎡ ⎤
0 √0 0 0
s35 = B1T P (D1 + α̃(l + 1, m)D̄1 , ⎢0 − α̂K̂1 0 ⎥
0
s36 = B1T P (D1 + α̃(l, m + 1)D̄2 , N32 =⎢⎣0
⎥,
⎦
0 0 √0
s45 = B2T P (D1 + α̃(l + 1, m)D̄1 , 0 0 0 − α̂K̂2
⎡ √ T ⎤
s46 = B2T P (D1 + α̃(l, m + 1)D̄2 , 0 − α̂C K̂1 0 0
⎢0 0 0 ⎥
s55 = (D1 + α̃(l + 1, m)D̄1 )T P (D1 + α̃(l + 1, m)D̄1 ) N12 =⎢ √ 0 ⎥,
⎣0 0 0 − α̂C T K̂2 ⎦
− γ 2 I, 0 0 0 0
s56 = (D1 + α̃(l + 1, m)D̄1 )T P (D2 + α̃(l, m + 1)D̄2 ), ⎡ ⎤
 0 0 0 0
s66 = (D2 + α̃(l, m + 1)D̄2 )T P (D2 + α̃(l, m + 1)D̄2 ) ⎢ ⎥
⎢ δ̃C T Ω 0 HT 0 ⎥
N13 =⎢ ⎥.
− γ 2 I. ⎣ 0  0 0 0 ⎦
0 δ̃C T Ω 0 HT
Using the same idea as Theorem 1, based on Schur com-
plement, we obtain that Proof: Based on Theorem 2, pre- and post- multiplying
(30) by the following symmetric matrix
E{ξ T (l, m)Θσ(l+1,m+1) ξ(l, m)|ℵ(l, m)} < 0, diag{I, I, I, I, I, I, P, P, P, Ω, Ω, I, I}

4562 2024 36th Chinese Control and Decision Conference (CCDC)

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on January 16,2025 at 15:44:06 UTC from IEEE Xplore. Restrictions apply.
 then equation (30) can be equivalently converted to derived, thereby endowing the system with robust H∞ per-
 formance. The gain matrix of the estimator is determined
Y11 Ȳ12 through the solution of matrix inequalities. Subsequently,
<0 (32)
∗ Ȳ22 the efficacy of the proposed scheme is validated through
a simulation example. This paper demonstrates that the
where Ȳ22 = diag{−I3 ⊗ P, −I2 ⊗ Ω, −I2 ⊗ I}, Ȳ12 = event-triggered mechanism can effectively mitigate the im-
Y12 diag{I3 ⊗ P, I2 ⊗ Ω, I2 ⊗ I}. Therefore, the display pact of DoS attacks on linear repetitive processes, offering
expression of the estimator is Ki = P2−1 K̂i , i = 1, 2. a novel approach to enhance the security and control per-
4 NUMERICAL EXAMPLE formance of 2-D systems. In future work, the exploration
will extend to linear repetitive processes with randomness,
In order to demonstrate the effectiveness of the designed aiming to establish an estimation system resilient against
estimator based on event triggering mechanism, a numeri- diverse network attacks.
cal example is given in this section to verify the effective-
ness of the scheme. Consider a linear repeating process REFERENCES
with the following parameters:
[1] E. Rogers, K. Galkowski and D. H. Owens, Control sys-
   
 1 = 0.34 0.43 , D
A  1 = 0.16 0.2 T ,
 1 = 0.05, L tems theory and applications for linear repetitive processes,
    Vol. 349, Berlin, Germany: Springer-Verlag, 2007.
 2 = −0.64 −0.2 , D
A  2 = −0.02, L 2 = −0.4 −0.1 T , [2] Y. Zhang and Z.-G. Wu, Asynchronous control of Markov
 3 = 0.65, L
L  4 = −0.12, . jump systems under aperiodic DoS attacks, IEEE Transac-
tions on Circuits and Systems II: Express Briefs, Vol. 70,
Let’s take ᾱ = 0.2, given the attenuation level γ = 1, No. 2, pp. 685-689, 2023.
using the above parameters, an estimator of form (15) is [3] S. Zhu, J. Lu, D.W.C. Ho and J. Cao, Minimal control nodes
designed to be a linear repetitive process (1) under DoS at- for strong structural observability of discrete-time iteration
tack. By solving the matrix inequality (36) in Theorem 3 systems: Explicit formulas and polynomial-time algorithms,
in Matlab, the gain matrix of the corresponding state esti- IEEE Transactions on Automatic Control, to be published,
mator can be given according to equation (37) doi: 10.1109/TAC.2023.3330263.
    [4] X. Li, P. Shi and W. Zhang, Adaptive event-triggered based
−0.1139 −0.0241 0.0183 0.0613 fault detection filtering for unmanned surface vehicles un-
K1 = , K2 = .
0.2100 0.0445 −0.0533 −0.1665 der DoS attacks, Journal of the Franklin Institute, Vol. 360,
No. 17, pp. 13051-13079, 2023.
where the optimal value of the H∞ performance index γ = [5] C. Peng, J. Wu and E. Tian, Stochastic event-triggered H∞
0.1288. Due to the limitation of space in this paper, a brief control for networked systems under denial of service At-
tacks, IEEE Transactions on Systems, Man, and Cybernetics:
Systems, Vol. 52, No. 7, pp. 4200-4210, 2022.
[6] X. Liu, B. Li, J. Zhong, D. Li and L. Lin, Event-triggered set
stabilization of impulsive logical control networks based on
the hybrid index model approach, Communications in Non-
0.5
linear Science and Numerical Simulation, Vol. 119, 107085,
2023.
[7] Y. Shen, Z. Wang, B. Shen, F. E. Alsaadi and F. E. Alsaadi,
Fusion estimation for multi-rate linear repetitive processes
η(l, m)

under weighted Try-Once-Discard protocol, Information Fu-

sion, Vol. 55, pp. 281-291, 2020.
20 [8] N. Zhang, J. Liang and D. Li, PID output-feedback control
-0.5
0
15 and filtering for positive Roesser system, Circuits System
5
10
15
10 Signal Process, to be published, doi: 10.1007/s00034-023-
20 5
25
30 m=0,1,...
02484-2.
l=0,1,... 35 0
[9] D. Li, J. Liang and F. Wang, Observer-based output feed-
back H∞ control of two-dimensional systems with periodic
Figure 1: Evolution of the η(l, m). scheduling protocol and redundant channels, IET Control
Theory & Applications, Vol. 14, No. 20, pp. 3713-3722,
2021.
explanation is given in describing the simulation. [10] Y. Wang, L. Xie, C. E. de Souza, Robust control of a class of
5 CONCLUSION uncertain nonlinear systems, Syst. Control Lett. 19 (2) (1992)
139-149.
This paper formulates an estimator with H∞ perfor-
mance based on an event-triggering mechanism for linear
repetitive processes under DoS attacks. An augmented esti-
mation error system is constructed, incorporating the event-
triggering mechanism and estimator parameters specifi-
cally tailored for 2-D DoS attacks. By leveraging Lya-
punov theory, the sufficient conditions ensuring the robust
stability of the 2-D augmented estimation error system are

2024 36th Chinese Control and Decision Conference (CCDC)

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on January 16,2025 at 15:44:06 UTC from IEEE Xplore. Restrictions apply.
4563