Professional Documents
Culture Documents
Journal Pre-proof
PII: S0016-0032(21)00496-8
DOI: https://doi.org/10.1016/j.jfranklin.2021.07.055
Reference: FI 5234
Please cite this article as: Guangtao Ran, Jian Liu, Chuanjiang Li, Hongtian Chen, Chunsong Han,
Finite-Time Filtering for Fuzzy Nonlinear Semi-Markov Jump Systems With Decep-
tion Attacks and Aperiodical Transmission, Journal of the Franklin Institute (2021), doi:
https://doi.org/10.1016/j.jfranklin.2021.07.055
This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.
Guangtao Rana , Jian Liub , Chuanjiang Lia,∗, Hongtian Chenc , Chunsong Hand
a Department of Control Science and Engineering, Harbin Institute of Technology, Harbin
150001, China
b School of Automation, Southeast University, Nanjing 210096, China
c Department of Chemical and Materials Engineering, University of Alberta, Edmonton, AB
Abstract
This paper investigates finite-time event-triggered filtering problems for fuzzy
nonlinear semi-Markov jump systems (S-MJSs) subject to deception attacks.
First, an interval type-2 (IT2) fuzzy S-MJSs is presented that can capture pa-
rameter uncertainties, in which the stochastic behavior is described by semi-
Markov process while the sojourn-time obeys non-exponential distribution. Sec-
ond, we design an adaptive event-triggered mechanism (AETM) for IT2 fuzzy
S-MJSs that can save network resources, in which the triggered parameter can
be automatically adjusted with the error information rather than a fixed value.
Third, a hidden-Markov model is employed to represent the asynchronous situa-
tion between the filter and system. Then, a new IT2 fuzzy filtering error system
is modeled by taking the effects of random deception attacks and AETM into
consideration. Moreover, finite-time boundedness (FTB) analysis and a filter
design method are proposed. Finally, an example is given to verify the effec-
tiveness of the finite-time event-triggered filtering algorithm.
Keywords: semi-Markov jump systems (S-MJSs), interval type-2 (IT2) fuzzy
model, hidden-Marov model, finite-time boundedness (FTB), event-triggered
filtering
1. Introduction
Over the past half-century, the type-1 Takagi-Sugeno (T-S) fuzzy systems
have received extensive attention in the control community because of their abil-
ity to deal with nonlinear systems well [1]. Although the type-1 fuzzy set theory
∗ Correspondingauthor
Email address: lichuan@hit.edu.cn (Chuanjiang Li)
2
was developed to further save computing resources and transmission resources
[37, 38, 39, 40, 41, 42]. Herein, the event-triggered function is developed by
designing the adaptive law for the threshold.
Furthermore, for the wireless channel, the cyber-attacks can not be ignored
55 when the signal is transmitted, especially considering the event-triggered con-
trol. Among the attacks, deception attacks are what we need to consider most
because it leads to serious threats to systems [43, 44, 45, 46]. In [43] an event-
triggered dynamic output feedback controller was designed for MJSs with de-
ception attacks. The nonhomogeneous MJSs with asynchronous modes was
60 considered in [44]. For the the Markov jump singularly perturbed systems, the
nonstationary quantized control problem was solved in [45]. In [47], the static
output feedback control problem was addressed for disturbed MJSs with ran-
domly deception attacks asynchronous mode information.
Although there are some results about filtering for S-MJSs, some issues
65 worthy to discuss further. The first one is that the designed filter is mode-
independent in [48], which means that all the system modes only correspond to
one filter gain. The second one is that the threshold parameter of the designed
ETS in [33] is set to be a fixed value, which cannot truly reflect the dynamics of
the system, thereby leading to a certain conservatism. The third one is that the
70 proposed method in [33, 48, 49] cannot be applied to S-MJSs with uncertainties,
but the uncertainties are inevitable in practical systems.
Motivated by these observations, this paper investigates the problem of
finite-time filter design for nonlinear S-MJSs with deception attacks and aperi-
odical transmission via IT2 T-S fuzzy models. The contributions are listed as
75 follows:
1) Different from [33, 48, 49], we consider a more practical IT2 fuzzy S-MJSs
that covers more complex situations: complex stochastic process, delay, decep-
tion attacks, asynchronous modes, and the limited network resources.
2) Compared with the existing works [33, 49], in this paper, the designed
80 filter can guarantee the FTB of the IT2 fuzzy S-MJSs, rather than infinite-time
interval. Moreover, the hidden-Markov model is introduced to describe the re-
lationships between system mode and filter mode.
3) The adaptive event-triggered mechanism (AETM) is designed to guaran-
tee that the IT2 T-S fuzzy S-MJSs are finite-time boundedness (FTB). Different
85 from the studies [48, 49], the proposed method can not only guarantee the per-
formance of the filtering error system but also reduce unnecessary waste of
communication resources.
The rest parts are organized as follows: In Sec. 2, some basic results are
presented. The main results are given in Sec. 3. Sec. 4 provides an example.
90 The conclusions are summarized in Sec. 5.
3
Fig. 1: The diagram of the system
4
PN1
115
n=1,n6=m ρmn (w). Inspired by [2, 30], the activation intensity corresponding to
the i-th rule can be written as ~i (x(t)) ∈ [~L
iQ(x(t)), ~U
i (x(t))], where ~i (x(t)) =
L
QK K
∂=1 µNi∂ (F∂ (x(t))) ≥ 0 and ~i (x(t)) =
L U U
∂=1 µNi∂ (F∂ (x(t))) ≥ 0 are lower
and upper grades (LAUGs) of membership function, respectively. µL Ni∂ and
µU
Ni∂ are lower and upper membership functions, respectively. Then, the overall
120 IT2 S-MJSs fuzzy model (1) can be inferred as
ẋ(t) = Amh x(t) + Bmh ν(t)
y(t) = Cmh x(t) + Dmh ν(t) (3)
z(t) = Emh x(t)
where P Pr Pr
r
Amh = Pi=1 ~i (x(t))Aϑt i , Bmh = Pi=1 ~i (x(t))Bϑt i , CP mh = i=1 ~i (x(t))Cϑt i ,
r r r
Dmh = i=1 ~i (x(t))Dϑt i , Emh = i=1 ~i (x(t))Eϑt i , i=1 ~i (x(t)) = 1,
~i (x(t)) = αiU (x(t))~U L L L U
i (x(t))+αi (x(t))~i (x(t)), in which αi (x(t)) and αi (x(t))
L U L
125 are nonlinear functions with 0 ≤ αi (x(t)) ≤ 1, 0 ≤ αi (x(t)) ≤ 1, αi (x(t)) +
αiU (x(t)) = 1.
2.2. AETM
Considering the limited network resources in the practice system, we employ
an AETM to alleviate the burden of communication network and save network
130 resources. As shown in Fig. 1, the sampled data y(sk h) can be transmitted only
violating the following condition:
where ey (t) = y(tk h) − y(sk h) is the error between latest transmitted signal
y(tk h) and current sampled signal y(sk h); tk h and sk h = tk h + Ch are latest
transmitted instant and current sampling instant, respectively; Xm denotes
135 the weighting matrix to be designed later for m ∈ O3 . ε(t) is the triggered
parameter, similar to the existing [50], which is determined by the following law
2
ε(t) = ε + (ε̄ − ε) exp−κky(sk h)−y(tk h)k (5)
where ε and ε̄ are predefined scalars that satisfy 0 < ε < ε̄ ≤ 1; κ > 0 is applied
2
to regulate the sensitivity of the ky(sk h) − y(tk h)k . According to (4) and (5),
the next triggered instant can be formulated as follows
tk+1 h = tk h + min Ch|eTy (t)Xm ey (t) > ε(t)y T (tk h)Xm y(tk h) (6)
C∈N
5
145 Define
t − tk h, t ∈ Ω0
d(t) = t − tk h − ih, t ∈ Ωi , i = 1, 2, ..., L0 − 1 (7)
t − tk h − L0 h, t ∈ ΩL0
which satisfies 0 ≤ d(t) ≤ h + d¯ = dM , t ∈ tk h + dtk , tk+1 h + dtk+1 . Overall,
we can derive y(tk h) = ey (t) + y(t − d(t)).
Remark 1. Different from the current results [33, 48, 49], this paper designs
the ATEM (4) to save network resources. It can be observed that the current
150 sampling signal can only be transmitted when the triggering condition (4) is
violated. In addition, it can be found that there is at least one sampling interval
h between two triggering interval. Therefore, the designed AETM (4) can avoid
Zeno behavior. Furthermore, the threshold of the designed AETM (4) can be
dynamically adjusted with the system. Therefore, the method proposed in this
155 paper contains the static case [33, 51].
where f (t) ∈ Rny is the spurious signal sent by attacker, which satisfies the
170 following energy constraint [51]:
Remark 3. It should worth noticing that χ(t) = 1 means that deception at-
175 tacks occur. On the contrary, no deception attacks occur when χ(t) = 0.
6
2.4. IT2 Fuzzy Filter
In this paper, we consider the following filter:
Filter Rule j: IF E1 (x(tk h)) is Mj1 , E2 (x(tk h)) is Mj2 ,..., and Eq (x(tk h)) is
Mjq , THEN
(
ẋf (t) = Af$t j xf (t) + B$
f
tj
ỹ(t)
f (11)
zf (t) = E$t j xf (t)
The following interval sets present the firing strength of the j-th rule:
185 Let $t = l and xµ = x(tk h). Then, by using fuzzy inference, we have
ẋf (t) = Alλ xf (t) + Blλ ỹ(t)
(14)
zf (t) = Elλ xf (t)
where
P
r P
r P
r
Alλ = λ̄j Af$t j , Blλ = f
λ̄j B$ t j , Elλ = f
λ̄j B$ tj
,
j=1 j=1 j=1
βjL (xµ )λ̄L U U
j (xµ )+βj (xµ )λ̄j (xµ )
0 < λ̄j = λ̄j (xµ ) = Pr L L U U ,
j=1 (βj (xµ )λ̄j (xµ )+βj (xµ )λ̄j (xµ ))
L U L U
0 ≤ βj (xµ ) ≤ 1, 0 ≤ βj (xµ ) ≤ 1, βj (xµ ) + βj (xµ ) = 1.
T _
Let x(t) = xT (t) xTf (t)
_ _
190 , z (t) = z(t) − zf (t), χ̄(t) = 1 − χ(t), and ν (t) =
T T
ν (t) ν T (t − d(t)) . For the sake of simplicity, ~i (x(t)) is replaced by ~i .
Combining with (3), (8), and (14), the filtering error system can be represented
as:
_˙ _ _
x(t) = Ãmlhλ x(t) + B̃mlhλ1 x(t − d(t)) + B̃mlhλ2 f (t)
_
+ B̃mlhλ3 ey (t) + B̃mlhλ4 ν (t) (15)
_ _
z (t) = Ẽmlhλ x(t)
where
P
r P
r P
r P
r
195 Ãmlhλ = ~iλ̄j Amlhλ , B̃mlhλ1 = ~iλ̄j Bmlhλ1 ,
i=1 j=1 i=1 j=1
Pr P r P
r P
r
Ẽmlhλ = ~iλ̄j Emlhλ , B̃mlhλ3 = ~iλ̄j Bmlhλ3 ,
i=1 j=1 i=1 j=1
Pr P r
B̃mlhλ4 = ~iλ̄j Bmlhλ4 ,
i=1 j=1
7
Ami 0 0 0 0 0
Amlhλ = f , Bmlhλ1 = f + χ̃(t) f ,
0 Alj χ̄1 Blj Cmi 0 −Blj Cmi 0
0 0 0 0
Bmlhλ2 = χ̃(t) f + f , Bmlhλ3 = f + χ̃(t) f ,
Blj χ1 Blj χ̄1 Blj −Blj
Bmi 0 0 0
200 Bmlhλ4 = f + χ̃(t) f ,
0 χ̄1 Blj Dmi 0 −Blj Dmi
h i
f
Emlhλ = Emi −Elj , χ̃(t) = χ(t) − χ1 , Ξ {χ̃(t)} = 0, Ξ {χ̃(t)χ̃(t)} = χ1 χ̄1 .
Before we give the main result, the following definitions and lemma are listed
for the convenience of later analysis.
Definition 1. Given a positive matrix P , positive scalars c1 , c2 , cv , and Tf ,
205 with c1 < c2 , the filtering error system (15) is FTB w.r.t. (c1 , c2 , cν , Tf , P ) for
any t ∈ [0, Tf ], if the following inequalities hold:
_T _ ˙T
_ ˙
_
sup Ξ x (θ)P x(θ), x (θ)P x(θ) ≤ c1 , (16)
−dM ≤θ≤0
Z Tf n T o
_T _ _ _
ν (θ) ν (θ)dθ ≤ cν , Ξ x (t)P x(t) ≤ c2 . (17)
0
3. Main Results
This section firstly presents a sufficient condition that the filtering error
system (15) remains FTB via the AETM and has H∞ performance. Then, the
existence condition of the filter is provided for the filtering error system (15).
8
220 3.1. H∞ FTB Performance Analysis
Theorem 1. Given scalars µ > 0, c1 > 0, c2 > 0, cv > 0, Tf > 0, 0 ≤ χ1 ≤
1, 0 < ε̄ ≤ 1, and matrix P > 0, the filtering error systems (15) is FTB w.r.t.
(c1 , c2 , cν , Tf , P ) and satisfy with a prescribed H∞ performance
level γ̃ > 0, if
R W
there exist matrices Gmli , Q > 0, R > 0, Ym > 0, ≥ 0, Xm > 0, for
∗ R
225 each i, j ∈ O1 , m ∈ O3 , l ∈ O4 , such that the following conditions hold:
XN2
ςml Gmli − R − Q < 0 (19)
l=1
c1 (b1 + b2 + b3 ) + γ 2 cν
≤ c2 (22)
exp−µTf λmin P −1/2 Ym P −1/2
where
Λ11
mlij Λ12
mlij Λ13
mlij Λ14
mlij
∗ −R 0 0
Λmlij = ∗
,
0
230
∗ −R
∗ ∗ ∗ −I
ψ11 ψ12 WT ψ14 ψ15 ψ16
∗ ψ22 W T − RT ψ24 0 ψ26
0
Λ11 = ∗ ∗ ψ33 0 0 ,
mlij ∗ ∗ ∗ ψ44 0 ψ46
∗ ∗ ∗ ∗ −χ1 I 0
∗ ∗ ∗ ∗ ∗ ψ66
12
T
Λmlij = dM Amlhλ B̄mlhλ1 0 B̄mlhλ3 B̄mlhλ2 B̄mlhλ4 ,
13 √ T
Λmlij = dM χ1 χ̄1 0 Σ1 0 Σ2 Σ3 Σ4 ,
14
T
Λmlij = Emlhλ 0 0 0 0 0 ,
T
PN1 Cmi
235 ψ11 = Q − R + He {Ym Amlhλ } + n=1 ρ̄mn Yn + χ1 F T F Cmi 0 ,
0
ψ12 = Ym B̄mlhλ1 + RT − W T ,
ψ14 = Ym B̄mlhλ3 , ψ15 = YmTB̄mlhλ2
,
Cmi
ψ16 = Ym B̄mlhλ4 + χ1 F T F 0 Dmi ,
0
T
Cmi
ψ22 = W T + W − 2R + ε̄ Xm Cmi 0 ,
0
T T
Cmi Cmi
240 ψ24 = ε̄ Xm , ψ26 = ε̄ Xm 0 Dmi ,
0 0
PN2
ψ33 = − l=1 ςml Gmli , ψ44 = (ε̄ − I) Xm , ψ46 = ε̄Xm 0 Dmi ,
9
0 T
0
ψ66 = χ1 T F F 0 Dmi + ε̄ T Xm 0 Dmi − γ 2 I,
Dmi Dmi
0 0 0
B̄mlhλ1 = f , B̄mlhλ2 = f ,
χ̄1 Blj Cmi 0 χ1 Blj
0 Bmi 0
B̄mlhλ3 = f , B̄mlhλ4 = f ,
χ̄1 Blj 0 χ̄1 Blj Dmi
0 0 0 0 0 0
245 Σ1 = f , Σ 2 = f , Σ 3 = f , Σ 4 = f ,
−Blj Cmi 0 −Blj Blj 0 −Blj Dmi
−1/2
b1 = λmax P Ym P −1/2 , b2 = dM expµdM λmax P −1/2 QP −1/2 ,
d 3
b3 = 2M expµdM λmax P −1/2 W P −1/2 .
250 where
_ _T _
V1 ( x(t), m, t) = x (t)Ym x(t),
_
R t _T _
V2 ( x(t), m, t) = t−dM expµ(t−s) x (s)Q x(s)ds,
_
R0 Rt _˙T ˙
_
V3 ( x(t), m, t) = dM −dM t+θ expµ(t−s) x (s)R x(s)dsdθ,
By computing the weak infinitesimal operator ` [·] [53] of (23), it yields that
h i X3 h i
_ _
` V ( x(t), m, t) = ` Vg ( x(t), m, t) , (24)
g=1
255 where
h i
_
` V1 ( x(t), m, t)
T
_ _
= ` x (t)Ym x(t)
h nP
1 N1
= lim+ ∆ Ξ n,n6=m Pr {ϑp+1 = n, δp+1 ≤ w + ∆|ϑp = m, δp+1 > w}
∆→0
_T _
× x (t + ∆)Yn x(t + ∆) +
o Pr {ϑp+1 = m, δip+1 > w + ∆|ϑp = m, δp+1 > w}
_T _ _T _
260 × x (t + ∆)Ym x(t + ∆) − x (t)Ym x(t)
h nP
1 N1 Pr{ϑp+1 =n,ϑp =m}
= lim ∆ Ξ n,n6=m Pr{ϑp =m}
∆→0+
Pr{δp+1 ≤w+∆,δp+1 >w|ϑp+1 =n,ϑp =m} _T _
× Pr{δp+1 >w|ϑp =m} x (t + ∆)Yn x(t + ∆)
o i
Pr{δp+1 >w+∆|ϑp =m} _T _ _T _
+ Pr{δ p+1 >w|ϑp =m}
x (t + ∆)Y m x(t + ∆) − x (t)Ym x(t)
h nP
1 N1 Fm (w+∆)−Fm (w) _T _
= lim ∆ Ξ n,n6=m gmn 1−Fm (w) x (t + ∆)Yn x(t + ∆)
∆→0+ o i
T _T
+ 1−F m (w+∆) _ _ _
265
1−Fm (w) x (t + ∆)Ym x(t + ∆) − x (t)Ym x(t) ,
h i Rt
_ _T _ _T _
` V2 ( x(t), m, t) = µ t−dM expµ(t−s) x (s)Q x(s)ds + x (t)Q x(t)
10
_T _
− x (t − dM )Q x(t − dM ),
h i R0 Rt
_ _˙T ˙
_ ˙T
_ ˙
_
` V3 ( x(t), m, t) = µ −dM t+θ expµ(t−s) x (s)R x(s)dsdθ + d2M x (t)R x(t)
Rt ˙T
_ ˙
_
− dM t−dM x (s)R x(s)ds.
270 where Fm (w) denotes the cumulative distribution function of the residence time
Pr{ϑp+1 =n,ϑp =m}
w when the system is maintained on the mode m. gmn := Pr{ϑp =m} =
Pr{ϑp+1 = n|ϑp = m} stands for the probability density of the system switching
from m to n. According to Taylor expansion, the following formula can be
derived for giving small constant ∆.
_ _ ˙
_
x(t + ∆) = x(t) + x(t)∆ + o(∆) (25)
1 − Fm (w + ∆) Fm (w) − Fm (w + ∆)
lim = 1, lim = ρm (w) (26)
∆→0+ 1 − Fm (w) ∆→0+ ∆(1 − Fm (w))
_
x(t − dM ) ∗ ∗ R (29)
_
x(t)
× x(t − d(t))
_
_
x(t − dM )
11
Recalling (9), we can obtain
−χ1 f T (t)f T T
( (t) + χ1y (t)F F y(t) T
Pr
_T
T
Cmi T
Cmi _T
= ~i χ1 x (t) FTF x (t)
i=1 0 0
( T T )
_T Cmi 0 _ (30)
+He χ1 x (t) FTF T ν (t)
0 Dmi
T )
_T 0 T 0 _ T
+χ1 ν (t) T F F T ν (t) − χ1 f (t)f (t) ≥ 0
Dmi Dmi
Pr
Define Gmlh = i=1 Gmli and combine (19), we can compute that
XN2
ςml Gmlh − R − Q < 0 (31)
l=1
Using Schur complemet lemma, Λmlij < 0 is equal to the following condition:
−R 0 0
Λ̄mlij = Λ11
mlij − Πmlij
∗ −R 0 ΠTmlij < 0 (33)
∗ ∗ −I
where Πmlij = Λ12 Λ13 Λ14
mlij .
h mlij mlij i
T _T _T _T
Define ΘT (t) = _ x (t) x (t − d(t)) x (t − dM ) eTy (t) f T (t) ν (t) .
It follows from (23) to (33) that
n h i o
_ _ _T _ _T _
Ξ ` V ( x(t), m, t) − µV ( x(t), m, t) + z (t) z (t) − γ 2 ν (t) ν (t)
P
r P r (34)
≤ ~iλ̄j ΘT (t)Λ̄mlij Θ(t) < 0
i=1 j=1
12
285 The following inequality can be obtained by integral calculation of (34):
n o
_ _
Ξ exp−µt V ( x(t), m, t) − V ( x(0), ϑ0 , 0)
n R Rt o (35)
t _T _ _T _
< Ξ − 0 exp−µθ z (θ) z (θ)dθ + γ 2 0 exp−µθ ν (θ) ν (θ)dθ
Similarly, we have
n o n n o T o
Ξ V ( x(t), m, t) ≥ Ξ λmin P −1/2 Ym P −1/2 x (t)P x(t)
_ _ _
(37)
Additionally, under zero initial condition, it can get from (35) that
(Z ) ( Z )
Tf Tf
_T _ 2 _T _
Ξ z (θ) z (θ)dθ < Ξ γ̃ ν (θ) ν (θ)dθ (39)
0 0
290 According to the Definition 2, the filtering error system is of FTB w.r.t.
(c
p1 c2 , cν , Tf , P ) and satisfies with a prescribed H∞ performance level γ̃ =
,
γ 2 expµTf . This completes the proof.
13
Theorem 2. Given scalars s1 > 0, µ > 0, c1 > 0, c2 > 0, cv > 0, Tf > 0,
0 ≤ χ1 ≤ 1, 0 < ε̄ ≤ 1, and matrix P > 0, the filtering error systems is
FTB w.r.t. (c1 , c2 , cν , Tf , P ) and satisfy with a prescribed H∞ performance level
305 γ̃ > 0, if there exist matrices
Gmli, Q > 0, R > 0, Ym > 0, Ym1 > Ym2 > 0,
R W
Q > 0, R > 0, Ym > 0, ≥ 0, Xm > 0, for each i, j ∈ O1 , m ∈ O3 ,
∗ R
l ∈ O4 , such that the following conditions hold:
XN2
ςml Gmli − R − Q < 0 (40)
l=1
R W
≥0 (41)
∗ R
c1 (b1 + b2 + b3 ) + γ 2 cν
−µT
≤ c2 (44)
exp fλ −1/2 Y P −1/2
min P m
where
Λ̃11
mlij Λ̃12
mlij Λ̃13
mlij Λ14
mlij
∗ 2
−2s1 Ym + s1 R 0 0
Λ̃mlij = ,
∗ ∗ −2s1 Ym + s21 R 0
∗ ∗ ∗ −I
ψ̃11 ψ̃12 WT ψ̃14 ψ̃15 ψ̃16
∗ ψ22 W T − RT ψ24 0 ψ26
∗ ∗ ψ33 0 0 0
11
Λ̃mlij = ,
∗ ∗ ∗ ψ44 0 ψ46
∗ ∗ ∗ ∗ −χ1 I 0
∗ ∗ ∗ ∗ ∗ ψ66
T
315 Λ̃12
mlij = dM ζ1 ζ2 0 ζ3 ζ4 ζ5 ,
13 √ T
Λ̃mlij = dM χ1 χ̄1 0 Σ̃1 0 Σ̃2 Σ̃3 Σ̃4 ,
T
PN1 Cmi
ψ̃11 = Q − R + He {ζ1 } + n=1 ρ̄mn Yn + χ1 FTF Cmi 0 ,
" # 0
Ym1 Ami −Āflj
ζ1 = , ψ̃12 = ζ2 + RT − W T ,
−Ym2 Ami Āflj
" # " #
f f
−B̄lj Cmi 0 −B̄lj
ζ2 = χ̄1 f , ψ̃14 = ζ3 , ζ3 = χ̄1 f ,
B̄lj Cmi 0 B̄lj
" # T
f
−B̄lj Cmi
320 ψ̃15 = ζ4 , ζ4 = χ1 f , ψ̃16 = ζ5 + χ1 FTF 0 Dmi ,
B̄lj 0
14
" # " #
f f
Ym1 Bmi −χ̄1 B̄lj Dmi B̄lj Cmi 0
ζ5 = f , Σ̃1 = f ,
0 χ̄1 B̄lj Dmi −B̄lj Cmi 0
" # " #
f f
B̄lj 0 −B̄lj Dmi
Σ̃2 = f , Σ̃ 3 = − Σ̃ 2 , Σ̃ 4 = f ,
−B̄lj 0 B̄lj Dmi
Then, the corresponding parameters of the filter can be written as
Aflj = Ym2
−1 f f
Ālj , Blj −1 f
= Ym2 f
B̄lj , Elj f
= Ēlj . (45)
Proof:
For the sake
of analysis, we decompose the matrix Ym into Ym =
Ym1 −Ym2
325 > 0. We can easily observe that Ym1 > Ym2 > 0. More-
−Ym2 Ym2
over, some variables are defined as follows:
Remark 5. Note that Theorem 2 gives the existence conditions of the filter
parameters, and the filter gain can be obtained by selecting appropriate pa-
rameters. However, there is no optimal method for the selection of parameters,
335 which prompts us to conduct further research in future work.
4. Numerical Example
where Lm (t) and vcm (t) (m = 1, 2) represent the inductor current and ca-
pacitor voltage, respectively. σ1 = 0.01, σ2 = 0.005, R = 2; when m = 1,
15
Fig. 2: Electric circuit model with switching modes.
345 L1 = 0.1, C1 = 0.1; when m = 2, L1 = 0.2, C1 = 0.5. Let x1 (t) = IL (t) and
T
x2 (t) = vc (t). Then state vector can be expressed as x(t) = x1 (t) x2 (t) .
Moreover, U consider as disturbance input. By adopting IT2 fuzzy sets with
r = 2, we have
Plant Rule 1: IF x2 (t) is 0, THEN
350
ẋ(t) = Am1 x(t) + Bm1 ν(t)
y(t) = Cm1 x(t) + Dm1 ν(t) (49)
z(t) = Em1 x(t)
where
−20 −10 0.1
A11 B11
= 10 2 0.05 , E11 = 0.3 −0.2 ,
C11 D11
0.5 −0.4 −0.3
−10 −5 0.05
A12 B12
= 2 0.4 0.01 , E12 = 0.2 −0.2 ,
C12 D12
0.3 −0.1 −0.2
−20 −10 0.1
A21 B21
355 = 10 −16 0.05 , E21 = 0.1 −0.5 ,
C21 D21
0.25 −0.2 −0.2
−10 −5 0.05
A22 B22
= 2 −3.2 0.01 , E22 = 0.1 −0.5 .
C22 D22
0.15 −0.3 −0.1
The LAUGs of membership functions of the plant and filter are given in the
following:
16
Table 1: The values of a and b for different modes
modes a b
m=1 1 2
m=2 2 1
2 2
~L1 (x2 (t)) = 1 − exp , ~1 (x2 (t)) = 1 − exp−x2 (t)/5 ,
−x2 (t)/9 U
αiL (x2 (t)) = 0.6 × sin (x2 (t))2 , αiU (x2 (t)) = 1 − αiL (x2 (t)),
−x22 (t)/7 U 2
λ̄L
1 (x2 (t)) = 1 − exp , λ̄1 (x2 (t)) = exp−x2 (t)/6 ,
λ̄L U U L
2 (x2 (t)) = 1 − λ̄1 (x2 (t)), λ̄2 (x2 (t)) = 1 − λ̄1 (x2 (t)),
βjL (x2 (t)) = 0.6 × sin (x2 (t))2 , βjU (x2 (t)) = 1 − βjL (x2 (t)).
365 Moreover, the switch modes obey the semi-Markov process and take values
in with the following transition matrix
η11 (w) η12 (w) −2w 2w
η(w) = =
η21 (w) η22 (w) 0.5w −0.5w
Assume that the sojourn time obeys the Weibull distribution and accords with
b
the probability density function fm (w) = abb wb−1 exp−(w/a) , w ≥ 0. Table I
gives the values of a and b for different switch modes. From Table I, we can
2
370 obtain f1 (w) = 2wexp−w and f2 (w) = 0.5exp
−0.5w
. Based
on the results, we
−1.7725 1.7725
then can calculate that Ξ {η(w)} = .
1 −1
Let the disturbance input ν(t) = 0.2 sin(t) × exp−0.3t , the upper bound of
delay d¯ = 0.1 , dM = 0.2, ε̄ = 0.3, ε = 0.1, κ = 0.5, χ1 = 0.5 and deception
attacks function f (t) = 0.2sin(x2 (t)) with the upper bounded F =0.2. Besides,
0.5 0.5
the conditional probability matrix is chosen as Υ = . According
0.4 0.6
to Definition 2, set (c1 , c2 , cν , Tf , P ) = (0.1, 1, 1, 10, I). By solving the linear
matrices inequalities in Theorem 2, we can derive that
−18.4693 2.0369 −10.8753 −1.0381
Af11 = , Af12 = ,
4.0141 −11.3711 0.7056 −2.7149
−18.4693 2.0369 −10.8753 −1.0381
Af21 = , Af22 = ,
4.0141 −11.3711 0.7056 −2.7149
f 1.0163 f 0.6356 f 1.0163
B11 = , B12 = , B21 = ,
−0.4204 −0.2489 −0.4204
T T
f 0.6356 f −0.0497 f 0.0468
B12 = , E11 = , E12 = ,
−0.2489 −0.0426 −0.1425
T T
f −0.0497 f 0.0468 X1 0.4842
E21 = , E12 = , = .
−0.0426 −0.1425 X2 0.8455
17
2
System modes
1.5
0 1 2 3 4 5 6 7 8 9 10
2
Filter modes
1.5
0 1 2 3 4 5 6 7 8 9 10
Time(s)
5. Conclusion
18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
-0.02
-0.04
0 1 2 3 4 5 6 7 8 9 10
Time(s)
0.15
0.1
Estimate error
0.05
0 1 2 3 4 5 6 7 8 9 10
Time(s)
19
0.3
0.25
0.2
0.15
0.1
0.05
0
0 1 2 3 4 5 6 7 8 9 10
Time(s)
1.4
The release instants and release intervals
1.2
0.8
0.6
0.4
0.2
0
0 1 2 3 4 5 6 7 8 9 10
Time(s)
20
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0 1 2 3 4 5 6 7 8 9
Time(s)
395 Acknowledgments
This work was supported in part by the National Natural Science Foun-
dation of China under Grant 61876050, the China Scholarship Council under
Grant 202006120100, the Jiangsu Planned Projects for Postdoctoral Research
Funds under Grant 2020Z081, the Fundamental Research Funds for the Cen-
400 tral Universities under Grant 2242021R20002, the China Postdoctoral Science
Foundation funded project under Grant 2021M690032, and the Equipment Pre-
research Field Fund (JZX7Y20190243001201).
References
References
405 [1] H. Li, J. Yu, C. Hilton, H. Liu, Adaptive sliding-mode control for nonlinear
active suspension vehicle systems using T–S fuzzy approach, IEEE Trans.
Ind. Electron. 60 (8) (2012) 3328–3338.
[2] H. K. Lam, L. D. Seneviratne, Stability analysis of interval type-2 fuzzy-
model-based control systems, IEEE Transactions on Systems, Man, and
410 Cybernetics, Part B (Cybernetics) 38 (3) (2008) 617–628.
[3] H. Li, Y. Pan, Q. Zhou, Filter design for interval type-2 fuzzy systems with
D stability constraints under a unified frame, IEEE Trans. Fuzzy Syst.
23 (3) (2014) 719–725.
[4] H. Li, S. Yin, Y. Pan, H. K. Lam, Model reduction for interval type-2
415 takagi–sugeno fuzzy systems, Automatica 61 (2015) 308–314.
[5] W. Zheng, Z. Zhang, H. Wang, H. Wang, Robust H∞ dynamic output
feedback control for interval type-2 T-S fuzzy multiple time-varying delays
21
systems with external disturbance, Journal of the Franklin Institute 357 (6)
(2020) 3193–3218.
420 [6] W. Zheng, H. Wang, C. Hua, Z. Zhang, H. Wang, Dynamic output-feedback
control for nonlinear continuous-time systems based on parametric uncer-
tain subsystem and interval type-2 fuzzy model, Journal of the Franklin
Institute 355 (16) (2018) 7962–7984.
[7] L. Wang, C. L. P. Chen, H. Li, Event-triggered adaptive control of saturated
425 nonlinear systems with time-varying partial state constraints, IEEE Trans.
Cybern. 50 (4) (2020) 1485–1497.
[8] H. Chen, B. Jiang, S. X. Ding, B. Huang, Data-driven fault diagnosis for
traction systems in high-speed trains: A survey, challenges, and perspec-
tives, IEEE Transactions on Intelligent Transportation Systems (2020) 1–
430 17doi:10.1109/TITS.2020.3029946.
[9] H. Dong, Z. Wang, H. Gao, Distributed H∞ filtering for a class of Marko-
vian jump nonlinear time-delay systems over lossy sensor networks, IEEE
Trans. Ind. Electron. 60 (10) (2013) 4665–4672.
[10] Y. Zhang, Y. He, M. Wu, J. Zhang, Stabilization for markovian jump
435 systems with partial information on transition probability based on free-
connection weighting matrices, Automatica 47 (1) (2011) 79–84.
[11] D. Yang, G. Zong, S. K. Nguang, X. Zhao, Bumpless transfer H∞ anti-
disturbance control of switching markovian LPV systems under the hybrid
switching, IEEE Transactions on Cybernetics (2020) 1–13doi:10.1109/
440 TCYB.2020.3024988.
[12] X. Yang, Q. Zhu, Stabilization of stochastic retarded systems based on
sampled-data feedback control, IEEE Transactions on Systems, Man, and
Cybernetics: Systems (2019) 1–10doi:10.1109/TSMC.2019.2956757.
[13] H. Shen, Z. Wu, J. H. Park, Reliable mixed passive and filtering for semi-
445 markov jump systems with randomly occurring uncertainties and sensor
failures, International Journal of Robust and Nonlinear Control 25 (17)
(2015) 3231–3251.
[14] L. Zhang, H. K. Lam, Y. Sun, H. Liang, Fault detection for fuzzy semi-
markov jump systems based on interval type-2 fuzzy approach, IEEE Trans-
450 actions on Fuzzy Systems 28 (10) (2020) 2375–2388.
[15] Y. Xia, J. Xia, Z. Wang, H. Shen, Extended non-fragile dissipative esti-
mation for nonlinear semi-markov jump systems, Journal of the Franklin
Institute 357 (1) (2020) 457–472.
[16] W. Qi, G. Zong, H. R. Karimi, SMC for nonlinear stochastic switching
455 systems with quantization, IEEE Transactions on Circuits and Systems II:
Express Briefs (2020) 1–1doi:10.1109/TCSII.2020.3047785.
22
[17] W. Qi, J. H. Park, G. Zong, J. Cao, J. Cheng, Filter for positive stochas-
tic nonlinear switching systems with phase-type semi-markov parameters
and application, IEEE Transactions on Systems, Man, and Cybernetics:
460 Systems (2021) 1–12doi:10.1109/TSMC.2020.3049137.
[18] G. Zong, W. Qi, H. R. Karimi, L1 control of positive semi-markov jump
systems with state delay, IEEE Transactions on Systems, Man, and Cyber-
netics: Systems (2020) 1–10doi:10.1109/TSMC.2020.2980034.
[19] B. Wang, Q. Zhu, Stability analysis of discrete-time semi-markov jump
465 linear systems, IEEE Transactions on Automatic Control 65 (12) (2020)
5415–5421. doi:10.1109/TAC.2020.2977939.
23
[29] G. Ran, J. Liu, C. Li, L. Chen, D. Li, Event-based finite-time consensus
495 control of second-order delayed multi-agent systems, IEEE Transactions
on Circuits and Systems II: Express Briefs 68 (1) (2021) 276–280. doi:
10.1109/TCSII.2020.2999480.
[30] G. Ran, C. Li, H. K. Lam, D. Li, C. Han, Event-based dissipative con-
trol of interval type-2 fuzzy markov jump systems under sensor satura-
500 tion and actuator nonlinearity, IEEE Trans. Fuzzy Syst., to be published,
doi:10.1109/TFUZZ.2020.3046335.
[31] Q. Zhu, Stabilization of stochastic nonlinear delay systems with exogenous
disturbances and the event-triggered feedback control, IEEE Transactions
on Automatic Control 64 (9) (2019) 3764–3771. doi:10.1109/TAC.2018.
505 2882067.
[32] J. Liu, Y. Yu, H. He, C. Sun, Team-triggered practical fixed-time consen-
sus of double-integrator agents with uncertain disturbance, IEEE Transac-
tions on Cybernetics 51 (6) (2021) 3263–3272. doi:10.1109/TCYB.2020.
2999199.
510 [33] J. Wang, M. Chen, H. Shen, Event-triggered dissipative filtering for net-
worked semi-Markov jump systems and its applications in a mass-spring
system model, Nonlinear Dyn. 87 (4) (2017) 2741–2753.
[34] G. Zong, H. Ren, Guaranteed cost finite-time control for semi-markov jump
systems with event-triggered scheme and quantization input, International
515 Journal of Robust and Nonlinear Control 29 (15) (2019) 5251–5273.
[35] H. Shen, M. Chen, Z. G. Wu, J. Cao, J. H. Park, Reliable event-triggered
asynchronous extended passive control for semi-markov jump fuzzy systems
and its application, IEEE Transactions on Fuzzy Systems 28 (8) (2020)
1708–1722.
520 [36] J. Wang, T. Ru, J. Xia, H. Shen, V. Sreeram, Asynchronous event-triggered
sliding mode control for semi-markov jump systems within a finite-time
interval, IEEE Transactions on Circuits and Systems I: Regular Papers
68 (1) (2021) 458–468. doi:10.1109/TCSI.2020.3034650.
24
[39] Y. Wang, Y. Xia, C. K. Ahn, Y. Zhu, Exponential stabilization of Takagi–
Sugeno fuzzy systems with aperiodic sampling: An aperiodic adaptive
event-triggered method, IEEE Trans. Syst., Man, Cybern., Syst. 49 (2)
535 (2018) 444–454.
[40] H. Zhang, Z. Wang, H. Yan, F. Yang, X. Zhou, Adaptive event-triggered
transmission scheme and H∞ filtering co-design over a filtering network
with switching topology, IEEE Trans. Cybern. 49 (12) (2019) 4296–4307.
[41] Z. Gu, P. Shi, D. Yue, Z. Ding, Decentralized adaptive event-triggered H∞
540 filtering for a class of networked nonlinear interconnected systems, IEEE
Trans. Cybern. 49 (5) (2018) 1570–1579.
25
[50] X. Xie, S. Li, B. Xu, Adaptive event-triggered H∞ fuzzy filtering for in-
570 terval type-2 T–S fuzzy-model-based networked control systems with asyn-
chronously and imperfectly matched membership functions, Journal of the
Franklin Institute 356 (18) (2019) 11760–11791.
[51] J. Wu, C. Peng, J. Zhang, B. L. Zhang, Event-triggered finite-time H∞
filtering for networked systems under deception attacks, Journal of the
575 Franklin Institute 357 (6) (2020) 3792–3808.
[52] Z. Lu, G. Ran, F. Xu, J. Lu, Novel mixed-triggered filter design for interval
type-2 fuzzy nonlinear Markovian jump systems with randomly occurring
packet dropouts, Nonlinear Dyn. 97 (2) (2019) 1525–1540.
26