Received: 4 September 2019 Revised: 9 November 2019 Accepted: 10 November 2019

DOI: 10.1002/rnc.4837

RESEARCH ARTICLE

Resilient control of networked control systems under

deception attacks: A memory-event-triggered
communication scheme

Kunyu Wang1,4 Engang Tian1,2 Jinliang Liu3 Linnan Wei4 Dong Yue5

1
School of Optical-Electrical and
Computer Engineering, University of Summary
Shanghai for Science and Technology, This paper considers resilient event-triggered control problem for a class of net-
Shanghai, China
2
worked systems subject to randomly occurring deception attacks. First, a new
College of Automation Electronic
Engineering, Qingdao University of memory event-triggered scheme (METS) is proposed to reduce the utilization
Science and Technology, Qingdao, China of communication resources while maintaining desired system performance.
3
College of Information Engineering, Different from some existing event-triggered schemes, some recently released
Nanjing University of Finance and
Economics, Nanjing, China
packets are firstly utilized in the proposed METS, which provides a better flexi-
4
School of Electrical and Automation bility to improve the system dynamics. Second, considering the security problem
Engineering, Nanjing Normal University, of the networked control systems, a randomly occurring deception attack model
Nanjing, China
is employed where the bounded malicious signals are injected by the adver-
5
Institute of Advanced Technology,
sary. Considering both the effects of METS and deception attack, new type of
Nanjing University of Posts and
Telecommunications, Nanjing, China networked control system model is constructed and the corresponding mem-
ory state-feedback controller is designed. Then, sufficient conditions for the
Correspondence
Engang Tian, School of Optical-Electrical asymptotical stability of the systems are derived by using a Lyapunov functional
and Computer Engineering, University of technique. Finally, the obtained results are verified through a pendulum system,
Shanghai for Science and Technology,
which demonstrates the effectiveness of the proposed methods.
Shanghai 200093, China; or
College of Automation Electronic
K E Y WO R D S
Engineering, Qingdao University of
Science and Technology, Qingdao 266061, deception attacks, Lyapunov functional, memory event-triggered control, networked
China. control systems
Email: tianengang@[Link]

Funding information
National Natural Science Foundation of
China, Grant/Award Number: 61773218
and 61903252; Natural Science
Foundation of Jiangsu Province,
Grant/Award Number: BK20161561;
Program for Professor of Special
Appointment (Eastern Scholar) at
Shanghai Institutions of Higher Learning

1 I N T RO DU CT ION

Control systems over a certain digital networks are referred as networked control systems (NCSs); in an NCS, the com-
ponents including the sensors, controllers, or actuators are usually distributed and connected to the plant through
wired/wireless network. NCSs have been applied in lots of different areas such as environmental monitoring, smart grids,

Int J Robust Nonlinear Control. 2019;1–15. [Link]/journal/rnc © 2019 John Wiley & Sons, Ltd. 1
2 WANG ET AL.

teleoperation control, artificial vehicles, industrial automation, and unmanned marine vehicles.1-9 In spite of the benefits
of NCSs, there are still many tough challenges in NCSs which has to be dealt with. In recent years, great efforts have been
paid to NCSs and a rich body of pioneer research studies has emerged to solve the modeling, control, filtering, and estima-
tion problems in NCSs (see other works10-18 and the references therein). To name a few, Cloosterman et al10 investigated
the stabilization problem of NCSs with time-varying delays. The H∞ filtering problems of NCSs have been studied in the
works of Huang et al11 and Jiang et al.12 Using an improved Riccati equation method, the authors considered both the
network-induced delays and packets dropout in NCSs simultaneously.13 It should be noted that a time-triggered scheme
is widely used in above mentioned investigations, ie, the sensor samples the output periodically and transmits all the sam-
pled signal to the controller. To guarantee desired system dynamics, the sampling period is chosen in the worst case of
the system. As a result, the sampling period is set small at most of the times and a great number of redundant packets are
released, which will waste the scarce network resources. In order to overcome this problem, lots of research studies have
been developed on how to make better use of the scarce resources while still guaranteeing the system performance.19-24
So far, the so-called event-triggered scheme (ETS) has been designed to reduce the packet transmission in the network
(see other works19,21,22,24-27 and the references therein). In ETS strategies, the transmission or not of a packet is deter-
mined by a predesigned triggering condition that depends on the state or output of the system. In this way, the number of
transmissions is reduced while desired system performance is maintained. For example, in some literature, a typical ETS
is given as tk+1 = min{t ∈ R|(||x(tk ) − x(t)||) ≥ 𝜎(||x(t)||)}, where x(t) is the current state, x(tk ) is the last released packet,
tk is the most recent released instant, tk+1 is the next released instant, and 𝜎 is a constant threshold parameter. In the past
two decades, different types of triggering conditions have been explored, which can be roughly classified into several cat-
egories, such as the absolute ETS,28 mixed ETS,29 hybrid ETS,30 static ETS,31 dynamic ETS,32 periodic/discrete ETS,21,24
adaptive ETS,33 and distributed ETS.34,35 It is worth noting that, in order to avoid ZENO phenomenon, Heemels et al21
and Yeu et al24 made a trade-off between the time-triggered control and event-triggered control and proposed a peri-
odic/discrete event-triggered control for linear systems. More recently, an adaptive ETS is developed in other works,33,36-38
wherein the threshold parameter can be adjusted adaptively according to the system dynamics.
It should be pointed out that most of the triggering conditions are designed based on the difference between the newly
sampled signal x(t) and the last released packet x(tk ), that is, x(t) − x(tk ). In those ETSs, when x(t) − x(tk ) is small, the packet
x(t) is not likely to be released. However, only the error information is not enough to reflect the full dynamic character
of the system. A “smart” ETS should consider more system dynamics so as to make a well trade-off between system
performance and communication resource utilization. For example, when the curve of the system dynamic reaches the
vertex of the response, the relative error between two sampled signals becomes small therefore the ETS is not likely to
release this packet. However, we expect that, at these instants, more sampled signals would be released so as to shorten
the transient process. For this purpose, we make one of the first several attempts to design a memory ETS (METS) by
using some historic released signals in the triggering condition. Two simple cases are proposed in Section 2.1 to illustrate
the effects of the proposed METS.
Along with the pervasive utilization of open yet unprotected communication network, the NCSs are vulnerable to cyber
attacks. It is reported in Guo et al39 that a successful cyber attacks may lead to great serious consequences, including
customer information leakage, destruction of infrastructure, and even threat of human lives. As a result, the security prob-
lems in NCSs have stimulated increasing attention and numerous outstanding results have been published.40-45 Among
them, deception attacks, as one of the utmost important cyber attacks, can wreck the integrity of the data by modifying
its content. More recently, increasing attention has been devoted to the exploration of the deception attacks (see other
works39,46-50 ).
Motivated by above observation, the goal of this paper is to design a resilient memory controller for the proposed NCSs
subject to randomly occurring deception attacks. A new type of ETS, called METS, is proposed to reduce the number of
redundant transmission. In order to implement the METS, two buffers are needed at the event-generator side and the
controller side, respectively, which are used to store the historic released signals. Furthermore, considering the security
problem of the NCSs, the effect of randomly occurring deception attacks is considered. By using the proposed memory
feedback controller, sufficient conditions are obtained to guarantee the asymptotic stability of NCSs. The feedback con-
troller and trigger parameters can be codesigned through solving a set of linear matrix inequalities (LMIs). Finally, a
pendulum system example is proposed to validate the feasibility and practicality of the proposed METS and controller
design method.
Notation. The notation used here is standard except where otherwise stated. Rn and Rn×m denote the n dimensional
Euclidean space and n × m real matrices, respectively. AT denotes the transpose of the matrix A. Im×m denotes a matrix
WANG ET AL. 3

with all elements are 1. X ≥ Y (respectively, X > Y) represent X − Y is positive semidefinite (respectively,
√ positive
definite), where X and Y are symmetric matrices. For a vector x ∈ R , we denote by ||x|| ∶= x x its 2-norm. In
n T

symmetric block matrices, “∗” is used to represent an ellipsis for terms that is induced by symmetry.

2 PROBLEM FO RMULATION

Considering the following system:

.
x(t) = Ax(t) + Bu(t), (1)
where x(t) ∈ Rn
is the system state vector, u(t) ∈ Rm
is control input. A and B are parameters with appropriate dimensions,
and u(t) is assumed to be 0 before the first packet is received.

2.1 The memory event-triggered scheme

Before proposing the METS, we first recall the periodic ETS proposed in the work of Yue et al24
{ }
tk+1 h = tk h + min lh|eTk,l Wek,l > 𝜎xT (tk h)Wx(tk h) , (2)
l∈N

where eTk,l ∶= x(tk h) − x(tk h + lh) and h is the sampling period, tk is a integer and {t0 , t1 , … } ⊂ {0, 1, … }, which denotes
the sequence the sampling instant, tk h denotes the most recent triggered instant. tk h + lh represents the sampling time.
l ∈ N satisfying l < tk+1 − tk , 0 < 𝜎 < 1 is a predefined threshold parameter, and W is a positive symmetric matrix to be
designed.
Remark 1. According to the ETS (2) proposed in the work of Yue et al,24 it can be found that whether a new sampled
packet can be released mainly determines on two factors, the first is the threshold parameter 𝜎, a smaller 𝜎 leads to
more triggering instants. When 𝜎 is fixed, it determines on the error ek,l = x(tk h) − x(tk h + lh), a larger ek,l means that
||x(tk h+lh)−x(tk h)||
the new packet x(tk h + lh) is more likely to be released. In this paper, an index is defined as 𝜆tk h ∶= ||x(tk h)||
, the
values of 𝜆tk h demonstrate whether a new sampled signal deserves to be released, obviously, a larger 𝜆tk h means the
sampled signal is more likely to be transmitted.
In order to show the limitation of the ETS in (2), a simple case of the system response is given in Figure 1. Assume that
the values of the sampled responses are listed in Table 1.

FIGURE 1 A simple case of the system response [Colour figure

can be viewed at [Link]]

x(t1 h) x(t2 h) x(t3 h) x(t4 h) x(t5 h) TABLE 1 The values of the system response
values 0.1 0.2 0.6 0.7 0.8
𝜆tk h − 1.000 2.000 0.167 0.143
4 WANG ET AL.

From Table 1, it can be found that index 𝜆tk h for packets x(t2 h)and x(t3 h) are large but those for the packets x(t4 h) and
x(t5 h) are very small. Therefore, as described above, x(t4 h) and x(t5 h) are not likely to be released to the controller. However,
the values of x(t4 h) and x(t5 h) are much larger than x(t2 h). Generally speaking, when the response of the system reach
the vertex, we usually expect the ETS releasing more packets thus generating more control input, such that the transient
process maybe shorten.
From above observation, it can be concluded that some existing ETSs cannot release more packets when the system
curve is at the vertex. The reasons may be that only the current and most released packets information is utilized in the
ETS. To overcome this problem, a METS is proposed by utilizing the historic released signals, which is described as

tk+1 h = tk h + min {lh|𝜑̂ (x(tk h), ei (t), 𝜎) > 0} , (3)

l∈N

where

∑
m
𝜑̂ (x(tk h), ei (t), 𝜎) ∶= 𝜇i eTi (t)Wei (t) − 𝜎 x̄ T W x̄
i=1

1 ∑m ∑m
with x̄ ∶= m i=1 x(tk+1−i h), ei ∶= x(tk−i+1 h) − x(tk h + lh), 𝜇 i ∈ [0, 1] are the weighting parameters and i=1 𝜇i = 1.
Remark 2. In the METS (3), m historic released packets {x(tk h), … , x(tk−m+1 h)} are utilized, and if m = 1, the METS
reduces to the ETS in (2). The weighting parameters 𝜇 i are selected to demonstrate the weights of the released packets.
Generally, we regard that the more recent released packets are more important than others. In this paper, we assume
that 𝜇i ≥ 𝜇i+1 (i = 1, … , m − 1).
∑m
𝜇 |e (t)|
Using the proposed METS (3), for 𝜇1 = 0.6, 𝜇2 = 0.2, 𝜇 3 = 0.2, the index 𝜆̄ tk h = i=1 |̄xi| i for the packets x(t4 h) and
x(t5 h) in Figure 1 and Table 1 is, respectively, 0.933 and 0.44. Obviously, they are much larger than those obtained using
the ETS (2), that is, the packets x(t4 h) and x(t5 h) are more likely to be released in the METS (3) than by using ETS (2).
In NCSs, the controller will hold the same in the interval [tk h + 𝜏tk , tk+1 h + 𝜏tk+1 ) ≜ , which can be divided into several
⋃l=t −t −1
subintervals as  = l=0k+1 k l , l = [tk h + lh + 𝛿tk , tk h + (l + 1)h + 𝛿tk ) with

{
𝜏t k , l = 0, 1, … , tk+1 − tk − 2,
𝛿tk =
𝜏tk+1 , l = tk+1 − tk − 1,

where 𝜏tk and 𝜏tk+1 are transmission delay of the packets x(tk h) and x(tk+1 h), respectively. Define 𝜏(t) = t−tk h−lh for t ∈ l ,
it is clear that 𝜏(t) is a piecewise function and 0 ≤ 𝜏tk ≤ 𝜏(t) ≤ h + 𝜏̄ ≜ 𝜏M , where 𝜏̄ is the upper bound of the transmission
delay {𝜏tk }.
Corresponding to the METS, a memory controller is proposed as

∑
m
u(t) = Ki x(tk−i+1 h), t ∈ . (4)
i=1

Considering the METS (3) and memory controller (4), for t ∈ i , system (1) can be rewritten as

. ∑
m
x(t) = Ax(t) + BKi [x(t − 𝜏(t)) + ei (t)],
i=1 (5)
x(t) = 𝜙(t), t ∈ [−𝜏M , 0),

where 𝜙(t) is the initial condition of x(t). The structure of the NCS with METS is illustrated in Figure 2. In the considered
system, in order to implement the METS and memory controller, two buffers are needed to store the past released samples
and the length of the buffer is m (1 ≤ m).
WANG ET AL. 5

FIGURE 2 The structure of memory event-triggered networked control system under deception attacks. METS, memory event-triggered
scheme

2.2 The deception attack model

In this paper, it is assumed that the attacker can capture the system dynamic x(tk h) and release aggressive signals
f (x(t − d(t)) in a random way. When deception attacks happen, the faked signal would join in the buffer of controller
together with true signal. It is hard to distinguish the faked signals from the nonattacked one. To get rid of this dilemma,
we design the following scheduling rules: at every released instant, the event-generator will packet m historic released sig-
nal {x(tk h), … , x(tk−m+1 h)} and send them to the controller together. In this way, the NCSs (5) under randomly occurring
deception attacks can be expressed as

. ∑
m
∑
m
x(t) = Ax(t) + 𝜃(t)BKi 𝑓i (x(t − d(t))) + (1 − 𝜃(t)) BKi (x(t − 𝜏(t)) + ei (t)), (6)
i=1 i=1

where 𝜃(t) ∈ {0, 1} denotes the occurring probability of deception attacks. When 𝜃(t) = 0, it means there are no attacks,
when 𝜃(t) = 1, the original signal [x(tk h), … , x(tk−m+1 h)] is captured by the attacker and replaced by a group of aggressive
signal [ f1 (x(t − d(t)), … , fm (x(t − d(t))]. The mathematical characters of 𝜃(t) are assumed to be known as

E(𝜃(t)) = 𝜃, E(𝜃(t) − 𝜃)2 = 𝜃(1 − 𝜃) ≜ 𝛾 2 .

The aggressive signals fi (x(t − d(t))(i ∈ {1, 2, … , m}) are assumed to satisfy

‖𝑓i (x(t − d(t))‖2 ≤ ‖Gi x(t − d(t))‖2 , (7)

where Gi is respective known matrix representing the upper bound of the nonlinearity fi (·), 0 ≤ d(t) ≤ dM . Without loss
of generality, assume that Gmax is the largest one among Gi .
The following infinitesimal operator  is needed in the main results.
Definition 1 (See the works of Liu et al51 ). For a given function V ∶ CFB ([−𝜏M , 0], Rn ) × S, its infinitesimal operator
0
 is defined as

1
(V𝜂 (t)) = lim + [E(V(𝜂t + Δ)|𝜂t ) − V(𝜂t )] . (8)
Δ→+0 Δ
6 WANG ET AL.

3 MAIN RESULTS

The purpose of this section is to design the memory controller (4) and METS (3) such that system (6) under randomly
occurring deception attacks is asymptotically stable. Firstly, by using Lyapunov functional method, sufficient conditions
for the stability of system (6) with given controller feedback gains are derived.
Theorem 1. For given parameters m, 𝜎, 𝜃, 𝛾, 𝜇𝛼 and matrix sequence K𝛼 , 𝛼 ∈ {1, . … , m}, system (6) is asymptotically
stable if there exist matrices P > 0, Q1 > 0, Q2 > 0, R1 > 0, R2 > 0, W > 0 and N, M, T, S with appropriate dimensions
such that

⎡ Ω11 ∗ ∗ ∗ ∗ ⎤
⎢ Ω (i) −R ∗ ∗ ∗ ⎥
⎢ 21 1
⎥
Ω(i) = ⎢ Ω31 (1) 0 −R1 ∗ ∗ ⎥ < 0, i = 1, 2 (9)
⎢ Ω41 (1) 0 0 −R1 ∗ ⎥
⎢ Ω 0 −P ⎥⎦
⎣ 51 0 0

⎡ Ω11 ∗ ∗ ∗ ∗ ⎤
⎢ Ω21 ( 𝑗) −R2 ∗ ∗ ∗ ⎥
⎢ ⎥
Ω( 𝑗) = ⎢ Ω31 (2) 0 −R2 ∗ ∗ ⎥ < 0, 𝑗 = 3, 4, (10)
⎢ Ω41 (2) 0 0 −R2 ∗ ⎥
⎢ ⎥
⎣ Ω51 0 0 0 −P ⎦

where

[ ]
Ω11 = Ψ + Γ + ΓT , Ψ = Ψi𝑗 4×4

∑
m

11 = PA + A P + Q1 + Q2 , Ψ11 = (1 − 𝜃)
Ψ11 T 21
K𝛼T BT P,
𝛼=1

11 = 𝜎W, Ψ22 = diag{−Q1 , 0, −Q2 },

Ψ22
[ 𝜎 ] [ ]
Ψ31 = (1 − 𝜃)K𝛼T BT P m W m×2 , Ψ41 = 𝜃K𝛼T BT P 0 m×2 ,
𝜎
Ψ33 = W · Im + diag{−𝜇1 W, … , −𝜇m W},
m2
√ √
Ψ44 = diag{−𝜃P, … , −𝜃P}, Ω21 (1) = 𝜏M N T , Ω21 (2) = 𝜏M M T ,
√ √ √
Ω21 (3) = dM T T , Ω21 (4) = dM ST , Ω31 (1) = 𝜏M R1 ,
√ √ √
Ω31 (2) = dM R2 , Ω41 (1) =
𝜏M R1 , Ω41 (2) = dM R2 ,
[ √ ]
Ω51 = 0 0 0 m𝜃Gmax P 0 · · · 0 0 ,
[ ]
Γ = N + T −N + M −M −T + S −S 0 · · · 0 ,
[ ∑ ]
 = A (1 − 𝜃) m ̄ ̄
𝛼=1 BK𝛼 0 0 0 (1 − 𝜃)K 𝜃 K ,
[ ∑ ]
 = 0 −𝛾 m ̄ ̄
𝛼=1 BK𝛼 0 0 0 −𝛾 K 𝛾 K ,
[ ]
K̄ = BK1 · · · BKm .

Proof. Construct the following Lyapunov functional as

V(t) = V1 (t) + V2 (t) + V3 (t), (11)

WANG ET AL. 7

where

V1 (t) = xT (t)Px(t)
t t
V2 (t) = xT (s)Q1 x(s) ds + xT (s)Q2 x(s)ds
∫t−𝜏M ∫t−dM
t t t t
. . . .
V3 (t) = xT (v)R1 x(v) dvds + xT (v)R2 x(v)dvds.
∫t−𝜏M ∫s ∫t−dM ∫s

From the definition of e𝛼 , 𝜏(t), x̄ and the triggering algorithm (3), it can be seen that, for t ∈ i ,

[ m ]T [ m ]
∑m
1 ∑ 1 ∑
𝜇𝛼 eT𝛼 We𝛼 ≤ 𝜎 (x(t − 𝜏(t))) + e𝛼 W (x(t − 𝜏(t))) + e𝛼 .
𝛼=1
m 𝛼=1 m 𝛼=1

From the definition of fi (x(t − d(t))) in (7), the following inequality exists:

∑
m
∑
m
T
− x (t − d(t))Gi PGi x(t − d(t)) + 𝑓iT (x(t − d(t)))P𝑓i (x(t − d(t))) ≤ 0.
i=1 i=1

Applying the infinitesimal operator defined in Definition 1 and taking expectation on both sides of (11), we can obtain

[ (m ]
∑ ∑
m
E((V(t))) ≤ 2xT (t)P Ax(t) + 𝜃B K𝛼 (𝑓𝛼 (x(t − d(t)))) + (1 − 𝜃)B K𝛼 (x(t − 𝜏(t)) − e𝛼 )
𝛼=1 𝛼=1

+ x (t)Q1 x(t) − x (t − 𝜏M )Q1 x(t − 𝜏M ) + x (t)Q2 x(t) − xT (t − dM )Q2 x(t − dM )

T T T

t t
. . . . . . . .
+ 𝜏M xT (t)R1 x(t) − xT (s)R1 x(s)ds + dM xT (t)R2 x(t) − xT (s)R2 x(s)ds
∫t−𝜏M ∫t−dM
[ t ]
.
− 2𝜉 T (t)N x(t) − x(t − 𝜏(t)) − x(s)ds − 2𝜉 T (t)M·
∫t−𝜏(t)
[ t−𝜏(t) ] [ t ]
. .
x(t − 𝜏(t)) − x(t − 𝜏M ) − x(s)ds − 2𝜉 T (t)T x(t) − x(t − d(t)) − x(s)ds
∫t−𝜏M ∫t−d(t)
[ ]
t−d(t)
.
− 2𝜉 T (t)S x(t − d(t)) − x(t − dM ) − x(s)ds
∫t−dM
[ ]T [ ]
∑
m
1∑
m
1∑
m
− 𝜇𝛼 eT𝛼 (t)We𝛼 (t) +𝜎 (e𝛼 + x(t − 𝜏(t))) W (e𝛼 + x(t − 𝜏(t)))
𝛼=1
m 𝛼=1 m 𝛼=1
∑ m
∑
m
+ 𝜃xT (t − d(t))Gi PGi x(t − d(t)) − 𝜃 𝑓iT (x(t − d(t)))P𝑓i (x(t − d(t))),
i=1 i=1

where
[ ]
𝜉 T (t) = xT (t) xT (t − 𝜏(t)) xT (t − 𝜏M ) xT (t − d(t)) xT (t − dM ) eT1 … eTm 𝑓1T … 𝑓mT .

Notice that
( . . ) ( )
E 𝜏M xT (t)R1 x(t) = 𝜉 T (t) ΩT31 (1)R−1
1 Ω31 (1) + Ω41 (1)R1 Ω41 (1) 𝜉(t)
T −1

( . . ) ( )
E dM xT (t)R2 x(t) = 𝜉 T (t) ΩT31 (2)R−1
2 Ω31 (2) + Ω41 (2)R2 Ω41 (2) 𝜉(t)
T −1
8 WANG ET AL.

Defining
[ . ]
𝜂 T (t, s) = 𝜉 T (t) xT (s) .

We can further conclude that

t t−𝜏(t)
1 1
E((V(t))) < 𝜂 T (t, s)F1 𝜂(t, s)ds + 𝜂 T (t, s)F2 𝜂(t, s)ds
∫
𝜏M t−𝜏(t) 𝜏M ∫t−𝜏M
t t−d(t)
1 1
+ 𝜂 T (t, s)F3 𝜂(t, s)ds + 𝜂 T (t, s)F4 𝜂(t, s)ds
∫
dM t−d(t) dM ∫t−dM

+ m𝜃xT (t − d(t))Gmax PGmax x(t − d(t)),

where
[ ]
Ω11 + Ω31 (1)T R−1 T −1
1 Ω31 (1) + Ω41 (1) R1 Ω41 (1) ∗
Fi = , i = {1, 2}.
Ω21 (i) −R1
[ ]
Ω11 + Ω31 (2)T R−1 Ω31 (2) + Ω41 (2)T R−1 Ω41 (2) ∗
F𝑗 = 2 2 , 𝑗 = {3, 4}.
Ω21 ( 𝑗) −R2

Using Schur Complement, one can obtain from (9) and (10) that, for t ∈ [tk h + 𝜏 k , tk+1 h + 𝜏 k+1 ],

E((V(t))) < 0.

This completes the proof.

Remark 3. Generally speaking, the conservatism of the results mainly comes from two aspects, the first is the selection
of the parameters 𝜇i and the second is the used Lyapunov functional technique. As for the first aspect, how to choose a
group of parameters 𝜇i to obtain an optimal result is still open. In this paper, we have tried different groups parameters
t . . t . .
to obtain a relative good one. For the second aspect, to deal with cross items ∫t−𝜏 xT (s)R1 x(s)ds and ∫t−d xT (s)R2 x(s)ds,
M M
a simple but efficient method has been used in the proof, however, the conservatism can be further improved by using
some new proposed methods, such as the method in the work of Zhang and Han.52
In the following, a criterion is provided to design the state-feedback gains under deception attacks and METS (3).
Theorem 2. For given parameters 𝜎, m, 𝜃, 𝛾, 𝜇 i , 𝜌1 , and 𝜌2 , system (6) under randomly occurring deception attacks and
METS is asymptotical stability, if there exist matrices X > 0, Q ̃ 2 > 0, R̃ 1 > 0, R̃ 2 > 0, W
̃ 1 > 0, Q ̃ > 0, Ñ, M,
̃ T, ̃
̃ S,
Y𝛼 , (𝛼 = 1, 2, … , m) of appropriate dimensions such that

⎡ Ξ11 ∗ ∗ ∗ ∗ ⎤
⎢ Ξ21 (i) −R̃ 1 ∗ ∗ ∗ ⎥
⎢ ⎥
Ξ(i) = ⎢ Ξ31 (1) 0 Φ(R̃ 1 ) ∗ ∗ ⎥ < 0, i = 1, 2 (12)
⎢ Ξ41 (1) 0 0 Φ(R̃ 1 ) ∗ ⎥
⎢ ⎥
⎣ Ξ51 0 0 0 −X ⎦

⎡ Ξ11 ∗ ∗ ∗ ∗ ⎤
⎢ Ξ21 ( 𝑗) −R̃ 2 ∗ ∗ ∗ ⎥
⎢ ⎥
Ξ( 𝑗) = ⎢ Ξ31 (2) 0 Φ(R̃ 2 ) ∗ ∗ ⎥ < 0, 𝑗 = 3, 4, (13)
⎢ Ξ41 (2) 0 0 Φ(R̃ 2 ) ∗ ⎥
⎢ ⎥
⎣ Ξ51 0 0 0 −X ⎦
WANG ET AL. 9

the memory controller gains are K𝛼 = Y𝛼 X−1 , where

[ ]
̃ + Γ̃ + Γ̃ T , Ψ
Ξ11 = Ψ ̃ = Ψ
̃ i𝑗 ̃ 11 = AX + XAT + Q
,Ψ ̃1 +Q
̃ 2,
4×4 11

∑
m
̃ 21 = (1 − 𝜃)
Ψ ̃ 22 = 𝜎 W,
Y𝛼T BT , Ψ ̃ Ψ̃ 22 = diag{−Q
̃ 1 , 0, −Q
̃ 2 },
11 11
𝛼=1

Φ(R̃ 1 ) = 𝜌21 R̃ 1 − 2𝜌1 X, Φ(R̃ 2 ) = 𝜌22 R̃ 2 − 2𝜌2 X,

[ 𝜎 ̃ ] 𝜎 ̃ { }
̃ 31 = (1 − 𝜃)Y𝛼T BT
Ψ W ̃ 33
,Ψ = ̃ … , −𝜇m W
W · Im + diag −𝜇1 W, ̃ ,
m m×2 m2
[ ]
Ψ̃ 41 = 𝜃Y𝛼T BT 0 , Ψ44 = diag{−𝜃X, … , −𝜃X},
m×2
√ √ √ √
T
Ξ21 (1) = 𝜏M Ñ , Ξ21 (2) = 𝜏M M ̃ T , Ξ21 (3) = dM T̃ T , Ξ21 (4) = dM S̃ T ,
[ ]
Γ̃ = Ñ + T̃ −Ñ + M ̃ −M ̃ −T̃ + S̃ −S̃ 0 0 · · · 0 ,
√ √ √ √
Ξ31 (1) = 𝜏M Ã, Ξ31 (2) = dM Ã, Ξ41 (1) = 𝜏M , ̃ Ξ41 (2) = dM , ̃
[ √ ]
Ξ51 = 0 0 0 m𝜃Gmax X 0 0 0 · · · 0 0 ,
m×(5+2m)

with
[ ∑ ]
à = AX (1 − 𝜃) m ̃ ̃
𝛼=1 BY𝛼 0 0 0 (1 − 𝜃)Y 𝜃 Y ,
[ ∑ ]
̃ = 0 −𝛾 m ̃ ̃
𝛼=1 BY𝛼 0 0 0 −𝛾 Y 𝛾 Y ,
[ ]
Ỹ = BY1 · · · BYm .

Proof. Define X = P−1 , premultiply and postmultiply (9) and (10) with diag(X, … , X, R−1
l
, R−1
l
, X), respectively, and
define new matrix variables Q̃ l = XQl X, R̃ l = XRl X(l = 1, 2), W
̃ = XWX, Υ̃ =diag(X, … , X)ΥX(Υ = N, M, T, S), (12)
and (13) can be obtained together with the following inequality:

−X R̃ −1 2̃
i X ≤ 𝜌i Ri − 2𝜌i X, i = 1, 2.

Remark 4. Theorem 2 has proposed a sufficient condition to deal with codesign problem of controller and METS
parameters, which are shown in terms of LMIs. Generally speaking, the interior-point LMI solvers are faster than
some convex optimization algorithms. The complexity of LMI computations is polynomial time which is bounded by
( 3 log(∕𝜀)), where 𝜀 is relative accuracy set for algorithm,  is data-dependant scaling factor,  is the total
row size of LMIs, and  is the number of scalar decision variables. Here, we assume that the system's dimension is
n and involved variable's dimension can be determined by x𝑗 (t) ∈ R and 𝑓i (x(t)) ∈ Rn , i ∈ {1, 2, … , m}. Then, for
Theorem 2,  = n(2m+8)+mn and  = n[3n+2+m+4(5+2m)]+5+m . Therefore, the computational complexity
of Theorem 2 can be expressed as (n7 ), which depends polynomially on the system's size.
For m = 1 and u(t) = Kx(tk h), a corollary can be easily derived from Theorem 2.
Corollary 1. For given parameters 𝜎, 𝜃, 𝛾, and 𝜌, system (6) is asymptotical stability if there exist matrices of X > 0,
Q ̃ 2 > 0, R̃ 1 > 0, R̃ 2 > 0, W
̃ 1 > 0, Q ̃ > 0, Ñ, M,
̃ T, ̃ Y of appropriate dimensions such that i = 1, 2, j = 3, 4:
̃ S,

⎡ Ξ̂ 11 ∗ ∗ ∗ ∗ ⎤
⎢ ̃ ̃ ⎥
Ξ
⎢ 21 (i) − R 1 ∗ ∗ ∗ ⎥
̂ = ⎢ Ξ̂ 31 (1) 0 Φ(R̃ 1 ) ∗
Ξ(i) ∗ ⎥ < 0, i = 1, (14)
⎢ ⎥
⎢ Ξ̂ 41 (1) 0 0 Φ(R̃ 1 ) ∗ ⎥
⎢ ̂ ⎥
⎣ Ξ51 0 0 0 −X ⎦
10 WANG ET AL.

⎡ Ξ̂ 11 ∗ ∗ ∗ ∗ ⎤
⎢̃ ̃ ⎥
⎢ Ξ21 ( 𝑗) −R2 ∗ ∗ ∗ ⎥
̂Ξ( 𝑗) = ⎢ Ξ̂ 31 (2) 0 Φ(R̃ 2 ) ∗ ∗ ⎥ < 0, 𝑗 = 3, 4, (15)
⎢ ⎥
⎢ Ξ̂ 41 (2) 0 0 Φ(R̃ 2 ) ∗ ⎥
⎢ ̂ ⎥
⎣ Ξ51 0 0 0 −X ⎦
where
[ ]
̂ + Γ̂ + Γ̂ T , Ψ
Ξ̂ 11 = Ψ ̂ = Ψ
̂ i𝑗 ̂ 11 = Ψ
,Ψ ̃ 11 , Ψ
̂ 22 = Ψ
̃ 22 ,
11 11 11 11
4×4
[ ]
Ψ ̃ 22 , Ψ
̂ 22 = Ψ ̂ 21 = (1 − 𝜃)Y T BT , Ψ
̂ 31 = (1 − 𝜃)Y T BT 𝜎 W
̃ ,
11
[ ]
̂ 33 = −W
Ψ ̃ + 𝜎 W,
̃ Ψ̂ 41 = 𝜃Y T BT 0 , Ψ̂ 44 = −𝜃X,
[ ]
Γ̂ = Ñ + T̃ −Ñ + M ̃ −M̃ −T̃ + S̃ −S̃ 0 0 ,
√ √ √
Ξ̂ 31 (1) = ̂ Ξ̂ 31 (2) = dM A,
𝜏M A, ̂
̂ Ξ̂ 41 (1) = 𝜏M ,
√ [ √ ]
Ξ̂ 41 (2) = ̂ Ξ̂ 51 = 0 0 0 𝜃GX 0 0 0 ,
dM ,
[ ]
 = AX (1 − 𝜃)BY 0 0 0 (1 − 𝜃)BY 𝜃BY ,
[ ]
̂ = 0 −𝛾BY 0 0 0 −𝛾BY 𝛾BY ,

the controller feedback gain can be designed as K = YX−1 .

4 AN ILLUSTRAT IVE EXAMPLE

Considering a pendulum system borrowed from Yue et al,24 the system parameters are

⎡0 1 0 0⎤ ⎡ 0 ⎤
⎢0 0 M 0⎥
−mg ⎢ 1 ⎥
A=⎢ ⎥, B = ⎢ M ⎥,
⎢0 0 0 1⎥ ⎢ 0 ⎥
⎢ g ⎥ ⎢ −1 ⎥
⎣0 0 l 0⎦ ⎣ Ml ⎦

where M = 10 is the cart mass and m = 1 is the mass of the pendulum bob, l = 3 is the length of the pendulum arm and
g = 10 is gravitational acceleration. The eigenvalue of A is {0, 0, 1.8257, −1.8257}; obviously, the system is unstable without
. .
a controller. The states [x1 x2 x3 x4 ] = [𝑦 𝑦 𝜔 𝜔], where xi (i = 1, 2, 3, 4), are the cart's position, the cart's velocity,
the pendulum bob's angle, and the pendulum's angular velocity, respectively. The initial state 𝜙(t) = [0.98 0 0.2 0]T .
In this example, the deception attacks are assumed to be

𝑓i (x(t)) = [tanh(0.05x1 (t); − tanh(0.1x2 (t)); tanh(0.05x3 (t); − tanh(0.1x4 (t))] .

The upper bound matrices Gi = Gmax = diag{0.05, 0.1, 0.05, 0.1} which satisfies the constraint ||fi (x(t − d(t))||2 ≤ ||Gi x(t −
d(t))||2 . Other parameters are set as dM = 0.06 s and 𝜌1 = 𝜌2 = 0.53.
In order to show the effectiveness of the proposed METS and the resilient control design method, two cases are proposed.
In Case 1, we assume that there is no deception attack in the system. In Case 2, when considering the randomly occurring
deception attacks, the simulation results show that the proposed resilient control method is effective.
Case 1. Suppose that there are no deception attacks occurring. Set 𝜏M = h + 𝜏̄ = 0.14 s, which shows a trade-off
between the sampling period h and allowable communication delay bound 𝜏. ̄ In this case, we choose h = 0.1 s and
WANG ET AL. 11

3.5 3.5

3 3

2.5 2.5
0.2
0.2
States response

States response
2 2
0
0
1.5 1.5
-0.2
-0.2
1 1
22 23 24 25 26 22 23 24 25 26
0.5 0.5

0 0

-0.5 -0.5
0 5 10 15 20 25 30 0 5 10 15 20 25 30
t(s) t(s)

(A) (B)

2.5 2

2
Release instants and release interval

Release instants and release interval

1.5

1.5
1

0.5
0.5

0
0

-0.5 -0.5
0 5 10 15 20 25 30 0 5 10 15 20 25 30
t(s) t(s)

(C) (D)

FIGURE 3 Simulation results of Case 1. (A) State responses of memory event-triggered scheme (METS) with m = 3; (B) State responses of
event-triggered scheme (ETS) in the work of Yue et al24 ; (C) Release intervals of METS with m = 3; (D) Release intervals of ETS in the work
of Yue et al24 [Colour figure can be viewed at [Link]]

𝜏̄ = 0.04 s. Firstly, recalling the ETS method in the work of Yue et al,24 for 𝜎 = 0.2, the state-feedback controller gain
and the triggering parameter W are obtained as

[ ]
K = 1.3305 8.2129 301.1510 167.1794 ,
⎡ 0.0009 0.0046 0.1514 0.0841 ⎤
⎢ 0.0046 0.0289 0.9162 0.5104 ⎥
W =⎢ ⎥.
⎢ 0.1514 0.9162 32.9649 18.3005 ⎥
⎢ ⎥
⎣ 0.0841 0.5104 18.3005 10.1673 ⎦

In this simulation, 55 packets are released by using the ETS in the work of Yue et al,24 the state responses and release
intervals are shown in Figure 3B and Figure 3D, respectively.
12 WANG ET AL.

Next, by using the proposed METS, for m = 3 and the weighting parameters 𝜇1 = 0.5, 𝜇 2 = 0.3, 𝜇3 = 0.2. Applying
Theorem 2, one can obtain the feedback gains and the triggering parameter W as
[ ]
K1 = 0.5090 3.2000 118.6998 65.9017 ,
[ ]
K2 = 0.3060 1.9231 71.3362 39.6064 ,
[ ]
K3 = 0.2045 1.2852 47.6650 26.4648 ,
⎡ 0.0001 0.0004 0.0144 0.0080 ⎤
⎢ 0.0004 0.0029 0.0867 0.0485 ⎥
W =⎢ .
0.0144 0.0867 3.1386 1.7420 ⎥
⎢ ⎥
⎣ 0.0080 0.0485 1.7420 0.9693 ⎦

In this case, by using the proposed METS, there are 68 packets released to the controller, the state responses and release
intervals of the METS are shown in Figure 3A and Figure 3C, respectively.
Compared the simulation results of the ETS in the work of Yue et al24 and METS, it can be found that more packets
are released by the METS at the beginning time, especially during the interval t ∈ [0, 5]; as a result, the settling time can
be obviously shortened. That is, the METS can improve the transient performance by adjusting the packet transmission
distribution according to the dynamics of the system.
Case 2. Considering the randomly occurring deception attacks, the occurring probability is chosen as 𝜃 = 0.1 and
𝜎 = 0.1. Set 𝜏M = h + 𝜏̄ = 0.11 s, h = 0.07 s and 𝜏̄ = 0.04 s. Firstly, if we do not utilize the historic release infor-
mation, by using Corollary 1, there is no feasible solution for the system with deception attacks. Next, by using the
proposed METS, for the same 𝜇i , (i = 1, 2, 3) in Case 1, by using Theorem 2, the controller feedback gains and triggering
parameters W are obtained as
[ ]
K1 = 1.4496 6.6060 147.6022 82.4861 ,
[ ]
K2 = 1.2446 5.5255 122.9560 68.6888 ,
[ ]
K3 = 1.0550 4.6204 102.6059 57.3089 ,
⎡ 0.0003 0.0008 0.0156 0.0087 ⎤
⎢ 0.0008 0.0034 0.0645 0.0363 ⎥
W = 104 × ⎢ .
0.0156 0.0645 1.3667 0.7637 ⎥
⎢ ⎥
⎣ 0.0087 0.0363 0.7637 0.4275 ⎦

3 1.2

2.5 1
Release instants and release interval

2 0.8
States response

1.5 0.6

1 0.4

0.5 0.2

0 0

-0.5 -0.2
0 5 10 15 20 25 30 0 5 10 15 20 25 30
t(s) t(s)

(A) (B)

FIGURE 4 Simulation results of Case 2. (A) State responses of memory event-triggered scheme (METS) with m = 3; (B) Release intervals
of METS with m = 3; [Colour figure can be viewed at [Link]]
WANG ET AL. 13

0.14
Attacks on x1(t)
0.12 Attacks on x2(t)
Attacks on x3(t)
Attacks on x4(t)
0.1

0.08

0.06

0.04

0.02

-0.02
FIGURE 5 The occurring instants of the random deception
0 5 10 15 20 25 30 attacks [Colour figure can be viewed at [Link]]

The state responses and release instants are shown in Figure 4 and the deception attack instants are illustrated in
Figure 5. From Figures 4 and 5, it can be found that, by using the proposed METS and resilient memory control method,
the system can be stabilized in the case of random deception attacks.

5 CO N C LU S I O N

In this paper, a METS has been proposed for the resilient control of NCSs under deception attacks. The information of
some historic released packets has been employed in the proposed METS to trigger more packets at some special instants,
especially, at the crest or trough of the system curve. Considering both the METS and deception attacks, new kind of
networked control systems model has been built. Based on a Lyapunov functional method, a resilient memory control
technique has been designed, sufficient conditions for the asymptotical stability have been derived. Then, controller feed-
back gains and triggering matrix can be codesigned by solving some LMIs. A simulation example has been proposed to
demonstrate the effectiveness of the proposed design method.

ACKNOWLEDGEMENTS
This work was supported in part by the National Natural Science Foundation of China (61773218 and 61903252), Natural
Science Foundation of Jiangsu Province (BK20161561), and the Program for Professor of Special Appointment (Eastern
Scholar) at Shanghai Institutions of Higher Learning.

REFERENCES
1. Ding L, Han Q, Zhang X. Distributed secondary control for active power sharing and frequency regulation in islanded microgrids using
an event-triggered communication mechanism. IEEE Trans Ind Inform. 2018;15(7):3910-3922.
2. Peng C, Zhang J, Han Q-L. Consensus of multiagent systems with nonlinear dynamics using an integrated sampled-data-based
event-triggered communication scheme. IEEE Trans Syst Man Cybern Syst. 2018;49(3):589-599.
3. Wang Y, Lei Y, Bian T, Guan Z. Distributed control of nonlinear multiagent systems with unknown and nonidentical control directions
via event-triggered communication. IEEE Trans Cybern. 2019. [Link]
4. Wang Y, Wei Y, Liu X, Zhou N, Cassandras CG. Optimal persistent monitoring using second-order agents with physical constraints. IEEE
Trans Autom Control. 2019;64(8):3239-3252.
5. Wang Y, Han Q. Network-based modelling and dynamic output feedback control for unmanned marine vehicles in network environments.
Automatica. 2018;91:43-53.
6. Wang Y, Han Q, Fei M, Peng C. Network-based T-S fuzzy dynamic positioning controller design for unmanned marine vehicles. IEEE
Trans Cybern. 2018;48(9):2750-2763.
7. Yang D, Zong G, Karimi HR. H∞ refined anti-disturbance control of switched LPV systems with application to aero-engine. IEEE Trans
Ind Electron. 2019. [Link]
8. Zhang D, Han Q, Zhang X. Network-based modeling and proportional-integral control for direct-drive-wheel systems in wireless network
environments. IEEE Trans Cybern. 2019. [Link]
14 WANG ET AL.

9. Zhang X, Han Q, Ge X, et al. Networked control systems: a survey of trends and techniques. IEEE/CAA J Autom Sin. 2019. [Link]
10.1109/JAS.2019.1911651
10. Cloosterman M, Wouw VDN, Heemels W, Nijmeijer H. Robust control of networked control systems, with uncertain time-varying delays.
IEEE Trans Autom Control. 2006;54(7):1575-1580.
11. Huang S, He X, Zhang N. New results on H∞ filter design for nonlinear systems with time delay via T-S fuzzy models. IEEE Trans Fuzzy
Syst. 2011;19(1):193-199.
12. Jiang X, Han Q, Liu S. A new H∞ stabilization criterion for networked control systems. IEEE Trans Autom Control. 2008;53(4):1025-1032.
13. Liang X, Xu J, Zhang H. Optimal control and stabilization for networked control systems with packet dropout and input delay. IEEE Trans
Circuits Syst II Express Briefs. 2016;64(9):1087-1091.
14. Ren H, Zong G, Karimi HR. Asynchronous finite-time filtering of networked switched systems and its application: an event-driven method.
IEEE Trans Circuits Syst I Regul Pap. 2018;66(1):391-402.
15. Tian E, Wang Z, Zou L, Yue D. Chance-constrained H-infinity control for a class of time-varying systems with stochastic nonlinearities:
the finite-horizon case. Automatica. 2019;107:296-305.
16. Zhang S, Xie D, Yan W. Decentralized event-triggered consensus control strategy for leader-follower networked systems. Phys A Stat Mech
Appl. 2017;479:498-508.
17. Zhang WA, Yu L. Modelling and control of networked control systems with both network-induced delay and packet-dropout. Automatica.
2008;44(12):3206-3210.
18. Zhang X, Han Q, Wang J. Admissible delay upper bounds for global asymptotic stability of neural networks with time-varying delays.
IEEE Trans Neural Netw Learn Syst. 2018;29(11):5319-5329.
19. Gawthrop PJ, Wang L. Event-driven intermittent control. Int J Control. 2009;82(12):2235-2248.
20. Gu Z, Zhang T, Zhao H, Yang F, Shen M. A novel event-triggered mechanism for networked cascade control system with stochastic
nonlinearities and actuator failures. J Frankl Inst. 2019;356(4):1955-1974.
21. Heemels WPMH, Donkers MCF, Teel AR. Periodic event-triggered control for linear systems. IEEE Trans Autom Control.
2013;58(4):847-861.
22. Lunze J, Lehmann D. A state-feedback approach to event-based control. Automatica. 2010;46(1):211-215.
23. Peng C, Wu M, Xie X, Wang Y. Event-triggered predictive control for networked nonlinear systems with imperfect premise matching.
IEEE Trans Fuzzy Syst. 2018;26(5):2797-2806.
24. Yue D, Tian E, Han Q. A delay system method for designing event-triggered controllers of networked control systems. IEEE Trans Autom
Control. 2013;58(2):475-481.
25. Hu S, Yue D, Xie X, Ma Y, Yin X. Stabilization of neural-network-based control systems via event-triggered control with nonperiodic
sampled data. IEEE Trans Neural Netw Learn Syst. 2018;29(3):573-585.
26. Ren H, Zong G, Li T. Event-triggered finite-time control for networked switched linear systems with asynchronous switching. IEEE Trans
Syst Man Cybern Syst. 2018;48(11):1874-1884.
27. Zong G, Ren H. Guaranteed cost finite-time control for semi-Markov jump systems with event-triggered scheme and quantization input.
Int J Robust Nonlinear Control. 2019;29(15):5251-5273.
28. Li F, Fu J, Du D. A novel decentralised event-triggered control for network control systems with communication delays. Int J Syst Sci.
2016;47(13):3098-3115.
29. Donkers MCF, Heemels WPMH. Output-based event-triggered control with guaranteed L∞ -gain and improved and decentralized
event-triggering. IEEE Trans Autom Control. 2012;57(6):1362-1376.
30. Liu J, Wei L, Xie X, Tian E, Fei S. Quantized stabilization for T-S fuzzy systems with hybrid-triggered mechanism and stochastic
cyber-attacks. IEEE Trans Fuzzy Syst. 2018;26(6):3820-3834.
31. Tian E, Wang Z, Zou L, Yue D. Probabilistic-constrained filtering for a class of nonlinear systems with improved static event-triggered
communication. Int J Robust Nonlinear Control. 2019;29(5):1484-1498.
32. Girard A. Dynamic triggering mechanisms for event-triggered control. IEEE Trans Autom Control. 2013;60(7):1992-1997.
33. Hu S, Yue D, Yin X, Xie X, Ma Y. Adaptive event-triggered control for nonlinear discrete-time systems. Int J Robust Nonlinear Control.
2016;26(18):4104-4125.
34. Ge X, Han Q. Distributed formation control of networked multi-agent systems using a dynamic event-triggered communication mecha-
nism. IEEE Trans Ind Electron. 2017;64(10):8118-8127.
35. Ge X, Han Q, Yang F. Event-based set-membership leader-following consensus of networked multi-agent systems subject to limited
communication resources and unknown-but-bounded noise. IEEE Trans Ind Electron. 2017;64(6):5045-5054.
36. Gu Z, Shi P, Yue D, Ding Z. Decentralized adaptive event-triggered H∞ filtering for a class of networked nonlinear interconnected systems.
IEEE Trans Cybern. 2018;49(5):1570-1579.
37. Li T, Zhang L, Li Z, Yuan R. Adaptive event-triggered output feedback control of networked control systems. Paper presented at: Eighth
International Conference on Intelligent Control and Information Processing; 2017; Hangzhou, China.
38. Peng C, Zhang J, Yan H. Adaptive event-triggering H∞ load frequency control for network-based power systems. IEEE Trans Ind Electron.
2018;65:1685-1694.
39. Guo Z, Shi D, Johansson KH, Shi L. Optimal linear cyber-attack on remote state estimation. IEEE Trans Control Netw Syst. 2017;4(1):4-13.
40. Ge X, Han Q, Zhong M, Zhang X. Distributed Krein space-based attack detection over sensor networks under deception attacks.
Automatica. 2019. [Link]
WANG ET AL. 15

41. Hu S, Yue D, Xie X, Chen X, Yin X. Resilient event-triggered controller synthesis of networked control systems under periodic DOS
jamming attacks. IEEE Trans Cybern. 2018;49(12):4271-4281.
42. Li X, Zhang W, Wang Y. Simultaneous fault estimation for Markovian jump systems with generally uncertain transition rates: a
reduced-order observer approach. IEEE Trans Ind Electron. 2019. [Link]
43. Peng C, Li JC, Fei M. Resilient event-triggered H∞ load frequency control for networked power systems with energy-limited DoS attacks.
IEEE Trans Power Syst. 2017;32(5):4110-4118.
44. Teixeira A, Sou KC, Sandberg H, Johansson KH. Secure control systems: a quantitative risk management approach. IEEE Control Syst.
2015;35(1):24-45.
45. Wang Y, Wang HO, Xiao J, Guan Z. Synchronization of complex dynamical networks under recoverable attacks. Automatica.
2010;46(1):197-203.
46. Jithish J, Sankaran S. Securing networked control systems: modeling attacks and defenses. Paper presented at: IEEE International
Conference on Consumer Electronics-Asia; 2017; Bangalore, India.
47. Liu J, Gu Y, Zha L, Liu Y, Cao J. Event-triggered H∞ load frequency control for multiarea power systems under hybrid cyber attacks. IEEE
Trans Syst Man Cybern Syst. 2019. [Link]
48. Liu J, Yang M, Tian E, Cao J, Fei S. Event-based security control for state-dependent uncertain systems under hybrid-attacks and its
application to electronic circuits. IEEE Trans Circuits Syst Regul Pap. 2019. [Link]
49. Ma L, Wang Z, Han Q-L. Variance-constrained distributed filtering for time-varying systems with multiplicative noises and deception
attacks over sensor networks. IEEE Sens J. 2017;17(7):2279-2288.
50. Rodriguez-Seda EJ, Severson T, Kiriakidis K. Recovery after attacks of deception on networked control systems. Paper presented at:
Resilience Week; 2016; Chicago, IL.
51. Liu J, Xia J, Tian E, Fei S. Hybrid-driven-based H∞ filter design for neural networks subject to deception attacks. Appl Math Comput.
2018;320:158-174.
52. Zhang X, Han Q. State estimation for static neural networks with time-varying delays based on an improved reciprocally convex inequality.
IEEE Trans Neural Netw Learn Syst. 2018;29(4):1376-1381.

How to cite this article: Wang K, Tian E, Liu J, Wei L, Yue D. Resilient control of networked control sys-
tems under deception attacks: A memory-event-triggered communication scheme. Int J Robust Nonlinear Control.
2019;1–15. [Link]