Electric Power Systems Research 236 (2024) 110912

Contents lists available at ScienceDirect

Electric Power Systems Research

journal homepage: www.elsevier.com/locate/epsr

Mitigating the impact of false data in wide area control of power systems
Priyatosh Mahish a ,∗, Ashok Kumar Pradhan c , Sukumar Mishra b
a
School of Computing and Electrical Engineering, Indian Institute of Technology, Mandi, India
b
Department of Electrical Engineering, Indian Institute of Technology, Delhi, India
c
Department of Electrical Engineering, Indian Institute of Technology, Kharagpur, India

ARTICLE INFO ABSTRACT

Keywords: The potential of wide area control (WAC) to improve power system stability encourages utility incorporating
Cyber attack phasor measurement units (PMUs) at different nodes. Higher usage of PMU makes the power network more
Wide area control vulnerable to cyber attack (CA) which is a concern for the system operator. Mitigation of the CA is important
Phasor measurement unit
for any WAC operation. In this paper, a false data mitigation (FDM) strategy is proposed for WAC to maintain
False data injection
stability of the power system even during CA. The proposed FDM filters out the data received from the attacked
State space model
PMUs, if any, and accordingly modifies the WAC model in the phasor data concentrator. The WAC continues its
operation with the modification and using the data from the unattacked PMUs. The proposed FDM technique
is tested in the New England 39 bus system and compared with the available dynamic state estimation (DSE)
method.

1. Introduction control signal at an operating system. The FDI attack manipulates the
power system measurement data for malfunction of monitoring and
Today power system exploits the usage of synchrophasor data based control system. CP attack is significantly different from FDI or DOS
monitoring and wide area control (WAC). The communication system, attack. In CP, the physical network of power system is manipulated.
which is backbone of WAC, is integrated part of a power system. Depen- On the other hand, in both the FDI and DOS, attackers play with
dency on communication system widens the way for cyber attack (CA) the network information [16]. An overview of different FDI attack is
leading to the power system in vulnerability [1–3]. Several incidences
available in [17].
of CA in power system are reported. The slammer worm was injected
To overcome the different type of CA issues, the literatures provide
in a local computer network at Ohio’s Davis-Besse nuclear power plant
the solutions which are classified into two categories [18,19]: (i) pro-
in USA in January, 2003 [4]. Following the attack, the monitoring
tection and (ii) detection based approach. The objective of the former
system was deactivated for around 5 h. In 2010, Natanz nuclear fuel-
enrichment facility was attacked with injecting Stuxnet worm in their category is to make specific meters secured by encryption or temper
computer system [5]. A well planned CA was applied to the supervi- proof communication system [20]. In [21], a least-budget defense
sory control and data acquisition (SCADA) system, resulting Ukraine strategy is available which solves a mixed integer nonlinear programme
blackout on 23 December, 2015 [6]. An intelligent manipulation in to find the critical meters to be protected from CA. The later category
the enrgy measurements has been implemented in an utility company, detects anomaly to identify the manipulated data [22,23]. In [22], a
resulting loss of approximately $100 million [7,8]. To avoid such events transmission line impedance model based technique is established to
in future, North American Electric Reliability Corporation (NERC) rec- detect CA. State estimation based CA detection techniques are also
ommends several standards on Critical Infrastructure Protection (CIP) effective and well-known [24]. Kullback–Leibler distance method is
subject to future enforcement, such as cyber security (CS) on informa- useful to estimate the difference between the true and false data [25,
tion protection [9], CS on electronic security perimeters (ESP) [10], and 26]. However, these techniques fail to capture the dynamics of nonlin-
so on [11]. The ESP must adapt with multiple approaches to detect and ear power system, due to slow update rates of supervisory control and
mitigate known/unknown/suspected/malicious/false communications
data acquisition (SCADA) system. In [23], a dynamic state estimator
for both incoming and outgoing channels.
(DSE) is established to detect CA, which uses phasor measurement unit
CAs are typefied as denial of service attack (DOS) [12], false data
(PMU) data to estimate the power system states through sliding mode
injection attack (FDI) [13] and cyber–physical (CP) attack [14,15]. For
DOS attack, the target of attacker is to compromise the availability of observer (SMO) based model.

∗ Corresponding author.
E-mail addresses: [email protected] (P. Mahish), [email protected] (A.K. Pradhan), [email protected] (S. Mishra).

https://doi.org/10.1016/j.epsr.2024.110912
Received 29 February 2024; Received in revised form 13 July 2024; Accepted 16 July 2024
Available online 24 July 2024
0378-7796/© 2024 Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
P. Mahish et al. Electric Power Systems Research 236 (2024) 110912

Fig. 1 shows the block diagram of a centralized WAC [32]. In this

diagram, two different components of WAC is represented. Each com-
ponent receives the input signal from their pre-scheduled PMU and
other component of WAC. The WAC model provides set of state space
equation of each component. The individual components are designed
for the respective generator substation/bus. A PDC combines these sets
to represent the centralized WAC. Supercript 𝑐 stands for controller.
𝑥𝑐1 and 𝑥𝑐2 are state vector matrix of the WAC component-1 and 2,
respectively. 𝑦1 and 𝑦2 are input vector matrix of the components,
which are calculated with the measurements of PMU1 and PMU2 ,
respectively. 𝑦𝑐1 and 𝑦𝑐2 are control output vector matrix of the WAC
component-1 and 2, respectively. Solving the set of equations for WAC
component-1 and component-2 provided in Fig. 1, processing of WAC
Fig. 1. Block diagram of the centralized WAC.
can be centralized in the PDC as in (1) and (2). N = (1 − Dc12 Dc21 )−1 .
The method for parameter calculation of WAC is explained in [32]. In
the present work, rotor speed (𝜔) of the synchronous generators is the
Following detection, mitigation of FDI attack is required to main- component of state vector matrix of WAC. The voltage, current and
tain healthy condition of the power grid. A detection and mitigation angle information of the PMU connected generator buses are used to
approach of FDI on real time electricity market is reported in [27]. A calculate active and reactive power injection in the respective buses.
mixed integer linear programming based method is applied for solving These data are sent to PDC. The WAC generates electromagnetic torque
cyber-secured unit commitment problem [28]. The model based FDI (𝑇 ) signal which is used in the 𝛥𝜔-type power system stabilizer (PSS)
mitigation techniques are effectively applied for automatic generation at the generators to improve power system stability. Thereby, in the
control [29,30]. However, these methods use SCADA system and un- WAC component-1, 𝑥𝑐1 ⟹ 𝜔1 , 𝑦1 ⟹ 𝑃1 , and 𝑦𝑐1 ⟹ 𝑇1 ,
suitable for primary frequency control. The DSE method uses estimated where 𝜔1 , 𝑃1 , and 𝑇1 are the rotor speed of generator G1, active power
data using healthy PMU measurements to maintain primary frequency
injection at G1-connected bus, and electromagnetic torque of generator
regulation even during FDI attack [23]. This approach eliminates at-
G1, respectively. Similarly, WAC component-2 is also defined for the
tacked sensors, ensuring observability through sufficient healthy PMUs.
generator G2. For the 𝑛-generator based power system, the centralized
The DSE approximates power system model which can be erroneous for
WAC is extended up to 𝑛th component which is explained in Section 5.
highly nonlinear grid [31].
In this paper, a false data mitigation (FDM) strategy is proposed
in WAC scheme to mitigate the effect of FDI attacked PMUs in the 3. False data detection method
power systems. Following FDI detection, the proposed FDM eliminates
the involvement of the attacked PMUs, and simultaneously modifies the Detection of FDI attack for individual PMU is necessary before
state space model of the WAC in phasor data concentrator (PDC). Thus, processing the proposed FDM. Therefore, each control input signal of
the WAC operation remains secured from FDI. The proposed method the WAC is passed through a false data detection (FDD) filter [23,33].
uses healthy PMU data for the FDI mitigation process. Thus, the FDM The filter for control input of 𝑘th component of the WAC is expressed
provides a solution for continuing WAC operation, even if PMU data as,
estimation is erroneous due to unobservability of the power network
𝑓̇𝑘𝑐 (𝑡) = (A𝑐𝑘 + A𝑐𝑘 C𝑐𝑘 𝑇 C𝑐𝑘 )𝑓𝑘𝑐 (𝑡) + A𝑐𝑘 C𝑐𝑘 𝑇 𝑦𝑘 (𝑡)
with the available unattacked PMUs. The proposed FDM is tested for (3)
New England 39 bus system. Performance of the proposed technique is 𝑦𝑘 (𝑡) = 𝑎𝑘 (𝑡) − C𝑐𝑘 𝑓𝑘𝑐 (𝑡)
compared with available DSE method and found to be more effective where, 𝑓𝑘𝑐 is state vector matrix of the FDD filter for 𝑘th component of
for single and multiple PMU attack scenarios. Contribution of the paper the centralized WAC. 𝐴𝑐𝑘 and 𝐶𝑘𝑐 are the state matrix and output matrix
is summarized below. for 𝑘th component of WAC, respectively. 𝑦𝑘 is input vector matrix of
1. Extending the centralized WAC model up to 𝑛 number of compo- the 𝑘th component of WAC, which is measured by PMU at 𝑘th bus,
nents. i.e. PMU𝑘 . 𝑎𝑘 is the CA vector at PMU𝑘 . During no CA, 𝑎𝑘 (𝑡) = 0. In
2. Modification of the 𝑛-component based WAC model for elimina- such situation, (3) becomes,
tion of the control inputs from the attacked PMUs. 𝑓̇𝑘𝑐 (𝑡) = A𝑐𝑘 𝑓𝑘𝑐 (𝑡) (4)

2. Conventional WAC methodology involving two generator buses Thus, during no CA, the FDD becomes an open loop system and it has
no effect on the WAC performance.
A centralized WAC receives input signals from multiple PMUs.
Before using these signals at WAC, a PDC arranges sequence of data 3.1. Brief explanation of FDD mechanism [23]
of the signals using time information.
[ c ] [ c ][ c ]
𝑥̇ 1 (t) A1 + NBc12 Cc1 Dc21 NBc12 Cc2 𝑥1 (t) In (3), initial state 𝑓𝑘𝑐 (𝑡0 ) is provided by SMO. The SMO for the
= linearized FDD, as in (3), is expressed as,
𝑥̇ c2 (t) NBc21 Cc1 Ac2 + NBc21 Cc2 Dc12 𝑥c2 (t)
[ c ] [ ] (1) 𝑓̂𝑘𝑐 (𝑡) = A𝑐𝑘 𝑓̂𝑘𝑐 (𝑡) + 𝐿𝑘 (𝑦𝑘 (𝑡) − 𝑦̂𝑘 (𝑡)) − 𝐵𝑤 𝐸(𝑦̂𝑘 , 𝑦𝑘 , 𝜂)
B + NBc12 Dc11 Dc21 NBc12 Dc22 𝑦1 (t) (5)
+ 11 𝑦̂𝑘 = C𝑐 𝑓̂𝑐 (𝑡)
NBc21 Dc11 Bc22 + NBc21 Dc22 Dc12 𝑦2 (t) 𝑘 𝑘

In (5), selection of 𝐵𝑤 should be such that it must satisfy rank(C𝑐𝑘 𝐵𝑤 )

[ ] [ ][ ]
𝑦c1 (t) NCc1 NCc2 Dc12 𝑥c1 (t) = rank(𝐵𝑤 ).
= c
𝑦c2 (t) NCc1 Dc21 NCc2 𝑥2 (t) ⎧ 𝜂𝐹𝑘 (𝑦𝑘 −𝑦̂𝑘 )
[ ][ ] (2) ⎪ ‖𝐹 𝑦 −𝑦̂ ‖ +𝑣 , if 𝐹𝑘 (𝑦𝑘 − 𝑦̂𝑘 ) > 𝜖
NDc11 NDc12 Dc22 𝑦1 (t) 𝐸(𝑦̂𝑘 , 𝑦𝑘 , 𝜂) = ⎨ 𝑘 𝑘 𝑘 2 (6)
+ ⎪0, if 𝐹𝑘 (𝑦𝑘 − 𝑦̂𝑘 ) ≤ 𝜖.
NDc21 Dc11 NDc22 𝑦2 (t) ⎩

2
P. Mahish et al. Electric Power Systems Research 236 (2024) 110912

4. Problem statement and objective

In Section 3, the estimation based approach is explained for FDD.

During FDM process the same technique can be used to provide esti-
mated data as the substitute of attacked PMU measurements. However,
the effectiveness of this approach is vulnerable to accuracy level of the
estimation, as depicted in Fig. 2. Thereby, estimation based approach,
such as DSE [23], is vulnerable to accuracy level of the estimation
which may degrade the performance of the attack mitigation approach.
To overcome this issue, a suitable method is required to mitigate CA
vector in WAC, which is robust/independent to the estimation error
during false data attack at the PMUs.
Fig. 2. Deviation in estimated data at SMO w.r.t. measured data at PMU𝑘 during CAs.
The objective of the paper is to provide robust FDM models of
WAC at single/multiple PMU attacked scenario for different conditions
of the power system. These models avoid estimation of the attacked
where, 𝜖 is the permissible estimation error. 𝜂 > 1 is the gain of SMO PMU measured data, to overcome the impact of errors in the estimated
and 𝑣 is a small positive number used to control the convergence speed. data. Formulating such models requires proper identification, detection
𝐹𝑘 ∈ R𝑛x𝑘 satisfies the following matrix equality. and elimination of the attacked PMUs, which is possible through the
techniques available in [21–26]. In this paper, the method in [23] is
𝐹𝑘 C𝑐𝑘 = 𝐵𝑤
𝑇
𝐻 (7)
used for FDD purpose. Thereby, attacked PMU data estimation is used
𝐿𝑘 ∈ R𝑘x𝑛 is selected to make sure the asymptotic stability of 𝑓𝑘𝑐 −𝐿𝑘 C𝑐𝑘 . for FDD only. Further, in online, formulation of the proposed FDM
This can be satisfied by Lyapunov matrix equation as follows, models demands high computational processors which is expected to
be available at the PDC.
(𝑓𝑘𝑐 − 𝐿𝑐𝑘 C𝑐𝑘 )𝑇 𝐻 + 𝐻(𝑓𝑘𝑐 − 𝐿𝑐𝑘 C𝑐𝑘 ) = −𝑄 (8)
5. Proposed FDM in WAC involving 𝒏 number of generator buses
For any positive definite matrix 𝑄, there is an unique symmetric
positive definite matrix 𝐻 ∈ R𝑘x𝑘 , such that (8) satisfies. The equations Followed by the false data detection with the method explained in
expressed above are solved with Matlab LMI Solver. It is worth to be Section 3, the proposed FDM is applied to mitigate the CA vector in
noted that 𝐿𝑘 is well defined during healthy situation of WAC and 𝐿𝑘 WAC. Extending the centralized WAC in Fig. 1 up to 𝑛th component, the
is unknown to the attacker. expression of 𝑘th component is formulated with combination of (10)
and (11).
3.2. CA modeling ∑
𝑛
𝑥̇𝑐𝑘 = A𝑐𝑘 𝑥𝑐𝑘 + B𝑐𝑘𝑘 𝑦𝑘 + B𝑐𝑘𝑖 𝑦𝑐𝑖 ; 𝑖 ≠ 𝑘
𝑖=1
Each CA vector is designed for an individual PMU. A CA vector (10)
∑
𝑛
attacks the respective PMU, and injects a combination of random data = A𝑐𝑘 𝑥𝑐𝑘 + B𝑐𝑘𝑘 𝑦𝑘 − B𝑐𝑘𝑘 𝑦𝑐𝑘 + B𝑐𝑘𝑖 𝑦𝑐𝑖
with a sinusoidal function. By injecting random data, first objective 𝑖=1

of the attacker is to create a confusion among the bad data and the ∑
𝑛
𝑦𝑐𝑘 = C𝑐𝑘 𝑥𝑐𝑘 + D𝑐𝑘𝑘 𝑦𝑘 + D𝑐𝑘𝑖 𝑦𝑐𝑖 ; 𝑖≠𝑘
CA. In general, the attacker may not know the setting of FDD, such as 𝑖=1
(11)
permissible estimation error 𝜖. As a result, the FDD can discriminate ∑
𝑛

between a series of bad data with the random attack. By injecting = M𝑘 C𝑐𝑘 𝑥𝑐𝑘 + M𝑘 D𝑐𝑘𝑘 𝑦𝑘 + M𝑘 D𝑐𝑘𝑖 𝑦𝑐𝑖
𝑖=1
sinusoidal CA vector, second objective of the attacker is to make a con-
fusion between a natural disturbance event with the CA. Though such where, M𝑘 = (1 + D𝑐𝑘𝑘 )−1 . 𝑥𝑐𝑘 is state vector matrix of the WAC
sinusoidal vectors are easy to be detected in a steady state condition component-𝑘. 𝑦𝑐𝑘 , and 𝑦𝑐𝑖 are control output vector matrix of 𝑘th compo-
of power systems, the challenge arises for introducing the CAs after nent, and 𝑖th component of the centralized WAC, respectively. B𝑐𝑘𝑘 , and
a natural disturbance event, if the system states oscillate during the B𝑐𝑘𝑖 are the input matrix of 𝑘th input, and 𝑖th input of WAC component-
attack. 𝑘, respectively. D𝑐𝑘𝑘 , and D𝑐𝑘𝑖 are the feed-forward matrix of 𝑘th input,
and 𝑖th input of WAC component-𝑘, respectively. 𝑛 is the number of
PMUs involved with the centralized WAC, which is also the number of
3.3. SMO performance analysis for different CA vectors components of the WAC. Combining the 𝑛 number of components, the
linearized WAC is expressed as,
Three different CA vectors 𝑎𝑘1 , 𝑎𝑘2 and 𝑎𝑘3 are simulated to verify 𝑐
Ẋ = A𝑐 X𝑐 + BY + B𝑐 Y𝑐
the performance of SMO on detecting the attack. (12)
Y𝑐 = C𝑐 X𝑐 + DY + D𝑐 Y𝑐
𝑎𝑘1 (𝑡) = 0.1sin(3𝑡) + rand(1, 60)
where, X𝑐 = [𝑥𝑐1 ...𝑥𝑐𝑛 ]𝑇 ; Y = [𝑦1 ...𝑦𝑛 ]𝑇 ; Y𝑐 = [𝑦𝑐1 ...𝑦𝑐𝑛 ]𝑇 ;
𝑎𝑘2 (𝑡) = sin(2𝑡) + rand(1, 60) (9)
⎡A𝑐1 ⎤ ⎡B𝑐11 ⎤ ⎡C𝑐1 ⎤
𝑎𝑘3 (𝑡) = 10sin(𝑡) + rand(1, 60) ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
A𝑐 = ⎢ ⋱ ⎥ ; B = ⎢ ⋱ ⎥ ; C𝑐
= ⎢ ⋱ ⎥;
The CAs are simulated for the duration of 2 s. In (9), 𝑎𝑘1 , 𝑎𝑘2 and 𝑎𝑘3 ⎢ 𝑐⎥ ⎢ 𝑐 ⎥ ⎢ 𝑐⎥
consist of 60 false data each. These data interact with the 𝑘th PMU ⎣ A𝑛 ⎦ ⎣ B𝑛𝑛 ⎦ ⎣ C𝑛 ⎦
measurement, as in (3). The PMUs send data to PDC at 30 frames per ⎡D𝑐11 ⎤ ⎡0 D𝑐12 ... D𝑐1𝑛 ⎤
second. Fig. 2 shows the deviation in estimated data at SMO w.r.t. ⎢ ⎥ 𝑐 ⎢ ⎥
D=⎢ ⋱ ⎥ ; D = ⎢⋮ ⋮⋱ ⋮ ⎥ ;
measured data at PMU𝑘 , during CAs. 𝑡𝑑𝑒𝑡𝑒𝑐𝑡 is the time to detect the CAs. ⎢ ⎥ ⎢ 𝑐 ⎥
⎣ D𝑐𝑛𝑛 ⎦ 𝑐
⎣ D𝑛1 D𝑛2 ... 0 ⎦
𝑡𝑑𝑒𝑡𝑒𝑐𝑡 depends on permissible estimation error 𝜖 as in (6). The detection
time increases for higher value of 𝜖. On the other hand, for much ⎡0 B𝑐12 ... B𝑐1𝑛 ⎤
⎢ ⎥
smaller 𝜖, the FDD may misinterpret the noise in PMU measurements B𝑐 = ⎢⋮ ⋮⋱ ⋮ ⎥
as CA vector. Therefore, suitable choice of 𝜖 is necessary to detect CAs. ⎢ 𝑐 𝑐 ⎥
⎣ B𝑛1 B𝑛2 ... 0 ⎦

3
P. Mahish et al. Electric Power Systems Research 236 (2024) 110912

In the remaining part of the paper, X𝑐 ⟹ [𝜔1 ...𝜔𝑛 ]𝑇 , Y ⟹ [𝑃1 ...𝑃𝑛 ]𝑇 expressed by eliminating 𝑃𝑘 and 𝑃𝑙 as,
and Y𝑐 ⟹ [𝑇1 ...𝑇𝑛 ]𝑇 . Where, 𝜔𝑛 , 𝑃𝑛 , and 𝑇𝑛 are the rotor speed of 𝑛th ∑
𝑛 ∑
𝑛
generator, active power injection at 𝑛th generator-connected bus, and 𝜔̇ 𝑘 = (A1 )𝑘𝑘 𝜔𝑘 + (A2 )𝑘𝑖 𝜔𝑖 + (B1 )𝑘𝑖 𝑃𝑖 + (B7 )𝑘𝑙 𝑇𝑘
electromagnetic torque of 𝑛th generator, respectively. 𝑖=1 𝑖=1
( ) (18)
Substituting (11) in (10), for 𝑘th generator, ∑
𝑛
+ (B8 )𝑘𝑙 𝑇𝑙 + (B9 )𝑘𝑖 + (B9𝑙 )𝑘𝑖 𝑇𝑖 ; 𝑖 ≠ 𝑘 ≠ 𝑙
𝜔̇ 𝑘 =(A𝑐𝑘 − M𝑘 B𝑐𝑘𝑘 C𝑐𝑘 )𝜔𝑘 + B𝑐𝑘𝑘 (1 − M𝑘 D𝑐𝑘𝑘 )𝑃𝑘 𝑖=1

∑
𝑛
(13) where, (B7 )𝑘𝑙 = (B2 )𝑘𝑘 − D𝑐𝑙𝑘 (B1 )𝑘𝑙 (D𝑐𝑙𝑙 )−1 ; (B1 )𝑘𝑙 = M𝑙 D𝑐𝑙𝑙 (B𝑐𝑘𝑙 − M𝑘 B𝑐𝑘𝑘 D𝑐𝑘𝑙 );
+ (B𝑐𝑘𝑖 − M𝑘 B𝑐𝑘𝑘 D𝑐𝑘𝑖 )𝑇𝑖 (B8 )𝑘𝑙 = 2M𝑙 (B𝑐𝑘𝑙 − M𝑘 B𝑐𝑘𝑘 D𝑐𝑘𝑙 ) + B𝑐𝑘𝑘 D𝑐𝑘𝑙 ((D𝑐𝑘𝑘 M𝑘 )2 − D𝑐𝑘𝑙 ); (B9 )𝑘𝑖 =
𝑖=1
B𝑐𝑘𝑘 ((D𝑐𝑘𝑘 M𝑘 )2 − D𝑐𝑘𝑘 M𝑘 M𝑖 − 1) + M𝑖 B𝑐𝑘𝑖 ; (B9𝑙 )𝑘𝑖 = M𝑙 D𝑐𝑙𝑖 (B𝑐𝑘𝑙 − M𝑘 B𝑐𝑘𝑘 D𝑐𝑘𝑙 ).
In (13), the measurements from 𝑘th PMU is used to obtain active power Using (18), 𝜔̇ 𝑘 is unaffected during FDI attack at PMU𝑘 and PMU𝑙 ,
injection at 𝑘th generator-connecting bus, 𝑃𝑘 . Therefore, during CA at simultaneously. During simultaneous FDI attack at PMU set 𝑆, the
𝑘th PMU, the 𝑘th component of WAC may maloperate. Modification of expression for 𝜔̇ 𝑘 is formulated as,
(13) is required to make the WAC as robust against attack at PMU𝑘 as ( 𝑛 )
∑𝑛 ∑
𝑛 ∑
described in (14)–(16). 𝜔̇ 𝑘 = (A1 )𝑘𝑘 𝜔𝑘 + (A2 )𝑘𝑖 𝜔𝑖 + (B1 )𝑘𝑖 𝑃𝑖 + [B7 ]𝑘𝑙 𝑇𝑘
Expression of 𝑦𝑐𝑖 is formulated by replacing 𝑘 with 𝑖 in (11). Substi- 𝑖=1 𝑖=1 𝑙=1
∑
𝑛 𝑛 (
∑ ∑𝑛 )
tuting the expression of 𝑦𝑐𝑖 ⟹ 𝑇𝑖 in (13), the equation is reformulated (19)
+ (B8 )𝑘𝑙 𝑇𝑙 + (B9 )𝑘𝑖 + (B9𝑙 )𝑘𝑖 𝑇𝑖 ;
as, 𝑙=1 𝑖=1 𝑙=1

∑
𝑛 PMU𝑘 ∈ 𝑆; PMU𝑖 ∈ 𝑈 ∖𝑆; PMU𝑙 ∈ 𝑆
𝜔̇ 𝑘 = (A1 )𝑘𝑘 𝜔𝑘 + (A2 )𝑘𝑖 𝜔𝑖 + (B1 )𝑘 𝑃𝑘
𝑖=1 where, 𝑈 = {PMU1 , PMU2 , … , PMU𝑛 }.
(14) During simultaneous FDI attack at PMU set 𝑆1 , where 𝑆1 = 𝑆∖
∑
𝑛 ∑
𝑛
+ (B1 )𝑘𝑖 𝑃𝑖 + (B2 )𝑘𝑗 𝑇𝑗 ; 𝑖 ≠ 𝑘; 𝑗 ≠ 𝑖 {PMU𝑘 }, the expression of 𝜔̇ 𝑘 is formulated as,
𝑖=1 𝑗=1
∑
𝑛 ∑
𝑛

where, (A1 )𝑘𝑘 = A𝑐𝑘 − M2𝑘 B𝑐𝑘𝑘 C𝑐𝑘 D𝑐𝑘𝑘 ; (A2 )𝑘𝑖 = M𝑖 C𝑐𝑖 (B𝑐𝑘𝑖 − M𝑘 B𝑐𝑘𝑘 D𝑐𝑘𝑖 ); 𝜔̇ 𝑘 = (A1 )𝑘𝑘 𝜔𝑘 + (A2 )𝑘𝑖 𝜔𝑖 + (B1 )𝑘𝑖 𝑃𝑖
𝑖=1 𝑖=1
(B1 )𝑘 = B𝑐𝑘𝑘 (1 − M2𝑘 (D𝑐𝑘𝑘 )2 ); (B1 )𝑘𝑖 = M𝑖 D𝑐𝑖𝑖 (B𝑐𝑘𝑖 − M𝑘 B𝑐𝑘𝑘 D𝑐𝑘𝑖 ); (B2 )𝑘𝑗 =
∑
𝑛 ∑
𝑛
(20)
M𝑗 (B𝑐𝑘𝑗 − M𝑘 B𝑐𝑘𝑘 D𝑐𝑘𝑗 ); M𝑗 = (1 + D𝑐𝑗𝑗 )−1 . + (B4 )𝑘𝑘 𝑃𝑘 + (B5 )𝑘𝑙 𝑇𝑙 + (B10 )𝑘𝑚 𝑇𝑚 ;
The expression in (14) uses all the generator bus-connected PMUs 𝑙=1 𝑚=1
to provide input at 𝑘th component of the WAC. Now, elimination of 𝑦𝑘 PMU𝑖 ∈ 𝑈 ∖𝑆; PMU𝑙 ∈ 𝑆1 ; PMU𝑚 ∈ 𝑈 ∖𝑆1
from the 𝑘th component of WAC can be compensated by other PMU ∑
where, (B10 )𝑘𝑚 = M𝑚 (B𝑐𝑘𝑚 − M𝑘 B𝑐𝑘𝑘 D𝑐𝑘𝑚 ) − 𝑛𝑝=1 M𝑝 D𝑐𝑝𝑚 (B𝑐𝑘𝑝 − M𝑘 B𝑐𝑘𝑘 D𝑐𝑘𝑝 );
data as followed. 𝑐 −1
PMU𝑝 ∈ 𝑈 ∖𝑆1 ; M𝑝 = (1 + D𝑝𝑝 )
From (11), the 𝑦𝑘 is expressed as,
∑
𝑛 6. Proposed FDM model architecture for WAC under FDI attack at
𝑦𝑘 =(D𝑐𝑘𝑘 )−1 𝑦𝑐𝑘 − (D𝑐𝑘𝑘 )−1 D𝑐𝑘𝑖 𝑦𝑐𝑖 different PMUs
𝑖=1 (15)
− C𝑐𝑘 (D𝑐𝑘𝑘 )−1 𝑥𝑐𝑘 ; 𝑖 ≠ 𝑘
𝑘th component of the centralized WAC model with the proposed
Re-expressing (15) with 𝑦𝑘 ⟹ 𝑃𝑘 , 𝑦𝑐𝑘 ⟹ 𝑇𝑘 , 𝑦𝑐𝑖 ⟹ 𝑇𝑖 , 𝑥𝑐𝑘 ⟹ FDM strategy under FDI attack at PMU𝑘 and PMU𝑙 are shown in
𝜔𝑘 , and substituting the expression in (14), Fig. 3(a) and Fig. 3(b), respectively. 𝑃𝑘 is erroneous during the attack at
PMU𝑘 . Therefore, 𝑃𝑘 is removed from the set of inputs, {𝑃1 , 𝑃2 , … , 𝑃𝑛 },
∑
𝑛 ∑
𝑛
in the WAC. In such a situation, the proposed FDM in WAC uses
𝜔̇ 𝑘 = (A3 )𝑘𝑘 𝜔𝑘 + (A2 )𝑘𝑖 𝜔𝑖 + (B1 )𝑘𝑖 𝑃𝑖 + (B2 )𝑘𝑘 𝑇𝑘
𝑖=1 𝑖=1 (16). During the FDI at PMU𝑙 , 𝑃𝑙 is eliminated from the input set.
(16)
∑
𝑛 In this situation, the proposed FDM strategy replaces (16) with (17)
+ (B3 )𝑘𝑖 𝑇𝑖 ; 𝑖 ≠ 𝑘 in centralized WAC. 𝑘th component of the centralized WAC model
𝑖=1 with the proposed FDM strategy under simultaneous FDI attack at
where, (A3 )𝑘𝑘 = A𝑐𝑘 − B𝑐𝑘𝑘 C𝑐𝑘 (D𝑐𝑘𝑘 )−1 ; (B2 )𝑘𝑘 = B𝑐𝑘𝑘 ((D𝑐𝑘𝑘 )−1 + M𝑘 −2D𝑐𝑘𝑘 M2𝑘 ); PMU𝑘 and PMU𝑙 is depicted in Fig. 3(c). In this situation, 𝑃𝑘 and 𝑃𝑙
(B3 )𝑘𝑖 = B𝑐𝑘𝑘 D𝑐𝑘𝑖 (D𝑐𝑘𝑘 M2𝑘 − (D𝑐𝑘𝑘 )−1 ) − M𝑖 (B𝑐𝑘𝑖 − M𝑘 B𝑐𝑘𝑘 D𝑐𝑘𝑖 ). are pruned from input set of the WAC. During this attack, the WAC
In (16), 𝜔̇ 𝑘 is expressed with the elimination of 𝑃𝑘 . Therefore, using with the proposed FDM strategy operates using (18). Fig. 4(a) and
(16), the 𝜔𝑘 is unaffected during FDI attack at PMU𝑘 . Similarly, with Fig. 4(b) show 𝑘th component of the centralized WAC model with the
proposed FDM strategy under simultaneous attack at the set of PMUs
elimination of 𝑃𝑙 (= 𝑃𝑖,𝑖=𝑙≠𝑘 ) from (14), expression of 𝜔̇ 𝑘 is reformulated
𝑆 and 𝑆1 , respectively. In such situations, the WAC operates with the
as,
proposed FDM strategy using (19) and (20), respectively. Eq. (16)–(20)
∑
𝑛 ∑
𝑛
are independent of data sent from the FDI attacked PMUs. Thereby, the
𝜔̇ 𝑘 = (A1 )𝑘𝑘 𝜔𝑘 + (A2 )𝑘𝑖 𝜔𝑖 + (B1 )𝑘𝑗 𝑃𝑗 + (B1 )𝑘 𝑃𝑘
𝑖=1 𝑗=1
WAC with the proposed FDM remains immune from the CAs.
(17) The basic steps involved in the proposed FDM method are;
∑
𝑛
+ (B5 )𝑘𝑙 𝑇𝑙 + (B6 )𝑘𝑚 𝑇𝑚 ; 𝑖 ≠ 𝑘 ≠ 𝑙; 𝑗 ≠ 𝑙 ≠ 𝑘; 𝑚 ≠ 𝑙
1. Detect the CAs at the PMUs, with the FDD filter at the centralized
𝑚=1
WAC.
where, (B1 )𝑘𝑗 = M𝑗 D𝑐𝑗𝑗 (B𝑐𝑘𝑗 − M𝑘 B𝑐𝑘𝑘 D𝑐𝑘𝑗 ); (B5 )𝑘𝑙 = 2M𝑙 (B𝑐𝑘𝑙 − M𝑘 B𝑐𝑘𝑘 D𝑐𝑘𝑙 ); 2. The substations of the attacked PMUs must send the CA detec-
(B6 )𝑘𝑚 = M𝑚 (B𝑐𝑘𝑚 −M𝑘 B𝑐𝑘𝑘 D𝑐𝑘𝑚 )−M𝑙 D𝑐𝑙𝑚 (B𝑐𝑘𝑙 −M𝑘 B𝑐𝑘𝑘 D𝑐𝑘𝑙 ); M𝑚 = (1+D𝑐𝑚𝑚 )−1 ; tion information to the PDC, using a binary command through
M𝑙 = (1 + D𝑐𝑙𝑙 )−1 communication channels.
Using (17), the 𝑥𝑐𝑘 is unaffected during FDI attack at PMU𝑙 . For 3. Deactivate the WAC temporarily in the PDC.
simultaneous attack at PMU𝑘 and PMU𝑙 , the statespace equation of 𝑘th 4. Eliminate the WAC inputs from the attacked PMUs/substations.
component of WAC needs to be remodified as follows. 5. Modify the state space model of WAC with the proposed FDM
By replacing 𝑘 with 𝑙 in (15) and substituting the expression in (16) method, as explained in Section 5.
with 𝑦𝑘 ⟹ 𝑃𝑘 , 𝑦𝑐𝑘 ⟹ 𝑇𝑘 , 𝑦𝑐𝑖 ⟹ 𝑇𝑖 , 𝑥𝑐𝑘 ⟹ 𝜔𝑘 , the 𝜔̇ 𝑘 is 6. Reactivate the modified WAC.

4
P. Mahish et al. Electric Power Systems Research 236 (2024) 110912

Fig. 4. 𝑘th component of the centralized WAC model with the proposed FDM strategy
under simultaneous FDI attack at the set of PMUs (a) 𝑆 (b) 𝑆1 .

Fig. 3. 𝑘th component of the centralized WAC model with the proposed FDM strategy
under FDI attack at (a) PMU𝑘 (b) PMU𝑙 (c) PMU𝑘 and PMU𝑙 .

7. Verify CA with the FDD filter. If CA still exists, then continue with
the modified WAC. Otherwise, switch to conventional model of
WAC, as in (14).

7. Simulation case studies

The single line diagram of New England 39 bus test system is shown
in Fig. 5. Each of the 10 generators avails AVR and 𝛥𝜔-type PSS [34].
In this study, PMUs form the attack surface which are vulnerable to
CA vectors. An optimal placement technique is used to minimize the
number of PMU signals in WAC [35]. Total 13 PMUs are identified
at the optimal buses; bus 2, bus 6, bus 9, bus 10, bus 12, bus 14, Fig. 5. New England 39 bus test system.
bus 17, bus 19, bus 20, bus 22, bus 23, bus 25, and bus 29, for WAC
operation [35,36]. The PMUs at the optimal buses form different attack
surfaces, and are the least requirement to maintain full observability of
large error, which widens usability and impact of the proposed FDM
the 39 bus power system. By identifying the optimal buses/substations,
the attack surface is minimized. However, during CA at even one PMU approach. The FDM eliminates the requirement of estimated data,
within these optimal substations, observability of the power system and discards WAC inputs from the attacked PMUs/substations. Each
fails resulting erroneous estimation of PMU measured data. In such PSS operates with their local and WAC signal. The state space model
situation, the estimation-based approaches cannot be reliable due to parameters of WAC is available in the appendix.

5
P. Mahish et al. Electric Power Systems Research 236 (2024) 110912

Fig. 6. CA vector at PMU25 . Fig. 8. Rotor speed deviation of G8 at the steady power system for CA at PMU25 .

Fig. 7. Active power oscillation of G8 at the steady power system for CA at PMU25 .
Fig. 9. Active power oscillation of G8 at the disturbed power system for CA at PMU25 .

7.1. Single PMU attack scenario

In this scenario, a CA vector is introduced at PMU25 as shown in

Fig. 6. The attack is initialized at 2 s and continues till 4 s. The CA
vector is manipulating the active power flow measurements for each of
the lines connected at bus 25, in such a way that the net power injection
at the bus is unchanged at every instant during the attack. Hence, the
CA is not affecting load flow calculations of the test system, which
allows bad data detection (BDD) filter to bypass the attack vector. The
FDD filter takes 0.1 s to detect the CA with the mechanism explained in
Section 3. The proposed FDM technique takes additional 0.7 s for the
Fig. 10. Rotor speed deviation of G8 at the disturbed power system for CA at PMU25 .
computation of parameters of the modified WAC, in Intel(R) Core(TM)
i5 3.3 GHz processor. Therefore, the modified WAC is activated at
2.8 s. During 2.1 s to 2.8 s, the proposed FDM makes operation of
PSSs with local control signals only. Two different states of the power oscillation, proving the effectiveness of the proposed method. Fig. 10
system are considered to demonstrate performance of the proposed depicts rotor speed deviation of G8 at the steady power system for CA
FDM technique; (i) steady state (ii) disturbed power system. at PMU25 . The WAC performance is poor without mitigation techniques
(1) Results for the proposed FDM to mitigate CA at steady power system: of the CA. The WAC with the proposed FDM technique settles the rotor
The active power oscillation of G8 at the steady power system for CA at oscillation effectively within 3 s of detecting the CA. As compared to
PMU25 is shown in Fig. 7. G8 provides steady power generation during that, the DSE method takes more time to damp out the oscillation.
no CA. Without any mitigation technique of the CA, the WAC fails to
settle the power oscillation of G8 within 10 s. With the DSE method, 7.2. Multiple PMU attack scenario
WAC at the PSS of G8 takes more than 5 s after detecting the CA, to
settle the power generation. With the proposed FDM, the WAC at the In this scenario, PMU25 , PMU29 and PMU20 are attacked with
PSS damps out the oscillation within 2.5 s after detecting the CA. Fig. 8 multiple step CA, during 2 to 4 s. The CA vectors introduced at PMU25
depicts the rotor speed deviation of G8 at the steady power system for and PMU20 are same and shown in Fig. 6. The CA vector at PMU29
CA at PMU25 . WAC performance is poor for the CA without mitigation is depicted in Fig. 11, which is used to manipulate the measurements
techniques. The WAC settles the rotor oscillation with DSE method after of active power flow in the lines connected at bus 29, in a similar
5 s of detecting the CA. With the FDM technique, the WAC takes less way as explained for single PMU attack scenario. The attack vectors
than 3 s to damp out the rotor oscillation which shows effectiveness of are detected at 2.1 s. Followed by the detection, the proposed FDM
the proposed method. takes additional 0.7 s to modify the WAC model. Therefore, the WAC
(2) Results for the proposed FDM to mitigate CA at disturbed power is reactivated with the proposed FDM at 2.8 s. During 2.1 s to 2.8 s,
system In this study, an outage event of line 25–26 is simulated at 1 s, the proposed FDM makes operation of PSSs with local control signals
resulting the power system is in disturbed condition. Fig. 9 shows the only.
active power oscillation of G8 at the disturbed power system for CA (1) Results for the proposed FDM to mitigate CA at the steady power
at PMU25 . Without the CA mitigation techniques, the WAC takes more system: The active power oscillation of G8 at the steady power system
than 10 s to settle the power oscillation. WAC with the DSE method for CA at PMU25 , PMU29 and PMU20 is shown in Fig. 12(a). For CA
takes more than 7 s of detecting the CA, to damp out the oscillation. without mitigation techniques, the WAC is not able to damp out the
As compared to that, WAC with the FDM takes less time to damp out the oscillation within 10 s. Using the proposed FDM technique, the WAC

6
P. Mahish et al. Electric Power Systems Research 236 (2024) 110912

Fig. 11. CA vector at PMU29 .

Fig. 13. Rotor speed deviation of (a) G8 (b) G9 (c) G5 at the steady power system
for simultaneous CAs at PMU25 , PMU29 and PMU20 .

at 1 s to make the power system in disturbed condition. For the CAs,

the active power oscillation of G8, G9 and G5 are shown in Fig. 14(a),
Fig. 14(b) and Fig. 14(c), respectively. The performance of the WAC
at these generators is poor for the CAs without mitigation techniques.
With the proposed FDM technique for WAC, the active power oscilla-
tion of these generators mitigates within 5 s of detecting the CAs. With
DSE method, the WAC takes more than 8 s of detecting the CAs to damp
out the oscillations. Fig. 15(a), Fig. 15(b) and Fig. 15(c) depict the rotor
speed deviation of G8, G9 and G5 at the disturbed power system for
Fig. 12. Active power oscillation of (a) G8 (b) G9 (c) G5 at the steady power system
for simultaneous CAs at PMU25 , PMU29 and PMU20 .
the CAs, respectively. With using the DSE method in WAC, the settling
time of the rotor oscillation at these generators are more than 10 s. As
compared to that, the settling time reduces to less than 6 s with using
FDM technique in WAC, which shows the effectiveness of the proposed
settles the oscillation within 3 s of detecting the CAs. As compared
method.
to that, the WAC with DSE method takes more time to damp out the
oscillation. Fig. 12(b) and Fig. 12(c) depict the active power oscillation
at G9 and G5 for the CAs, respectively. The WAC with the proposed 7.3. Communication delayed scenario at the input channels of the PDC
FDM settles the power oscillation at both the generators within 4 s of
detecting the CAs. Whereas, with DSE method, the WAC takes more
The real-time performance of the proposed method is compared
time to mitigate the oscillations. Fig. 13(a), Fig. 13(b) and Fig. 13(c)
with the DSE method in OPAL-RT platform OP4510 [37]. Under the
show the rotor speed deviation of G8, G9 and G5 for the CAs, respec-
false data attack situation (Fig. 6), variable communication delays in
tively. With the proposed FDM technique, the rotor oscillation of G8
and G9 settle within 4 s of detecting the CAs in WAC. In this situation, the range of 40–100 ms are introduced at the different input channels
the rotor oscillation of G5 mitigates within 7 s. As compared to that, of PDC using the open source software Open-PDC manager. Data from
with DSE technique, the WAC takes more time to damp out the rotor this software are used to test the real-time performance of the proposed
oscillation of all these generators. method. The result for such a case is shown in Fig. 16. Damping
(2) Results for the proposed FDM to mitigate CAs at the disturbed power with the proposed FDM is more than the DSE method in the delayed
system: In this study, an outage event of the line 25–26 is simulated situation.

7
P. Mahish et al. Electric Power Systems Research 236 (2024) 110912

Fig. 14. Active power oscillation of (a) G8 (b) G9 (c) G5 at the disturbed power system Fig. 15. Rotor speed deviation of (a) G8 (b) G9 (c) G5 at the disturbed power system
for simultaneous CAs at PMU25 , PMU29 and PMU20 . for simultaneous CAs at PMU25 , PMU29 and PMU20 .

8. Conclusion

This work proposes the FDM technique to secure the WAC operation
during CA at PMUs in the power system. Following detection of the
CAs, the proposed method modifies the WAC model. In this process,
the respective weights of the data coming from the malicious PMUs
are reduced to zero. Thus, the effect of CAs at the PMUs are eliminated
in the WAC. The modified WAC model ensures power system stability
with the measured data at healthy PMUs. With increasing number of
attacked PMUs, the estimated data at the attacked PMUs are more
Fig. 16. Rotor speed deviation of G8 at the disturbed power system for CA at PMU25
erroneous. In such situation, the proposed method modifies the WAC
with variable communication delays of 40–100 ms at the input channels of PDC.
model using healthy PMU data. According to IEEE C37.118 standard,
total vector error (TVE) of the healthy PMUs must be less than or equal
to 1% [38]. Thereby, accuracy of FDM is better than the estimation-
based approaches. The proposed FDM is tested for New England 39 PMUs. The ANN usually demands large amount of data to train the
WAC model which can be availed from healthy PMUs.
bus system. Simulation results show that the performance of WAC
during CA is better with the proposed method as compared to the DSE
CRediT authorship contribution statement
approach.
The proposed method introduces a downtime of WAC to modify Priyatosh Mahish: Writing – original draft, Validation, Methodol-
the state space model, on detection and identification of the attacked ogy, Formal analysis. Ashok Kumar Pradhan: Supervision, Conceptu-
PMUs. This time duration could be significant for the case of weak grid alization. Sukumar Mishra: Conceptualization.
or/and low inertia in the power systems, which is a limitation of the
proposed approach. To reduce the WAC downtime, online formulation Declaration of competing interest
of the proposed FDM models requires high computational processors
which is expected to be available at the PDC. In this regard, artificial The authors declare that they have no known competing finan-
neural (ANN) network based techniques can be used in future to model cial interests or personal relationships that could have appeared to
such robust FDM strategy in WAC for different combination of attacked influence the work reported in this paper.

8
P. Mahish et al. Electric Power Systems Research 236 (2024) 110912

Data availability [12] C. Cameron, C. Patsios, P.C. Taylor, Z. Pourmirza, Using self-organizing archi-
tectures to mitigate the impacts of denial-of-service attacks on voltage control
schemes, IEEE Trans. Smart Grid 10 (3) (2019) 3010–3019.
No data was used for the research described in the article.
[13] L. Liu, M. Esmalifalak, Q. Ding, V.A. Emesih, Z. Han, Detecting false data
injection attacks on power grid by sparse optimization, IEEE Trans. Smart Grid
Appendix 5 (2) (2014) 612–621.
[14] G. Liang, S.R. Weller, J. Zhao, F. Luo, Z.Y. Dong, A framework for cyber-topology
attacks: Line-switching and new attack scenarios, IEEE Trans. Smart Grid 10 (2)
A𝑐 = 𝑑𝑖𝑎𝑔[ − 9.6 − 103.21 − 4.71 − 10.33 − 3.26 − 8.75 (2019) 1704–1712.
[15] R. Yadav, A.K. Pradhan, PCA-LSTM learning networks with Markov chain models
− 56.19 − 11.49 − 36.69 − 1.87]
for online classification of cyber-induced outages in power system, IEEE Syst. J.
B = 𝑑𝑖𝑎𝑔[0.47 0.41 0.73 0.9 0.63 0.55 1.31 0.64 0.91 15 (3) (2021) 3948–3957.
[16] J. Wang, D. Wang, L. Su, J.H. Park, Dynamic event-triggered H∞ load frequency
1.01] control for multi-area power systems subject to hybrid cyber attacks, IEEE Trans.
Syst. Man, Cybern. Syst. 52 (12) (2022) 7787–7798.
⎡ 0 1.13 0.09 0.23 0.13 0.45 0.69 0.16 0.14 0.19 ⎤
[17] L. Che, X. Liu, Z. Li, Y. Wen, False data injection attacks induced sequential
⎢ 0.25 0 0.12 0.2 0.62 0.18 0.12 0.49 0.08 0.25 ⎥
⎢ ⎥ outages in power systems, IEEE Trans. Power Syst. 34 (2) (2019) 1513–1523.
⎢ 0.68 0.72 0 0.14 0.19 0.29 1 0.1 0.2 0.12 ⎥ [18] M. Mohammadpourfard, A. Sami, Y. Weng, Identification of false data injec-
⎢ 0.43 0.2 0.22 0 1.07 0.09 0.21 0.25 0.14 0.12 ⎥ tion attacks with considering the impact of wind generation and topology
⎢ ⎥ reconfigurations, IEEE Trans. Sustain. Energy 9 (3) (2018) 1349–1364.
0.26 1.53 0.05 0.31 0 0.62 0.38 0.22 0.08 0.26 ⎥
𝐵𝑐 = ⎢ [19] Q. Yang, et al., On false data-injection attacks against power system state
⎢ 0.12 0.55 0.06 0.27 0.31 0 0.06 0.66 0.04 0.33 ⎥ estimation modeling and countermeasures, IEEE Trans. Parallel Distrib. Syst. 25
⎢0.88 − 0.31 0.17 0.06 0.25 0.12 0 0.04 0.26 0.05 ⎥⎥
⎢ (3) (2014) 717–729.
⎢0.55 0.07 0.27 0.03 1.38 0.03 0.27 0 0.18 0.04⎥ [20] S. Bi, Y.J. Zhang, Graphical methods for defense against false-data injection
⎢ 0.03 0.65 0.02 0.32 0.08 0.28 0.01 0.77 0 0.39 ⎥ attacks on power system state estimation, IEEE Trans. Smart Grid 5 (3) (2014)
⎢ ⎥ 1216–1227.
⎣ 0.01 0.3 0.01 1.5 0.02 0.29 0.01 0.2 0.05 0 ⎦
[21] R. Deng, G. Xiao, R. Lu, Defending against false data injection attacks on power
C𝑐 = 𝑑𝑖𝑎𝑔[34.52 32.57 2.28 19.17 16.96 1.08 20.18 27.14 system state estimation, IEEE Trans. Ind. Inform. 13 (1) (2017) 198–207.
[22] S. Pal, B. Sikdar, J.H. Chow, Classification and detection of PMU data manipu-
15.67 40.13] lation attacks using transmission line parameters, IEEE Trans. Smart Grid 9 (5)
(2018) 1216–1227.
D = 𝑑𝑖𝑎𝑔[1.4 2.4 0.4 2.7 0.5 0.8 1.1 1.7 2.5 0.4] [23] A.F. Taha, J. Qi, J. Wang, J.H. Panchal, Risk mitigation for dynamic state
estimation against cyber attacks and unknown inputs, IEEE Trans. Smart Grid 9
⎡ 0 0.1 0.01 0.12 0.15 0.14 0.16 0.12 0.01 0.16 ⎤ (2) (2018) 886–899.
⎢ 0.03 0 0.01 0.02 0.03 0.03 0.05 0.02 0 0.03 ⎥ [24] Z. Zhang, R. Deng, D. k. Y. Yau, P. Cheng, M.Y. Chow, Security enhancement
⎢ ⎥ of power system state estimation with an effective and low-cost moving target
⎢ 0.02 0.01 0 0.02 0.05 0.05 0.03 0.03 0 0.07 ⎥ defense, IEEE Trans. Syst. Man, Cybern. Syst. 53 (5) (2023) 3066–3081.
⎢ 0.02 0.03 0.03 0 0.32 0.24 0.05 0.07 0 2.02 ⎥ [25] G. Chaojun, P. Jirutitijaroen, M. Motani, Detecting false data injection attacks
⎢ ⎥ in AC state estimation, IEEE Trans. Smart Grid 6 (5) (2015) 2476–2483.
0 0 0 0.01 0 0.02 0.01 0.01 0 0.03 ⎥
𝐷𝑐 = ⎢ [26] S.K. Singh, K. Khanna, R. Bose, B.K. Panigrahi, R. Lu, Joint-transformation-based
⎢ 0.01 0.01 0.01 0.03 0.03 0 0.02 0.03 0 0.1 ⎥
⎢ 0.04 0.03 0.03 0.04 0.05 0.05 0 0.19 0.01 0.23 ⎥ detection of false data injection attacks in smart grid, IEEE Trans. Ind. Inform.
⎢ ⎥ 14 (1) (2018) 89–97.
⎢ 0.01 0 0.03 0.01 0.03 0.03 0.26 0 0 0.24⎥ [27] C. Liu, M. Zhou, J. Wu, C. Long, D. Kundur, Financially motivated FDI on SCED
⎢ 0 0 0.01 0.01 0.02 0.02 0.07 0.03 0 0.05⎥ in real-time electricity markets: Attacks and mitigation, IEEE Trans. Smart Grid
⎢ ⎥
⎣ 0.01 0 0.04 0.01 0.02 0.02 0.05 0.07 0.83 0 ⎦ 10 (2) (2019) 1949–1959.
[28] H. Shayan, T. Amraee, Network constrained unit commitment under cyber
attacks driven overloads, IEEE Trans. Smart Grid 10 (6) (2019) 6449–6460.
[29] S. Sridhar, M. Govindarasu, Model-based attack detection and mitigation for
References automatic generation control, IEEE Trans. Smart Grid 5 (2) (2014) 580–591.
[30] R. Tan, et al., Modeling and mitigating impact of false data injection attacks
[1] G. Liang, J. Zhao, F. Luo, S. Weller, Z.Y. Dong, A review of false data injection on automatic generation control, IEEE Trans. Inf. Forensics Secur. 12 (7) (2017)
attacks against modern power systems, IEEE Trans. Smart Grid 8 (4) (2017) 1609–1624.
1630–1638. [31] Y. Zhao, et al., Resilient adaptive wide-area damping control to mitigate false
[2] C. Peng, H. Sun, M. Yang, Y.L. Wang, A survey on security communication and data injection attacks, IEEE Syst. J. 15 (4) (2021) 4831–4842.
control for smart grids under malicious cyber attacks, IEEE Trans. Syst. Man, [32] P. Mahish, A.K. Pradhan, A.K. Sinha, Wide area predictive control of power
Cybern. Syst. 49 (8) (2019) 1554–1569. system considering communication delay and data drops, IEEE Trans. Ind.
[3] P. Ramanan, D. Li, N. Gebraeel, Blockchain-based decentralized replay attack Inform. 15 (6) (2019) 3243–3253.
detection for large-scale power systems, IEEE Trans. Syst. Man, Cybern. Syst. 52 [33] F. Pasqualetti, F. Dörfler, F. Bullo, Attack detection and identification in
(8) (2022) 4727–4739. cyber–physical systems, IEEE Trans. Autom. Control 58 (11) (2013) 2715–2729.
[4] K. Poulsen, Slammer worm crashed ohio nuke plant network, 2003, [Online]. [34] I. Hiskens, IEEE PES task force on benchmark systems for stability controls on
Available: https://www.securityfocus.com/news/6767. IEEE 10-generator, 39-bus system, 2013, [Online]. Available: http://www.sel.
[5] M. Holloway, Stuxnet worm attack on Iranian nuclear facilities, 2015, [Online]. eesc.usp.br/ieee/.
[35] B.K.S. Roy, A.K. Sinha, A.K. Pradhan, An optimal PMU placement technique
Available: http://large.stanford.edu/courses/2015/ph241/holloway1/.
for power system observability, Int. J. Elect. Power Energy Syst. 42 (1) (2012)
[6] G. Liang, S.R. Weller, J. Zhao, F. Luo, Z.Y. Dong, The 2015 Ukraine blackout:
71–77.
Implications for false data injection attacks, IEEE Trans. Power Syst. 32 (4)
[36] R. Sodhi, S.C. Srivastava, S.N. Singh, Optimal PMU placement to ensure system
(2017) 3317–3318.
observability under contingencies, in: Proc. IEEE Power Eng. Soc. General
[7] CBC News, Electricity theft by BC grow-ops costs $100 a year, 2010, [On-
Meeting, 2009, pp. 1–6.
line]. Available: https://www.cbc.ca/news/canada/britishcolumbia/electricity-
[37] OP4510 RT-LAB-rcp/HIL systems user guide, in: OPAL-RT Technol, Montreal,
theft-by-bc-grow-ops-costs-100m-ayear-1.969837.
QC, Canada, 2016.
[8] B. Krebs, FBI: Smart meter hacks likely to spread, 2012, [Online]. Available:
[38] IEEE standard for synchrophasor data transfer for power systems, in: IEEE Std
https://krebsonsecurity.com/2012/04/fbi-smart-meter-hacks-likely-to-spread/.
C37.118.2-2011 (Revision of IEEE Std C37.118-2005), 2011, pp. 1–53.
[9] CIP-011-3 - cyber security - information protection, 2021, [Online]. Available:
https://www.nerc.com/pa/Stand/Project201902BCSIAccessManagement/CIP-
011-3_Clean.pdf.
[10] CIP-005-7 - cyber security - electronic security perimeter(s), 2022, [Online]. Priyatosh Mahish is presently working as an assistant professor in the School of
Available: https://www.nerc.com/pa/Stand/Reliability%20Standards/CIP-005-7. Computing and Electrical Engineering at Indian Institute of Technology, Mandi, India.
pdf. He has received M.Tech in Electrical Power Systems from Birla Institute of Technology
[11] Standards, compliance, and enforcement bulletin, 2022, [Online]. Mesra in 2013. Dr. Mahish has received his Ph.D. degree in Electrical Power System
Available: https://www.nerc.com/pa/comp/news/Documents/2022_07_18_ control from the Indian Institute of Technology Kharagpur in 2021. He has also received
StandardsCompliance_Bulletin.pdf. the prestigious POSOCO power system award (2022) for his doctoral research work.

9
P. Mahish et al. Electric Power Systems Research 236 (2024) 110912

His research interests include smart grid operation, wide area power system monitoring Industries Platinum Jubilee Award for Application Oriented Innovation in Physical
and control, and grid-integration of renewable energy resources. Sciences (2019). He has been granted fellowships from academies like NASI (India),
INAE (India), and professional societies like IET (U.K.), IETE (India), IE (India). He
has also been recognized as the INAE Industry Academic Distinguished Professor. His
Ashok Kumar Pradhan received the Ph.D. degree in electrical engineering from Sam- research interests lie in the field of Power Systems, Power Quality Studies, Renewable
balpur University, Sambalpur, India, in 2001. Dr. Pradhan has been with the Depart- Energy and Smart Grid. He is currently acting as the ABB Chair professor and has
ment of Electrical Engineering, Indian Institute of Technology Kharagpur, Kharagpur, previously delegated as the NTPC, INAE, and Power Grid Chair professor. He has also
India, since 2002, where he is currently a Professor. He was with the Department of served as an Independent Director of the Cross Border Power Transmission Company
Electrical Engineering, VSS University of Technology, Burla, India from 1992 to 2002. Ltd. and the River Engineering Pvt. Ltd. And has carried out many important industrial
His research interests include power system relaying and monitoring and control. Prof. consultations with TATA Power, Microtek and others. He is the founder of SILOV
Pradhan is a Fellow of Indian National Academy of Engineering India. SOLUTIONS PRIVATE LIMITED, a company that specifically deals in products related to
renewable energy sources utilizable at household scale as well as at commercial setups.
From March 2020, he has also been functioning as the Associate Dean Research and
Sukumar Mishra received his M.Tech and Ph.D. in Electrical Engineering from National
Development of IIT Delhi. Prof. Mishra has been working in close association with IEEE
Institute of Technology, Rourkela in 1992 and 2000 respectively. Presently, Dr. Mishra
Delhi Section Executive Committee for the past few years. He has served as an Editor
is a Professor at the Indian Institute of Technology Delhi and has been its part
for the IEEE Transactions on Smart Grid, IEEE Transactions on Sustainable Energy and
for the past 17 years. He has won many accolades such as INSA Medal for Young
was an Area Editor for the IET Generation, Transmission Distribution journal.
Scientist (2002), INAE Young Engineer Award (2009), INAE Silver Jubilee Young
Engineer Award (2012), Samanta Chandra Shekhar Award (2016), IETE Bimal Bose
award (2019), National Mission Innovation Champion award (2019) and NASI-Reliance

10