Content uploaded by Prabhat Kumar
Author content
All content in this area was uploaded by Prabhat Kumar on Aug 31, 2023
Content may be subject to copyright.
Solar Energy 263 (2023) 111921
Available online 25 August 2023
0038-092X/© 2023 The Authors. Published by Elsevier Ltd on behalf of International Solar Energy Society. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).
Contents lists available at ScienceDirect
Solar Energy
journal homepage: www.elsevier.com/locate/solener
Digital twin-driven SDN for smart grid: A deep learning integrated
blockchain for cybersecurity
Prabhat Kumar a,∗, Randhir Kumar b, Ahamed Aljuhani c, Danish Javeed d, Alireza Jolfaei e,
A. K. M. Najmul Islam a
aDepartment of Software Engineering, LUT School of Engineering Science, LUT University, 53850 Lappeenranta, Finland
bDepartment of Computer Science and Engineering, SRM University AP, AP 522240, India
cDepartment of Information Technology, University of Tabuk, Tabuk 71491, Saudi Arabia
dSoftware College, Northeastern University, Shenyang 110169, China
eCollege of Science and Engineering, Flinders University, Adelaide, Australia
ARTICLE INFO
Keywords:
Blockchain
Deep learning
Digital twin
Internet of things
Smart grid
Software-defined networking
ABSTRACT
Internet of Things (IoT)-enabled Smart Grid (SG) network is envisioned as the next-generation network for
intelligent and efficient electric power transmission. In SG environment, the Smart Meters (SMs) mostly
exchange services and data from Service Providers (SPs) via insecure public channel. This makes the entire
SG ecosystem vulnerable to various security threats. Motivated from the aforementioned challenges, we
incorporate Digital Twin (DT) technology, Software-Defined Networking (SDN), Deep Learning (DL) and
blockchain into the design of a novel SG network. Specifically, a secure communication channel is first designed
using an authentication method based on blockchain technology that has the ability to withstand a number of
well-known assaults. Second, a new DL architecture that includes a self-attention mechanism, a Bidirectional-
Gated Recurrent Unit (Bi-GRU) model, fully connected layers, and a softmax classifier is designed to enhance
the attack detection process in SG environments. To deliver low latency and real-time services, the SDN is next
employed as the network’s backbone to send requests from SMs to a global SDN controller. DT technology is
finally integrated into the SDN control plane, which stores the operating states and behavior models of SMs
and communicates with SMs. The efficiency of the proposed framework is demonstrated by the blockchain
implementation used in the SG network to assess computing time for the various numbers of transactions per
block. Finally, the numerical results based on the N-BaIoT dataset shows better intrusion detection.
1. Introduction
The use of Internet of Things (IoT) has grown so rapidly due
to many advances in technology. This use of IoT devices has been
involved in a diverse range of critical sectors including healthcare,
agriculture, and energy sector [1,2]. For example, the deployment of
IoT technology in energy grid which enables to gather, share, and
analyze real-time data about energy consumption such as home and
street lights through different wireless devices such as sensors, gate-
ways, and routers. The integration of IoT technology into electric
grids has led to the IoT-enabled smart grid. When compared to tra-
ditional grids, the IoT-enabled smart grid can address several issues
such as power outages, solar flares, and security concerns [3], [4].
Additionally, IoT-enabled smart grid adopts alternative solutions such
as solar-based smart grids control in the event of widespread black-
outs or energy shortages. The IoT-enabled smart grid has become an
∗Corresponding author.
E-mail addresses: prabhat.kumar@lut.fi (P. Kumar), randhir.honeywell@ieee.org (R. Kumar), a_aljuhani@ut.edu.sa (A. Aljuhani), 2027016@stu.neu.edu.cn
(D. Javeed), alireza.jolfaei@flinders.edu.au (A. Jolfaei), najmul.islam@lut.fi (A.K.M.N. Islam).
essential part in the development of energy systems in modern cities,
with the goal of improving the safety, efficiency, and sustainability
of energy management [5]. The IoT-enabled smart grid relies on dif-
ferent telecommunication networks and smart emerging technologies
that cooperate together to improve the quality of energy services for
consumers. Such an emerging technology communicates with sensors,
actuators, machines, gateways, and heterogeneous networks that sense,
exchange, and process data via different communication technologies
(e.g., Bluetooth, ZigBee, WiMax, and 5G/LTE).
Although IoT-enabled smart grid has introduced many great fea-
tures for energy consumers, it carries with itself several security and
privacy issues [6,7]. As the nature of IoT-enabled smart grid, which
includes heterogeneous and homogeneous smart devices, networks, and
applications, sensing data in such environments is transmitted over an
https://doi.org/10.1016/j.solener.2023.111921
Received 24 January 2023; Received in revised form 5 June 2023; Accepted 28 July 2023
Solar Energy 263 (2023) 111921
2
P. Kumar et al.
insecure communication channel, exposing it to a number of privacy
and security issues [8]. For example, Man-in-the-Middle (MiTM) attacks
compromise the integrity and confidentiality of data exchanged among
legitimate entities in such an environment [9]. The availability of smart
grid is also critical in the preservation of security functionalities from
cyber threats. A well-known cyber-attack that threatens smart grid ser-
vice availability is the DDoS attack [10]. As the nature of IoT-enabled
smart grid requires all the smart things verifying their authenticity
and ensuring that smart devices can be trusted to communicate and
exchange information among such entities, the authentication of sens-
ing devices is crucial to avoiding cyberattacks in such a connected
environment [11].
Digital Twin (DT) has recently gained attention in a variety of fields
(e.g., healthcare and energy) [12,13]. A DT is a simulated version
of a real-world system. Such a technology assists in the conduct of
experiments, the testing of hypotheses, and the prediction of smart
grid system’s behavior. The DT has emerged as a promising solution
for IoT-enabled smart grid challenges such as smart grid management,
operations, and cybersecurity concerns. As IoT-enabled smart grids
become targets for various cyberattack threats, security measures such
as firewalls, IDS and IPS become increasingly important in protecting
such a connected environment. However, deploying and testing secu-
rity functions of running systems in smart grids is critical and time
consuming. The DT has the potential to improve security measures in
the IoT-enabled smart grid by maintaining and testing fully security
functional digital twins in an isolated environment to provide reliable,
resilient, and sustainable of IoT-enabled smart grid systems. Several
existing works on the use of DT as an enabler for improved security
have been proposed in the literature [14–19]. However, little work has
been done toward DT for IoT-enabled smart grid cybersecurity.
Recently, blockchain technology has been applied in a diverse range
of critical domains and integrated with different emerging technolo-
gies to provide secure and trustworthy systems. Along with other
smart technologies, blockchain technology has gained much attention
in the energy domain for privacy-preserving and secure data sharing
among different smart things in such a connected environment [20].
Blockchain is a peer to peer (P2P) network where peers can communi-
cate and do transactions without the need for a centralized authority.
Blockchain transactions are immutable and traceable, therefor any
attempt to change or manipulate data will be detected [21]. In addition,
the data integrity and confidentiality are ensured through various
cryptographic mechanisms such as symmetric/asymmetric encryption
methods and hash functions, which protect IoT-enabled smart grid data
from unauthorized access [22]. Blockchain can greatly enhance the
privacy and security of data sharing in IoT-enabled smart grid while
also improving interoperability and reliability in such connected envi-
ronments. As the communication among smart objects in IoT-enabled
smart grid environments is done through the public and insecure
channels [23], an adversary may exploit vulnerabilities and violate data
privacy among the smart objects in such a connected environment.
To overcome such serious threats, blockchain-based authentication
and key agreements ensure data sharing security and privacy in IoT-
enabled smart grid networks. Blockchain-based authentication allows
connected devices to authenticate each other with their own secret
credentials, while key agreements are involved to establish secret keys
among authenticated devices. Various existing blockchain-based au-
thentication and key agreement approaches have been proposed in
the literature [24–26]. However, less work has been done toward
blockchain for DT in IoT-enabled smart grid.
With the growing development of IoT-enabled smart grid, Artificial
Intelligence (AI) is becoming increasingly important in transforming
the traditional energy systems into cost-effective, autonomous, and in-
telligent systems [27]. The IoT-enabled smart grid can benefits from AI
in many different aspects, including monitoring and controlling grids,
fault detection, safety and security. Specifically, machine learning and
deep learning have sparked interest in protecting such a connected
environment from various types of cyberattacks [27]. Increasing the
security and privacy of sharing data among connected entities from
insider and outsider attacks is critical for providing sustainable and
secure smart grid systems. The enormous amount of heterogeneous,
homogeneous, and ambiguous data generated by IoT-enabled smart
grid networks necessitates a reliable, cost-effective, and robust de-
tection scheme to combat cyberattacks in such a connected network.
IDS becomes an efficient security tool for dealing with a wide range
of cyberattacks. Deep learning (DL) techniques have recently been
fully integrated with IDS and have become an integral part of other
security functions to protect intelligent systems from various types
of cyberattacks [27,28]. When compared to traditional ML, DL is a
better alternative for solving high-dimensional features and providing
more accurate models [29,30]. Although several works on IDS-based
deep learning schemes for IoT-enabled smart grid have been proposed
to ensure security and privacy-preserving of data sharing in such a
connected environment [31–33]. However, some limitations in the
existing approaches have been observed, such as low accuracy [34],
high complexity, outdated datasets, and being limited to specific types
of cyberattacks.
In addition to other emerging technologies, software defined net-
working (SDN) is a promising emerging technology with a high poten-
tial for deployment in IoT-enabled smart grid environments [35]. The
SDN architecture is designed to decouple control plane from network
hardware, allowing simplicity of network services, flexibility of net-
work management, and full network programmability [36]. The control
plane offers a full abstraction of the underlying SDN-enabled trans-
portation systems. Although the control plane of SDN-enabled smart
power grid is critical for networking management, routing, monitoring,
decision making, and so on, it is vulnerable to a single point of failure
and DDoS attacks [37,38]. Therefore, a centralized controller of SDN-
enabled smart power grid could be a target of different cyberattacks
aimed at disrupting the entire networking functionalities of a con-
nected system. Several existing works related to SDN-enabled smart
grids have been proposed, such as [39–41]; however, those works
lack scalability, have inadequate security analysis, and do not provide
a foundational integration framework with additional technologies in
IoT-enabled smart grid networks.
Although existing authentication and key agreements mechanisms
rely primarily on blockchain-based scheme for securing data sharing
in such connected networks. However, some of those solutions suffers
from consensus delay, high computational complexity, and lack of
scalability [42–44]. Additionally, blockchain suffers from data privacy
vulnerabilities that can be exploited by intruders [45]. The security
challenges of SDN controller is critical in ensuring network monitoring
and management of smart power grid systems. On the other hand, IDS
and deep learning techniques should be used to overcome such security
risks and ensure secure data sharing among sensing devices in such a
connected network. Meanwhile, many existing IDS-based schemes in
smart grid networks are designed to detect specific type of cyberattacks;
however, in practical uses and real application of such a connected
network, both internal and external IoT-enabled smart grid networks
are susceptible to a variety of malicious cyberattacks. As several ex-
isting security solutions rely on either a blockchain-based scheme or
IDS-based scheme; however, little research has been conducted on
blockchain and deep learning-enabled secure data sharing for DT in
IoT-enabled smart grid. We believe that Blockchain technology and
deep learning schemes can greatly complement each other to provide
cost-effective, secure, sustainable, and resilient IoT-enabled smart grid.
1.1. System model
1.1.1. Digital twin-driven software-defined networking for smart grid net-
works
The network model of DT-driven SDN for SG is shown in Fig. 1.
Three distinct planes with various participating entities make up this
Solar Energy 263 (2023) 111921
3
P. Kumar et al.
Fig. 1. DT-driven SDN for secure data sharing in smart grid network.
model. For instance, the data plane has a variety of intelligent sources
that are in charge of generating and transmitting data. This plane is
equipped with Smart Meters (SMs), Open Flow Switches (OFSs) and
Service Providers (SPs). The SMs are in charge of reading the amount
of electricity utilized and electricity used time. The SDN forwarding
devices, also referred to as OFSs, forward the generated or collected
data to SPs and SDN Controller (SDNCs) at the control plane using
an unsecured channel on a hop-to-hop basis. The SP, who are in
charge of organizing the system for allocating electricity and trading
energy, conducts an analysis of the data that they have collected. The
core decision-making body, the SDNC, keeps track of how SG’s global
traffic is routed. The DT is situated at the application plane and is
in charge of using the collected data to enhance the SG network’s
data analytics capabilities (such as grid states and electrical equipment
management). However, the reliability of the data collected cannot be
guaranteed, because communication occurs across insecure channels at
lower planes. Therefore, it is essential to develop a secure and efficient
access control with effective strategy across SG entities at the data
plane.
1.2. Research contribution
The prime contributions of this research are as follows:
(1) Secure DT-driven SDN under blockchain-based authentication sche-
me and DL-based IDS: The DT is incorporated in SG network and
its security is improved from two aspects. First by incorporating
mutual authentication and key agreement phase between SG
and SP and further by establishing a common session key for
secure communication. Second, DL-based IDS is incorporated
for attack detection. Finally, a SDN architecture is employed in
the proposed framework to reduce network latency and enhance
quality of service (QoS) in SG network.
(2) A novel blockchain-based authentication scheme: The blockchain-
based authentication scheme ensures a secure communication
and provides integrity for the exchanged messages between SG
and SP. A detailed method for creating and adding new block
to the blockchain using voting-based consensus algorithm is
presented.
(3) A novel DL-based architecture for intrusion detection: A DL-based
IDS is proposed by combining self-attention mechanism, Bi-
GRU model, fully connected layers and a softmax classifier. In
particular, a self-attention mechanism is used to determine the
spatial connections in each network, a Bi-GRU model is used
to capture the temporal dependency across networks, a fully
connected network made up of three fully connected layers, and
a softmax classifier is used to detect attacks.
The remainder of this work is organized as follows. Section 2presents
the proposed framework and its key elements for secure data shar-
ing in SG network. Section 3provides the performance analysis for
blockchain-based authentication scheme and DL-based IDS. Finally,
Section 4concludes the article.
Solar Energy 263 (2023) 111921
4
P. Kumar et al.
2. Proposed framework for secure data sharing
2.1. Blockchain-based authentication scheme
This phase includes blockchain based authentication of data source
i.e., smart meters 𝑆𝑀 and how this authentication is performed by the
various entities participating in the communication start from the smart
meters to SDN controllers. This phase makes the system more secure
by enabling four prominent approach namely System initialization, SM
data collection, and SM data aggregation, and Authentication of the
data source. Finally, the consensus approach is applied to create and
dissemination of the block in the network. The detailed process of each
approach is discussed below.
(𝑖)System Initialization: The initial constraints includes some known
parameters of various participating entities of the network namely
smart meters 𝑆𝑀𝑖, open flow switches 𝑂𝐹 𝑆𝑗, SDN controllers 𝑆 𝐷𝑁𝐶𝑘,
and service providers 𝑆𝑃𝑧. The initialization is performed by trusted
authority TS.
(𝑎)System Constraints: Let 𝑆𝑏represent a bilinear map 𝑆𝑏:𝐾1×𝐾2
→𝐾𝑇, a cyclic group with group order 𝑉𝑡, where 𝑉𝑡is a large prime
number and 𝛽is a generator of 𝐾1. Next, the message digest, i.e., a
one way cryptographic hash 𝑀𝑄1is computed. Further, the essential
constraints {𝑆𝑏,𝐾1,𝐾𝑇,𝑉𝑡,𝛽,𝑀𝑄1} are distributed to all network
entities.
(𝑏)𝑆𝑀𝑖,𝑂𝐹 𝑆𝑗,𝑆 𝐷𝑁𝐶𝑘, and 𝑆𝑃𝑧: smart meters 𝑆𝑀𝑖select a secret
key 𝑆𝑀 𝐾𝑖and compute a public key 𝑆 𝑁𝐾𝑖=𝛽𝑆𝑀 𝐾𝑖mod 𝑉𝑡, where 𝑖∈
{1, 𝑛}represents the number of smart meters. Next, 𝑆𝑀𝑖preserves the
𝑆𝑀 𝐾𝑖and shares 𝑆𝑁𝐾𝑖. Further, 𝑂𝐹 𝑆𝑗picks a secret key 𝑂𝐹 𝑀𝑗and
computes a public key 𝑂𝐹 𝑁𝑗=𝛽𝑂𝐹 𝑀𝑗mod 𝑉𝑡. Next, 𝑂𝐹 𝑆𝑗preserves
𝑂𝐹 𝑀𝑗and distributes 𝑂𝐹 𝑁𝑗. Further, 𝑆𝐷𝑁𝐶𝑘chooses a secret key
𝑆𝐷𝑁 𝑀𝑘and computes a public key 𝑆 𝐷𝑁 𝐾𝑘=𝛽𝑆𝐷𝑁 𝑀𝑘mod 𝑉𝑡. Next,
𝑆𝐷𝑁 𝐶𝑘preserves 𝑆 𝐷𝑁 𝑀𝑘and distributes 𝑆𝐷𝑁𝐾𝑘.
(𝑖𝑖)SM Data Collection: This phase details the SM data collection and
how those data is shared with 𝑂𝐹 𝑆𝑗after encryption and signature of
data, i.e., consumption of units. Next, 𝑂𝐹 𝑆𝑗verifies data and shares to
𝑆𝐷𝑁 𝐶𝑘and 𝑆 𝑃𝑧. Here, 𝑆𝑀𝑖picks a message 𝑀 𝑆 𝐺𝑖and encrypts the
message as 𝐸𝑀 𝑆𝐺𝑖and shares it with 𝑂𝐹 𝑆𝑗. Then, 𝑂𝐹 𝑆𝑗verifies the
signature 𝑆𝑀 𝑆𝐺𝑖and after a successful verification, it will share the
signature with 𝑆𝐷𝑁 𝐶𝑘and 𝑆 𝑃𝑧. Detailed steps are as follows:
•𝑆𝑀𝑖picks a random number 𝑆 𝑀𝑟𝑖from {0, 1, . . . , 𝑉𝑡−1} and en-
crypts the message 𝑀𝑆 𝐺𝑖as 𝐸𝑀𝑆 𝐺𝑖, where 𝐸𝑀𝑆 𝐺𝑖= (𝐸𝑀𝑆 𝐺1
𝑖,
𝐸𝑀 𝑆𝐺2
𝑖) = (𝛽𝑆𝑀 𝑟𝑖mod 𝑉𝑡,𝑆𝑀 𝑆 𝐺𝑖+𝑆𝑁 𝐾 𝑆𝑀 𝑟𝑖
𝑖mod 𝑉𝑡).
•𝑆𝑀𝑖signs 𝐸 𝑀𝑆𝐺𝑖to 𝑆 𝑀𝑆𝐺𝑖, where 𝑆 𝑀 𝑆𝐺𝑖=𝑠𝑖𝑔 𝑛𝑆𝑀 𝐾𝑖
(𝐸𝑀 𝑆𝐺2
𝑖).
•𝑆𝑀𝑖generates message digest using SHA-256, i.e., 𝑆 𝑀 𝐷𝑖=
𝑆𝑀 𝐷𝑄1(𝑆 𝑀𝑖∥𝐸𝑀 𝑆 𝐺𝑖∥𝑆𝑀 𝑆 𝐺𝑖∥TP)𝑆 𝑀 𝐾𝑖, where TP signi-
fies current timestamp. Next, parameters {𝐸𝑀 𝑆𝐺𝑖∥𝑆𝑀 𝑆𝐺𝑖∥
𝑆𝑀 𝐷𝑖∥TP} is shared with 𝑂𝐹 𝑆𝑗. Further, the same parame-
ters are disseminated to the 𝑆𝐷𝑁 𝐶𝑘and 𝑆 𝑃𝑧after a successful
verification.
•Once 𝑆𝐷𝑁 𝐶𝑘or 𝑆 𝑃𝑧receives {𝐸𝑀 𝑆 𝐺𝑖∥𝑆𝑀 𝑆 𝐺𝑖∥𝑆𝑀𝐷𝑖∥
TP}, it verifies three prominent attributes, i.e., (𝑖)Timestamp,
(𝑖𝑖)message digest SMDV(𝑆𝑀 𝐷𝑖,𝛽) = (𝑆𝑀𝐷𝑄1(𝑆𝑀𝑖∥𝐸 𝑀𝑆𝐺𝑖
∥𝑆𝑀 𝑆𝐺𝑖∥TP), 𝑆 𝑁 𝐾𝑖), and (𝑖𝑖𝑖)signature 𝛽𝐸𝑀 𝑆𝐺𝑖=𝑆 𝑁𝐾𝑠𝑖𝑔 𝑛𝑖
𝑖
(𝑠𝑖𝑔𝑛1
𝑖)𝑠𝑖𝑔𝑛2
𝑖. After a successful verification, 𝑆𝐷𝑁 𝐶𝑘or 𝑆 𝑃𝑧pre-
serves (𝐸𝑀 𝑆𝐺𝑖,𝑆𝑀 𝑆𝐺𝑖) for further communication in network.
(𝑖𝑖𝑖)SM Data Aggregation: This phase discusses about, how the en-
crypted data is collected and verified for respective 𝑆𝑀𝑖. The 𝑆 𝐷𝑁 𝐶𝑘
or 𝑆𝑃𝑧chooses the 𝐸 𝑀𝑆𝐺𝑖from the stored credential and computes.
𝑑=
𝑖∈[1,𝑛]
𝐸𝑀 𝑆𝐺2
𝑖𝑚𝑜𝑑 𝑉𝑡(1)
The 𝑆𝐷𝑁 𝐶𝑘and 𝑆 𝑃𝑧evaluates authentication code 𝐴𝑈 𝑇 𝐻𝑘=𝑆𝑀 𝐷𝑄1
(𝑑∥𝐸𝑀 𝑆𝐺1
𝑖∥𝑛∥TP)∗)𝑆 𝐷𝑁𝑀 𝐾𝑘and shares it with the 𝑆𝑀𝑖with pa-
rameter (𝑑,𝐸𝑀 𝑆𝐺1
𝑖,𝑛,TP∗,𝐴𝑈 𝑇 𝐻𝑘) for further communication. Simi-
larly, 𝑆𝑃𝑧computes the authentication code for further communication
in the network.
(𝑖𝑣)Authentication: This phase explores authentication process of mes-
sage source, i.e., 𝑆𝑀𝑖. As discussed in data collection phase, 𝑆 𝐷𝑁 𝐶𝑘
or 𝑆𝑃𝑧receives {𝐸 𝑀𝑆𝐺𝑖∥𝑆 𝑀𝑆𝐺𝑖∥𝑆 𝑀 𝐷𝑖∥TP } from the 𝑆𝑀𝑖. To
verify the message source, 𝑆𝐷𝑁 𝐶𝑘or 𝑆 𝑃𝑧verify SMDV(𝑆𝑀 𝐷𝑖,𝛽) =
(𝑆𝑀 𝐷𝑄1(𝑆 𝑀𝑖∥𝐸𝑀 𝑆 𝐺𝑖∥𝑆𝑀 𝑆 𝐺𝑖∥TP), 𝑆 𝑁 𝐾𝑖) using 𝑆𝑀𝑖public
key 𝑆𝑁 𝐾𝑖. If authentication successful, then 𝑆 𝐷𝑁𝐾𝑘share credential
to 𝑆𝑀𝑖for further communication.
(𝑣)Consensus for Block creation and verification: This phase states
about the block creation and verification by 𝑆𝐷𝑁 𝐶𝑘. The block consists
of 𝑠𝑖𝑔𝑛𝑆𝐷 𝑁𝐶𝑘,bk.weight,bk.parent,𝑆 𝐵𝑖∪𝑏𝑘,𝑇 𝐵𝑖∪𝑏𝑘.𝑝𝑎𝑟𝑒𝑛𝑡, and
TP [46]. The detailed block creation and verification is illustrated in
Algorithm 1.
2.2. Deep learning-based IDS
Deep Learning (DL) is a subset of Machine Learning (ML) that makes
extensive use of hidden layers. These methods outperform ML because
of their deep structure and inherent capacity to discover relevant
characteristics inside a dataset and produce an output. The proposed
DL-based IDS consist a self-attention mechanism, Bidirectional-Gated
Recurrent Unit (Bi-GRU) model, fully connected layers and a softmax
classifier. Specifically, a self-attention mechanism is used to determine
the spatial connections in each network, a Bi-GRU model is used to
capture the temporal dependency across networks, a fully connected
network made up of three fully connected layers, and a softmax clas-
sifier is used to detect attacks. The working for each of them are
explained below:
2.2.1. Self-attention mechanism for determining spatial connections
The Self-Attention 𝑆𝐴 module computes attention scores to provide
spatial attention weights, which are then automatically applied to each
connection in a network. This makes it possible for the suggested
model to pay greater attention to the discriminative connections in
an adaptive manner. The 𝑆𝐴 mechanism outperforms other attention
mechanisms in terms of computational efficiency by lowering reliance
on external input in theory. It determines the significance of features
for the GRU model’s input and hidden layers, forming a dual-stage self-
attention process 𝐷𝑆𝑆 𝐴𝑃 . The 𝐷𝑆𝑆 𝐴𝑃 ′𝑠key computation steps are as
follow. Eq. (2) is used to develop correlation between the features.
E𝑚
𝑡=F(W𝑡,𝑚[H1,𝑚 ,H2,𝑚,H𝑇 ,𝑀 ] + U𝑡,𝑚 X𝑘),
𝑡= 1,…, 𝑇 , 𝑚 = 1,…, 𝑀 (2)
where the hidden state is represented by the H𝑡,𝑚.Wand Urepresent
the parameters for learning in the training process. Fis the dense layer,
𝑇is the number of time steps, and 𝑀is the number of dimensions of the
hidden features accordingly. Further, the attention weight is calculated
as follows:
𝜎𝑚
𝑡=𝐸𝑥𝑝(E𝑚
𝑡)
𝑇
𝐾=1 𝐸𝑥𝑝(E𝑚
𝑘)
,where
𝑡
𝜎𝑚
𝑡= 1.(3)
The original input series X𝑝can be turned into X𝑡using the attention
mechanism as
X𝑡= (𝜎1
𝑡X𝑝
1, 𝜎2
𝑡X𝑝
2, 𝜎3
𝑡X𝑝
3,…, 𝜎𝑛
𝑡X𝑝
𝑛).(4)
Finally, the weights 𝜎𝑚
𝑡are averaged to 𝜎𝑚𝑒𝑎𝑛
𝑡to acquire the attention
weight of the 𝑡-th time step as
𝜎𝑚𝑒𝑎𝑛
𝑡=1
M
M
𝑚=1
𝜎𝑚
𝑡.(5)
Later on, the H𝑡can be converted to H𝑎𝑡𝑡
𝑡based on the hidden state’s
attention weight 𝜎𝑚𝑒𝑎𝑛
𝑡in the 𝑡-th time step.
H𝑎𝑡𝑡
𝑡=𝜎𝑚𝑒𝑎𝑛
𝑡H𝑡(6)
Solar Energy 263 (2023) 111921
5
P. Kumar et al.
Algorithm 1 Algorithm for Block Verification and Creation
1: State: 𝑆𝐷𝑁 𝐶𝑘∈𝐼 𝐷𝑚miners,
2: 𝐶𝐵𝑖=(𝑆 𝐵𝑖,𝑇 𝐵𝑖)𝑆𝐵𝑖is local blockchain of peer 𝑇 𝐵𝑖
3: 𝑏𝑘 ←Block records
4: parent ←previous node of 𝑏𝑘
5: miner ←mines and verify block 𝑏𝑘
6: numbers ←index of block
7: weight ←weight of block
8: blocktime ←timestamp between two different blocks
9: default time is 5 seconds.
10: vote, 𝑚𝑖𝑛𝑒𝑟+1
2
11: minerlimit ←among various successive block in which miner can
pick only one and sign it.
12: steps ←addition of new block in network
13: function latestsign(𝑠𝑖𝑔𝑛𝑆𝐷 𝑁𝐶𝑘,𝑍 𝑃 )𝑖
14: 𝛾←miner limit
15: resultflag = false
16: for 𝑥=𝑍𝑃 −𝛾to 𝑍 𝑃 do
17: if (𝑏𝑘𝑖.𝑛𝑢𝑚𝑏𝑒𝑟 mod 𝑚𝑖𝑛𝑒𝑟== 𝑖)then
18: flagresult = true
19: end if
20: end for
21: return flagresult
22: end function
23: function initialize(⋅)𝑤
24: while (True) do
25: 𝑍𝑃 ←previousblock(𝑠𝑖𝑔𝑛𝑆𝐷𝑁 𝐶𝑘). 𝑛𝑢𝑚𝑏𝑒𝑟
26: wait until latestsign(𝑠𝑖𝑔𝑛𝑆𝐷 𝑁𝐶𝑘,𝑍 𝑃 )
27: TP ←previous-timestamp (𝑠𝑖𝑔𝑛𝑆𝐷 𝑁𝐶𝑘)
28: wait until clock >=TP + blocktime
29: if (𝑆𝐷𝑁 𝐶𝑘+𝑖mod 𝑚𝑖𝑛𝑒𝑟== 𝑖)then
30: 𝑏𝑘.𝑤𝑒𝑖𝑔ℎ𝑡 = 2
31: else
32: delay(0, 500)* 𝑣𝑜𝑡𝑒
33: 𝑏𝑘.𝑤𝑒𝑖𝑔ℎ𝑡 = 1
34: end if
35: 𝑏𝑘.𝑛𝑢𝑚𝑏𝑒𝑟 =𝑍𝑃 + 1
36: 𝑏𝑘.𝑝𝑎𝑟𝑒𝑛𝑡 =previousblock(𝑠𝑖𝑔𝑛𝑆𝐷 𝑁𝐶𝑘)
37: 𝑏𝑘.𝑚𝑖𝑛𝑒𝑟 =𝑠𝑖𝑔𝑛𝑆𝐷 𝑁𝐶𝑘
38: 𝑠𝑖𝑔𝑛𝑆𝐷 𝑁𝐶𝑘
←(𝑆𝐵𝑖∪𝑏𝑘,𝑇 𝐵𝑖∪𝑏𝑘.𝑝𝑎𝑟𝑒𝑛𝑡)
39: end while
40: distribute (𝑠𝑖𝑔𝑛𝑆𝐷 𝑁𝐶𝑘)
41: end function
42: function WeightSum(𝑆𝐵𝑗,𝑇 𝐵𝑗)
43: return ∀𝑏𝑘∈𝑆𝐵𝑗𝑏𝑘.𝑤𝑒𝑖𝑔ℎ𝑡
44: end function
45: function Create(𝑆𝐵𝑗,𝑇 𝐵𝑗)
46: if WeightSum(𝑆𝐵𝑗,𝑇 𝐵𝑗)>WeightSum(𝑆 𝐵𝑖,𝑇 𝐵𝑖)then
47: WeigthSum(𝑆𝐵𝑖,𝑇 𝐵𝑖)←WeightSum(𝑆 𝐵𝑗,𝑇 𝐵𝑗)
48: end if
49: end function
50: function isCertian(𝑏𝑘)𝑘
51: VT ←{𝑏𝑘𝑖.𝑆𝐷𝑁 𝐶𝑘|𝑏𝑘𝑖∈𝑆𝐵𝑖}
52: return (𝑉 𝑇 >𝑣𝑜𝑡𝑒)
53: end function
2.2.2. Bi-GRU model for capturing temporal aspects
The proposed work used a DL-based detection scheme (Bi-GRU)
for an efficient threat detection in such a network. GRU is a type of
RNN with gating mechanism. A simple RNN can preserve history infor-
mation for an undetermined time in theory, however, it has gradient
exploding or gradient vanishing problems practically [17]. The GRU
is an enhanced edition of RNN having strong capacities for long-term
dependencies and is deemed to be less computationally complex due
Fig. 2. Network structure of Bi-GRU.
to its uncomplicated structure. The GRU uses two gates, reset Re𝑡and
update gate U𝑡, that results in a use of smaller number of parameters
to train it. In Bi-GRU, the final output at time 𝑡is concluded by the
preceding and next frame at time 𝑡− 1 and 𝑡+ 1. In Bi-GRU, one GRU
run forwards and one GRU runs in backward direction to calculate the
hidden states (h1,h2,h3, . . . , hn) of forward and backward directions.
A complete network structure of the Bi-GRU is shown in Fig. 2. The
following equations are used for computing the Bi-GRU hidden units:
Upt=𝜎(WXUpdt+WhUp ht−1+bUp ),(7)
Ret=𝜎(WXRedt+WhRe ht−1+bRe),(8)
Ct= tanh(WXc dt+Whc (Ret ⊙ht−1) + bC),(9)
ht= (1−Ut)⊙ht−1+Upt⊙Ct,(10)
where Up𝑡,Re𝑡are update and reset gates while C𝑡,h𝑡are candidate
cell and final state. The weight matrix for the input dtis denoted by
WXUp,WXRe , and WXc . Furthermore, WhUp ,WhRe and Whc denotes the
recurrent weight matrix between two consecutive h′
tsfor future prior
recurrent input ht−1and future recurrent input ht+1. The tanh denotes
the non-linear activation function, while ⊙represents the point-wise
multiplication.
2.2.3. Fully connected layers and softmax classifier for attack detection
The proposed IDS comprises two layers having 100 and 50 neurons
with a dropout rate of 0.2% respectively, which are set to prevent data
overfitting along with batch normalization to speed up the network
convergence. We have employed ADAM as an optimizer and RELU as
an activation function in the input layer and CC-E as a loss function.
The probability computation is then performed by a fully linked layer
followed by a Softmax layer. It computes a probability distribution over
a network’s non-normalized output for a given input sample. Finally,
the highest probability is calculated as the projected activity using
maximum likelihood estimation. The Eqs. (11) and (12) are used to
compute the probability and loss as
P𝑗=E𝑑𝑗
𝑁
𝑚=1 E𝑑𝑚 ,(11)
where P𝑗denotes the corresponding probability, is the sum of
multiple vectors, 𝑁represents the number of classes, and 𝑑𝑗 denotes
the input vector’s element.
L𝑜𝑠𝑠 = −
𝐶
I=1
Y𝑖𝑙𝑜𝑔
Y𝑖+𝜆1
L
L
𝐼=1
(1 −
T
𝑡=1
𝑚𝑡,1∕T)
+𝜆2
T
T
𝑡=1
(1 −
T
𝑖=1
𝑒𝑡,𝑖∕T) + 𝜆3
W𝑔,𝑓
2,
(12)
where Y𝑖is the true label,
Y𝑖is the 𝑖-th class sequence probability and
W𝑔,𝑓 denotes the BiGRU weight matrices.
Solar Energy 263 (2023) 111921
6
P. Kumar et al.
Algorithm 2 Proposed Detection Scheme
1: Input: Dataset= 𝐷𝑠𝑒𝑡
2: Output:𝐵𝑒𝑛𝑖𝑔𝑛 ←0,𝑀 𝑖𝑟𝑎𝑖.𝑎𝑐𝑘 ←1,𝑀𝑖𝑟𝑎𝑖.𝑆 𝑐𝑎𝑛 ←2, and so on.
3: Split 𝐷𝑠𝑒𝑡 in to 𝐷𝑠𝑒𝑡𝑇 𝑟 and 𝐷𝑠𝑒𝑡𝑇 𝑠
4: for each layer of BiGRU do
5: 𝐷𝑠𝑒𝑡′
𝑇 𝑟 =𝐷𝑠𝑒𝑡𝑇 𝑟 pre-processing
6: 𝐵𝑖𝐺𝑅𝑈𝑇 𝑟𝑀 𝑜𝑑𝑒𝑙 = Train BiGRU using 𝐷𝑠𝑒𝑡′
𝑇 𝑟
7: Upt=𝜎(WXUpdt+WhUp ht−1+bUp )
8: Ret=𝜎(WXRedt+WhReht−1+bRe )
9: Ct= tanh(WXc dt+Whc (Ret ⊙ht−1) + bC
10: ht= (1−Ut)⊙ht−1+Upt⊙Ct
11: end for
12: 𝐷𝑠𝑒𝑡′
𝑇 𝑠 =𝐷𝑠𝑒𝑡𝑇 𝑠 pre-processing
13: while True do
14: 𝑂𝑢𝑡𝑝𝑢𝑡 ←𝐵𝑖𝐺𝑅𝑈𝑇 𝑟𝑀 𝑜𝑑𝑒𝑙 (𝐷𝑠𝑒𝑡′
𝑇 𝑠)
15: if the predicted value = 0 then
16: Return Benign
17: else
18: Return the type of attack
19: end if
20: end while
3. Performance analysis
In this section, we discuss the details of our experiments and results
obtained from proposed framework. Specifically, this section is divided
into three parts: (𝐴)Experimental Setup, (𝐵)Numerical results of
blockchain-based authentication scheme and (𝐶)Numerical results of
DL-based IDS.
3.1. Experimental setup
The experiments are performed on a Tyrone PC with a 2 GHz
Intel(R) Xeon(R) Silver 4114 CPU, a RAM of 128 GB, and a hard
disk drive of 2 TB. The DL-based IDS is created using the TensorFlow
(TF) library Keras. Python programming language is used to run the
implementation scripts. The Ethereum Ropsten test network is used for
the blockchain experiment.
3.1.1. Overview of the dataset
This work used the N-BaIoT [47] dataset for the experimentation.
The NBaIoT dataset includes two distinct categories of IoT attack types:
Mirai and Gafgyt along with a Benign class. The Mirai and Gafgyt are
subdivided into multiple subclasses, i.e., Mirai.ack, scan, syn, UDP, UDP
plain, Gafgyt combo, junk, and TCP.
3.1.2. Data pre-processing
We removed any rows with blank or nan values, as they might have
an impact on data quality and the assessment model. Further, as the
DL algorithms primarily handle numeric data, we converted all non
numeric values to numeric ones using the label encoder, i.e., sklearn.
We also used the MinMax scalar function for data normalization [48].
3.1.3. Implementation details
The proposed IDS model comprises of two layers of Bi-GRU with
100 and 50 neurons and a dropout rate of 0.2%, respectively. The
model is set to prevent data overfitting along with batch normalization
to speed up the network convergence. We further employed an ADAM
optimizer. We used an RELU and a CC-E in the input layer as activation
and loss functions. The experiments were performed for 10 epochs with
a batch-size of 64.
Fig. 3. Analysis of blockchain-based authentication scheme in terms of transaction (tx)
upload and block mining time.
Fig. 4. Analysis of blockchain-based authentication scheme in terms of block creation
time and transaction (tx) storage.
3.1.4. Evaluation metrics
As the focus of this work was on multi-classification, macro-averaging
procedures were utilized to determine Accuracy (Ac), Detection Rate
(Dr), Precision (Pn), and F1-score (F1). In order to calculate these
metric, we have used various parameters named as True Positive Rate
(𝜖), True Negative Rate (𝛾), False Positive Rate (𝜂), and False Negative
Rate (𝜅). The Accuracy (Ac) is; 𝐴𝑐 =𝜖+𝛾
𝛾+𝜅+𝜖+𝜂. Detection Rate (Dr) or
Recall (Re) is 𝐷𝑟 =𝜖
𝜅+𝜖. Precision (Pn) is 𝑃 𝑛 =𝜖
𝜖+𝜂, and F1-Score is
𝐹1=2𝑃 𝑛×𝑅𝑒
𝑃 𝑛+𝑅𝑒 .
3.2. Numerical results of blockchain-based authentication scheme
Figs. 3 and 4illustrate the blockchain analysis in terms of execution
time for transaction (Tx) upload, mining of block, creation of block,
and the size (in KB) of transactions uploaded into off-chain storage.
The upload time of transaction indicates that original transactions
sharing over the off-chain storage layer. Figs. 3(b) and 4(a) illustrate
mining time of block and creation time of block with different set
of transactions and peers. The execution time analysis can be seen
growing with the amount of transactions and peers in the network.
Fig. 4(b) illustrates off-chain storage size (in KB) over the off-chain
storage layer for different set of transactions and peers. It can be seen
that the size is increasing as the number of transactions is increasing in
the network.
3.3. Numerical results of DL-based IDS
While evaluating a DL-based classification algorithm’s generalized
performance, confusion metrics are considered a substantial element. It
indicates the actual performance of an algorithm on the pre-determined
values of a test dataset. The proposed IDS is rationally examined to
monitor its confusion metrics performance, as shown in Fig. 5. The con-
solidated pattern of a true label and the predicted label demonstrated
Solar Energy 263 (2023) 111921
7
P. Kumar et al.
Fig. 5. Confusion matrix analysis.
Fig. 6. ROC analysis.
the efficiency of Bi-GRU. Further, the ROC curve exhibits conclu-
sive and comprehensive remarks regarding the stability of a DL-based
anomaly detection mechanism. It indicates the direct relationship be-
tween the TPR and FPR values accomplished by an algorithm. Fig. 6
exhibits the ROC curve analysis of the proposed IDS on various compo-
nents. It can be witnessed that we have obtained a stable straight line
that defines the appropriate ratio between TPR and FPR. The ROC is
further monitored at eight different classes, however, the conclusive
scale has projected equal area under all classes. That phenomenon
results in a straight end-to-end line to declare the reliability of the
proposed Bi-GRU.
While analyzing the performance efficiency of a DL-based threat
detection algorithm, Ac, Pn, Re, and F1 play a crucial role. From
Fig. 7, it can be seen that Bi-GRU achieves an Ac of 99.73%, while
LSTM and GRU achieve an Ac of 98.69% and 98.68% , respectively.
The higher Ac endorses the superiority of Bi-GRU over LSTM and
GRU. The same scenario is noticed regarding Pn, where Bi-GRU has
achieved a precision of 97.3%. The number is comparatively impressive
as competitive schemes such as LSTM and GRU could achieve 96.08%
and 96.38% Pn. A likewise sequence is monitored while investigating
the recall value. The proposed IDS has accomplished a Re value of
97.95% by beating the benchmarked schemes. The LSTM has shown a
92.37% Re score. However, the GRU only got stuck at 91.97%. The F1
determines the stability of a threat detection protocol. The proposed
Fig. 7. Overall comparison.
Table 1
Comparison with existing approaches.
Approaches Dataset Model SDN BC DT Ac (%)
[28] N-BaIoT Hybrid DL ✓× × 99.45
[30] N-BaIoT CNN-Stacked
LSTM
× × × 97.39
[34] N-BaIoT SVM × × × 95.90
Proposed Approach N-BaIoT SA-BiGRU ✓ ✓ ✓ 99.73
IDS has outclassed LSTM and GRU by achieving the F1 of 97.56%,
whereas the number was 93.87% for LSTM and 93.71% for GRU. The
exceptional performance exhibited by Bi-GRU in terms of Ac, Pn, Re,
and F1 advocates its efficiency over LSTM and GRU.
Likewise, other crucial performance parameters, TPR, TNR, and
NPV should also be included in ideal performance metrics. We have
investigated the performance of Bi-GRU in terms of these performance
parameters; additionally, the performance is then compared with its
competitive schemes, LSTM and GRU. It can be seen from Fig. 8 that
Bi-GRU has significantly achieved a 97.95% TPR, which is vividly
high as compared to the 91.87% TPR shown by LSTM. However, GRU
has demonstrated 91.97% TPR, which is slightly closer to the TPR
shown by LSTM; however still lower than the proposed Bi-GRU. That
phenomenon highly supports the validity of Bi-GRU over LSTM and
GRU. The performance of all these algorithms is evaluated in terms of
TNR, where Bi-GRU has demonstrated a 99.57% TNR by dominating
LSTM and GRU where the TNR was 99.2% and 99.17% respectively.
The more excellent value of TNR declares the higher efficiency of Bi-
GRU on a comparative scale with existing benchmarked schemes. NPV
is the following important performance parameter while interrogating
the real-time performance of a DL-based threat detection scheme. The
proposed IDS has substantially attained 99.63% NPV by suppressing
the performance of LSTM and GRU with 99.44% and 99.48% NPV,
respectively.
A DL-based intrusion detection model can be more accurately in-
spected by understanding its FDR, FNR, FOR, and FPR. The proposed
IDS is evaluated on these performance parameters, which are then ana-
lyzed on a comparative scale with competitive frameworks, e.g., LSTM
and GRU. Fig. 9 depicts that the Bi-GRU has 0.02668% FDR, which
is considerably lower than the 0.03612% FDR shown by GRU and
0.03894% FDR shown by LSTM. The lower FDR exhibited by Bi-GRU
is a solid declaration of its reliability. FNR determines the number of
adverse events that were negatively detected. Bi-GRU has exhibited
0.02041% FNR, remarkably reduced to the FNR shown by competitive
Solar Energy 263 (2023) 111921
8
P. Kumar et al.
Fig. 8. TPR, TNR and NPV comparison.
Fig. 9. FDR, DNR, FOR, and FRP comparison.
schemes. The LSTM has projected 0.07863% FNR; however, the FNR
value is 0.08021% for GRU. FOR is another performance measurement
standard that determines the actual validation of an algorithm. The
lower value of FOR demonstrates the efficiency of a DL-based threat
detection model. The proposed IDS has potentially achieved a lower
FOR of 0.000036% by beating the benchmarked schemes where GRU
has shown 0.0051% FOR; however, the value is 0.00556% for LSTM.
The proposed IDS has outpaced LSTM and GRU in terms of FPR.
The FPR has attained 0.000004% FPR, which is visually lower than
the 0.0073% FPR value achieved by LSTM and 0.00814% FPR value
demonstrated by GRU. Finally, we compare the performance of the
proposed IDS with some existing approaches, i.e., [28,30], and [34]
in Table 1. The table shows that the proposed IDS achieves a higher Ac
compared to existing approaches.
4. Conclusion
Smart grid (SG) aims to provide efficient energy management sys-
tem using SDN and DT technology. However, the communication be-
tween the SG entities take place using an insecure open channel and
DT heavily depends on the collected data for analysis. In this pa-
per, a blockchain-based authentication scheme and a deep learning-
based intrusion detection system were proposed. More specifically,
firstly, a mutual authentication between the smart meter and open flow
switches, and also between the service provider and open flow switches
were performed. After authentication, a session key was established
between them to securely transmit the collected data using the shared
session key. The authenticated data can be used by the upper planes
(i.e., control plane, digital twin and application plane) for further
analysis. The SPs use consensus algorithm to create and write data
into blockchain ledger to ensure privacy and integrity. Additionally,
a deep learning-based intrusion detection system was proposed by
combining a self-attention mechanism, a Bi-GRU model, fully connected
layers and a softmax classifier. The proposed approach was deployed
at control plane to further analyze the traffic in SG network. Numerical
results for blockchain and deep learning show the effectiveness of the
proposed framework. Future research will include the scalability testing
of the proposed framework using a larger number of smart meters and
different real-time datasets.
Declaration of competing interest
The authors declare that they have no known competing finan-
cial interests or personal relationships that could have appeared to
influence the work reported in this paper.
References
[1] R. Kumar, P. Kumar, A. Aljuhani, A.N. Islam, A. Jolfaei, S. Garg, Deep
learning and smart contract-assisted secure data sharing for iot-based intelligent
agriculture, IEEE Intell. Syst. (2022).
[2] S.A.A. Abir, A. Anwar, J. Choi, A. Kayes, Iot-enabled smart energy grid:
Applications and challenges, IEEE Access 9 (2021) 50961–50981.
[3] F. Al-Turjman, M. Abujubbeh, IoT-enabled smart grid via SM: An overview,
Future Gener. Comput. Syst. 96 (2019) 579–590.
[4] M.Y. Mehmood, A. Oad, M. Abrar, H.M. Munir, S.F. Hasan, H. Muqeet, N.A.
Golilarz, Edge computing for IoT-enabled smart grid, Secur. Commun. Netw.
2021 (2021).
[5] G. Fenza, M. Gallo, V. Loia, Drift-aware methodology for anomaly detection in
smart grid, IEEE Access 7 (2019) 9645–9657.
[6] P.H. Mirzaee, M. Shojafar, H. Cruickshank, R. Tafazolli, Smart grid secu-
rity and privacy: From conventional to machine learning issues (threats and
countermeasures), IEEE Access (2022).
[7] K.G. Boroojeni, M.H. Amini, S. Iyengar, Overview of the security and privacy
issues in smart grids, in: Smart Grids: Security and Privacy Issues, Springer,
2017, pp. 1–16.
[8] M. Babar, M.U. Tariq, M.A. Jan, Secure and resilient demand side management
engine using machine learning for IoT-enabled smart grid, Sustainable Cities Soc.
62 (2020) 102370.
[9] R. Khan, P. Maynard, K. McLaughlin, D. Laverty, S. Sezer, Threat analysis of
blackenergy malware for synchrophasor based real-time control and monitoring
in smart grid, in: 4th International Symposium for ICS & SCADA Cyber Security
Research 2016 4, 2016, pp. 53–63.
[10] D. Acarali, K.R. Rao, M. Rajarajan, D. Chema, M. Ginzburg, Modelling smart grid
IT-OT dependencies for DDoS impact propagation, Comput. Secur. 112 (2022)
102528.
[11] N. Saxena, B.J. Choi, State of the art authentication, access control, and secure
integration in smart grid, Energies 8 (10) (2015) 11883–11915.
[12] M. Alazab, L.U. Khan, S. Koppu, S.P. Ramu, I. M, P. Boobalan, T. Baker, P.K.R.
Maddikunta, T.R. Gadekallu, A. Aljuhani, Digital twins for healthcare 4.0 - recent
advances, architecture, and open challenges, IEEE Consum. Electron. Mag. (2022)
1–8, http://dx.doi.org/10.1109/MCE.2022.3208986.
[13] T. Cioara, I. Anghel, M. Antal, I. Salomie, C. Antal, A.G. Ioan, An overview
of digital twins application domains in smart energy grid, 2021, arXiv preprint
arXiv:2104.07904.
[14] C. Gehrmann, M. Gunnarsson, A digital twin based industrial automation and
control system security architecture, IEEE Trans. Ind. Inform. 16 (1) (2019)
669–680.
[15] R. Bitton, T. Gluck, O. Stan, M. Inokuchi, Y. Ohta, Y. Yamada, T. Yagyu, Y.
Elovici, A. Shabtai, Deriving a cost-effective digital twin of an ICS to facilitate
security evaluation, in: European Symposium on Research in Computer Security,
Springer, 2018, pp. 533–554.
[16] M. Eckhart, A. Ekelhart, Towards security-aware virtual environments for digital
twins, in: Proceedings of the 4th ACM Workshop on Cyber-Physical System
Security, 2018, pp. 61–72.
Solar Energy 263 (2023) 111921
9
P. Kumar et al.
[17] M. Eckhart, A. Ekelhart, A specification-based state replication approach for
digital twins, in: Proceedings of the 2018 Workshop on Cyber-Physical Systems
Security and Privacy, 2018, pp. 36–47.
[18] A. Bécue, Y. Fourastier, I. Praça, A. Savarit, C. Baron, B. Gradussofs, E.
Pouille, C. Thomas, CyberFactory# 1—Securing the industry 4.0 with cyber-
ranges and digital twins, in: 2018 14th IEEE International Workshop on Factory
Communication Systems (WFCS), IEEE, 2018, pp. 1–4.
[19] T. Zheng, M. Liu, D. Puthal, P. Yi, Y. Wu, X. He, Smart grid: Cyber attacks,
critical defense approaches, and digital twin, 2022, arXiv preprint arXiv:2205.
11783.
[20] M.B. Mollah, J. Zhao, D. Niyato, Y.L. Guan, C. Yuen, S. Sun, K.-Y. Lam, L.H.
Koh, Blockchain for the internet of vehicles towards intelligent transportation
systems: A survey, IEEE Internet Things J. 8 (6) (2020) 4157–4185.
[21] H.-N. Dai, Z. Zheng, Y. Zhang, Blockchain for s: A survey, IEEE Internet Things
J. 6 (5) (2019) 8076–8094.
[22] A.S. Musleh, G. Yao, S. Muyeen, Blockchain applications in smart grid–review
and frameworks, IEEE Access 7 (2019) 86746–86757.
[23] K. Park, J. Lee, A.K. Das, Y. Park, BPPS: Blockchain-enabled privacy-preserving
scheme for demand-response management in smart grid environments, IEEE
Trans. Dependable Secure Comput. (2022).
[24] Z. Wang, R. Huo, S. Wang, A lightweight certificateless group key agreement
method without pairing based on blockchain for smart grid, Future Internet 14
(4) (2022) 119.
[25] S.A.M. Taqi, S. Jalili, LSPA-SGs: A lightweight and secure protocol for authen-
tication and key agreement based elliptic curve cryptography in smart grids,
Energy Rep. 8 (2022) 153–164.
[26] A. Tomar, S. Tripathi, Blockchain-assisted authentication and key agreement
scheme for fog-based smart grid, Cluster Comput. 25 (1) (2022) 451–468.
[27] O.A. Omitaomu, H. Niu, Artificial intelligence techniques in smart grid: A survey,
Smart Cities 4 (2) (2021) 548–568.
[28] D. Javeed, T. Gao, M.T. Khan, D. Shoukat, A hybrid intelligent framework to
combat sophisticated threats in secure industries, Sensors 22 (4) (2022) 1582.
[29] M. Hanselmann, T. Strauss, K. Dormann, H. Ulmer, Canet: An unsupervised
intrusion detection system for high dimensional CAN bus data, IEEE Access 8
(2020) 58194–58205.
[30] D. Santhadevi, B. Janet, Stacked deep learning framework for edge-based
intelligent threat detection in IoT network, J. Supercomput. (2023) 1–34.
[31] P.I. Radoglou-Grammatikis, P.G. Sarigiannidis, Securing the smart grid: A com-
prehensive compilation of intrusion detection and prevention systems, IEEE
Access 7 (2019) 46595–46620.
[32] Q. Liu, V. Hagenmeyer, H.B. Keller, A review of rule learning-based intrusion
detection systems and their prospects in smart grids, IEEE Access 9 (2021)
57542–57564.
[33] Z. Lv, Y. Li, H. Feng, H. Lv, Deep learning for security in digital twins of
cooperative intelligent transportation systems, IEEE Trans. Intell. Transp. Syst.
(2021).
[34] T.B. Seong, V. Ponnusamy, N. Jhanjhi, R. Annur, M. Talib, A comparative
analysis on traditional wired datasets and the need for wireless datasets for
IoT wireless intrusion detection, Indonesian J. Electr. Eng. Comput. Sci. 22 (2)
(2021) 1165–1176.
[35] M.H. Rehmani, A. Davy, B. Jennings, C. Assi, Software defined networks-based
smart grid communication: A comprehensive survey, IEEE Commun. Surv. Tutor.
21 (3) (2019) 2637–2670.
[36] D. Javeed, T. Gao, M.T. Khan, SDN-enabled hybrid DL-driven framework for the
detection of emerging cyber threats in IoT, Electronics 10 (8) (2021) 918.
[37] U. Ghosh, P. Chatterjee, S. Shetty, Securing SDN-enabled smart power grids:
SDN-enabled smart grid security, in: Research Anthology on Smart Grid and
Microgrid Development, IGI Global, 2022, pp. 1028–1046.
[38] A. Aljuhani, Machine learning approaches for combating distributed denial of
service attacks in modern networking environments, IEEE Access 9 (2021)
42236–42264, http://dx.doi.org/10.1109/ACCESS.2021.3062909.
[39] Z. Lv, L. Wang, Z. Guan, J. Wu, X. Du, H. Zhao, M. Guizani, An optimizing and
differentially private clustering algorithm for mixed data in SDN-based smart
grid, IEEE Access 7 (2019) 45773–45782.
[40] H. Mahmood, D. Mahmood, Q. Shaheen, R. Akhtar, W. Changda, S-DPs: An
SDN-based DDoS protection system for smart grids, Secur. Commun. Netw. 2021
(2021).
[41] A. Jakaria, M.A. Rahman, A. Gokhale, Resiliency-aware deployment of SDN in
smart grid SCADA: A formal synthesis model, IEEE Trans. Netw. Serv. Manag.
18 (2) (2021) 1430–1444.
[42] L. Vishwakarma, D. Das, Blockchain-based security protocol for next-generation
smart grid transactions, TechScape (2020).
[43] B. Appasani, S.K. Mishra, A.V. Jha, S.K. Mishra, F.M. Enescu, I.S. Sorlei, F.G.
Bîrleanu, N. Takorabet, P. Thounthong, N. Bizon, Blockchain-enabled smart
grid applications: Architecture, challenges, and solutions, Sustainability 14 (14)
(2022) 8801.
[44] A.A.G. Agung, R. Handayani, Blockchain for smart grid, J. King Saud
Univ.-Comput. Inform. Sci. (2020).
[45] M.A. Ferrag, L. Maglaras, DeepCoin: A novel deep learning and blockchain-based
energy exchange framework for smart grids, IEEE Trans. Eng. Manage. 67 (4)
(2019) 1285–1297.
[46] P. Kumar, R. Kumar, S. Garg, K. Kaur, Y. Zhang, M. Guizani, A secure data
dissemination scheme for IoT-based e-health systems using AI and blockchain,
in: GLOBECOM 2022-2022 IEEE Global Communications Conference, IEEE, 2022,
pp. 1397–1403.
[47] Y. Meidan, M. Bohadana, Y. Mathov, Y. Mirsky, A. Shabtai, D. Breitenbacher,
Y. Elovici, N-baIoT: Network-based detection of IoT botnet attacks using deep
autoencoders, IEEE Pervasive Comput. 17 (3) (2018) 12–22.
[48] D. Javeed, M.S. Saeed, I. Ahmad, P. Kumar, A. Jolfaei, M. Tahir, An intelligent
intrusion detection system for smart consumer electronics network, IEEE Trans.
Consum. Electron. (2023).
