This article has been accepted for publication in IEEE Access.

This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3370239

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2017.DOI

A Novel Approach Based on Machine

Learning, Blockchain and Decision
Process for Securing Smart Grid
NABIL TAZI CHIBI1 , OMAR AIT OUALHAJ1 , WASSIM FASSI FIHRI1 , AND HASSAN EL
GHAZI1
1
STRS Laboratory, National Institute of Posts and Telecommunications, Rabat 10140, Morocco
Corresponding author: Nabil Tazi Chibi ([email protected]).

ABSTRACT Smart Grids (SGs) rely on advanced technologies, generating significant data traffic across
the network, which plays a crucial role in various tasks such as electricity consumption billing, actuator
activation, resource optimization, and network monitoring. This paper presents a new approach that
integrates Machine Learning (ML), Blockchain Technology (BT), and Markov Decision Process (MDP) to
improve the security of SG networks while ensuring accurate storage of events reported by various network
devices through BT. The enhanced version of the Proof of Work (PoW) consensus mechanism ensures data
integrity by preventing tampering and establishing the reliability of known and unknown attack detection.
The proposed versions of PoW, namely GPoW 1.0 and GPoW 2.0, aim to make the consensus process more
environmentally friendly.

INDEX TERMS Smart grid, cybersecurity, NIST, vulnerabilities, Markov decision process, blockcahin,
Machine Learning

I. INTRODUCTION Man in Middle attacks, Password theft, Spoofing, Replay,

Smart Grid (SG), also called Smart Electrical Network, was Data injection and Data modification.
first implemented in Italy in 2005 with the Telegestore project The objective of our paper is to employ Blockchain Tech-
[1]. The objective is, firstly, to promote the consumption nology (BT) in conjunction with the Machine Learning (ML)
of green energy in order to avoid the use of fossil fuels, paradigm and the Markov Decision Process (MDP). In the
which currently generate most of the world’s primary energy initial phase, ML will be utilized to categorize validators
[2]. Secondly, to enable consumers to become producers of participating in the Proof of Work (PoW) consensus into
this type of energy. The surplus of this energy could be fed two groups: Potential Winners and Potential Losers. Sub-
into the electricity grid. This concept is a revolution in the sequently, in the second phase, we will employ MDP to
field of electricity, where three levels interact: The electrical formulate the optimal policy for Potential Winners. This
network, the telecommunications network and information policy aims to streamline the competition by retaining only
technology. The control and monitoring of this network re- those validators with the highest likelihood of success in
quires the introduction of technologies used in Industry 4.0, validating and storing events in the BT Ledger. The ultimate
such as IIOT (Industrial Internet of Things), SCADA (System goal is to make our consensus model more environmentally
Control and Data Acquisition), SIEM (System Information sustainable.
Event Management) and data analytics [3]. This paper is structured as follows. In Section II, we
The big data (BD) generated by smart meters (installed in introduce the smart grid domains and their vulnerabilities.
homes to track and bill the amount of electricity consumed In Section III, we discuss some related work. Section IV
and fed into the grid), IIOT and various SG network devices presents our methodology. In Section V we highlight the
requires a very advanced security policy to guarantee the results obtained and finally in Section VI we conclude and
availability, confidentiality and integrity of the information. present some perspectives.
According to [4] and [5], the security of an SG could be
attacked mainly by Wormholes, Flooding, Puppet attacks, II. SMART GRID AND CYBER-SECURITY

VOLUME 4, 2016 1

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3370239

Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

A. SMART GRID OVERVIEW There are best practices in the literature to counter-attacks
The National Institute of Standard and Technology (NIST) such as [18] [23], disabling unused services and software,
[14] has proposed a conceptual model (see Fig. 1) which controlling access to resources, changing default passwords
makes it possible to clearly identify the different components, and accounts, using complex passwords, using logging to
actors and actions that are involved in an SG network. The monitor activity, restricting communication on a segmented
network, protecting against malware, sending regular signals
(Heartbeat Signals) by the components to indicate that their
state is normal, installing equipment that conforms to stan-
dards, using anonymous key agreement and mutual authen-
tication protocol[39], dynamic ephemeral and session key
generation protocol [40], lightweight authentication mecha-
nism [41], cryptographic for safeguarding information and
communication [42][43] ...etc.
In this work, we want to guarantee the reliability of the
events generated by the network equipment by storing these
data using the BT. We will therefore start by studying works
in which the BT has been used in the SG field.

III. RELATED WORK

FIGURE 1. Smart Grid’s framework based on NIST
Through the study of work carried out on BT and its use
in smart grid networks, we find that it mainly affects the
NIST model of the SG is organized into seven domains [19]:
operational aspect [6]. To summarize, this technology could
• Generation: The production of the electricity using tra- be used :
ditional and ecological technologies.
• Transmission: The transmission of electricity over long • To secure the AMI (Advanced Metering Infrastructure)
distances. With the ability to generate and store the transactions. The Blockchain has been tested in storing
electricity. data related to the energy consumed for billing reasons
• Distribution: distribution of electricity to and from the and applying the rules defined by the operator through
end users. smart contracts.
• Markets: Electricity market participants for better sales. • On Trading and Market, to secure energy sales transac-
• Consumer: Traditionally, the end user’s role was to con- tions between individuals without third-party intermedi-
sume electricity. In this new configuration, they could aries.
generate, store, deliver and manage energy. • For monitoring and control to enforce the reliability
• Operations: managing the flow of electricity. of data that maintains the proper functioning of the
• Provider: providing services to utilities and customers. infrastructure components.
• To ensure the non-alteration of data used in an electrical
protection platform, called an Adaptive Protection Plat-
B. SMART GRID VULNERABILITIES AND
form (APP) [7].
COUNTERMEASURES
• For storing data from smart sensors based on Ethereum
The cybersecurity objectives for the SG, as defined by NIST technology and applying certain rules using smart con-
[14] are: tracts [8].
• confidentiality : protecting the SG from attacks that tries • To regulate the communication between the utility and
to access to private data the smart meters based on the smart contract [9].
• Availability : keeping the information accessible • To Secure the storage of events coming from IOTs on a
• Integrity : ensuring that the data is not altered Smart Grid Network using PoW Consensus [10].
The main sources of vulnerability in an SG [16] are software • For the management, control and operation of an SG
with security flaws, incorrect configurations, unprotected Network [11].
communication lines, lack of maintenance, an unprotected Upon scrutinizing the existing literature and conducting a
network, etc. comparative analysis of the technologies enumerated in Table
Attacks on an SG could be at the level of components, 1 for securing the smart grid, it is evident that none of these
communications and protocols [15] [20] or topologies. These researches have integrated all three technologies: Blockchain
attacks could take several forms [24][25], such as : Malware Technology (BT), Machine Learning (ML), and Markov
and virus spreading, database links, communication hijack- Decision Process (MDP) to enhance the energy efficiency
ing, replay attack with false data injection, Modbus protocol of the Proof of Work (PoW) consensus while concurrently
privacy, interception of network traffic, unplanned shutdown establishing an effective security framework for the smart
[17]. grid. This unique combination is a novel aspect that our paper
2 VOLUME 4, 2016

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3370239

Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

endeavors to explore and contribute to the existing body of

knowledge.

TABLE 1. Comparison of the technologies used (BT, ML and MDP) to secure

the SG in related work.

Related Work BT ML MDP Comment

[33] X BT proposed as a solution to secure
SG
[34] X BT and smart contracts to improve
the reliability of transactions
[33] X Secure data awareness model using
blockchain
[36] X Proposal of an abstract system for
the prevention of data injection
[37] X X In this survey, AI/ML and BT were
identified as a solution for cyberse-
curity chalenges FIGURE 2. Machine learning workflow
[27] X X BT ledger with an AI-enabled se-
cure framework for SG
[28] X X Analysing solutions using different
methods based on IA/ML and BT
[29] X X Proposal of a model to improve the decentralized way, called DLT (Distributed Ledger Technol-
security of the SG based on the BT, ogy). The Blockchain has mechanisms that will ensure that
and industrial fault detection using
wireless sensors Network and deep this data will be unaltered within a redundant system that is
learning techniques resilient to failures and cyber-attacks. Blockchain technology
[44] X Securing industrial IoTs data in the
Electric SG using a blockchain- could be public or private and it is based on different types of
based system consensuses for the validation and insertion of new blocks in
the Blockchain, such as [6] and [12]:
Our objective and contribution revolve around enhancing
the energy efficiency of the Proof of Work (PoW) consensus, • Proof of Work (PoW): Here the nodes are called miners
a concept often acknowledged as energy-intensive in existing and must solve a calculation of a very complicated prob-
literature [31]. Simultaneously, we aim to establish a sus- lem, the node which finds the solution first broadcasts
tainable security model for the smart grid. To achieve this, it to the other nodes for verification and insertion in
we propose the utilization of Machine Learning (ML) and the distributed ledger (DL). The disadvantage of this
Markov Decision Process (MDP) to create two environmen- consensus is the very high energy consumption. This
tally friendly versions of PoW, namely GPoW 1.0 and GPoW consensus is used by the Bitcoin.
2.0. These versions will be instrumental in storing events gen- • Proof of Stake (PoS): Here the nodes are called Valida-
erated by Smart Grid (SG)-enabled devices. The integration tors. Block validation is randomly granted to nodes that
of Blockchain Technology (BT) ensures the protection of all have the most and oldest shares.
stored data, maintaining transaction integrity and eligibility. • Delegated Proof of Stake (DPoS): In this consensus,
the validators will be a subset of the PoS validators for
IV. METHODOLOGY whom the task of validation will be delegated.
Our approach will be based on blockchain technology (BT) to • Leased Proof of Stake (LPoS): Here, nodes lease their
store events from SG devices in a decentralized way employ- assets to other nodes to increase the probability of being
ing machine learning [26] and the Markov decision process a validator. Profits will be shared among the members of
in order to reduce energy consumption and for better load each node group.
balancing between nodes, ensuring validation and storage of • Proof of Activity (PoAc): It is a consensus that start by
these events. using PoW and once the miners reap enough rewards the
system switches to PoS using the rewards as stakes.
A. MACHINE LEARNING • Proof of Burn (PoB): Validators will need to regularly
The Machine Learning [21] Fig.2, in our approach, will help burn some of their own coins to increase their chances
us to develop a Prediction Model (PM) from experimental of being selected. This action is done by sending some
Data (Dataset) based on a Supervised Learning techniques coins to public and verifiable addresses. This consensus
using a Classification Model that allows to identify which is used by the Slimcoin.
category an object belongs to. • Proof of Authority (PoA): Validators will be pre-
selected and authorized to validate blocks based on their
B. BLOCKCHAIN TECHNOLOGY identity. Etherum is based on this type of consensus.
Blockchain technology [22] is based on a distributed data
storage and processing, which avoids the monopoly of power Other consensus variants exist like: Proof of Inclusion (PoI),
by a single authority. This storage and processing will be Proof of Elapsed Time (PoET) and Practical Byzantine Fault
carried out by several nodes/computers on the network. The Tolerance (PBFT).
data will be formatted and stored as a chain of blocks in a
VOLUME 4, 2016 3

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3370239

Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

C. MARKOV DECISION PROCESS APPROACH The events generated by the IoTs and Smart Meter will
The MDP (Markov Decision Process) [13] allows the mod- cross the sentinel, which is responsible for centralizing events
elling of a process for decision-making assistance. At a given and transferring them to the IoTSSP (IoT Security Service
time, the process is in a state S. Several actions will be Provider) which will either authorize the processing of the
possible in this state, the decision maker must choose an event or will block it based on a blocklist and the Machine
action a among them. The process will react by moving Learning Algorithms as presented in the flow chart Fig.4.
randomly (according to a probability Pa (S, S ′ )) to another
state S ′ and will offer a reward ra (S, S ′ ) to the decision
maker.
The goal will be to find the optimized policy for the
decision maker.

D. OUR APPROACH
The advantage of our work is to propose a model based on
PoW consensus enhanced by some adaptations to make it
very energy efficient. In our model, the authority will be
granted to a certain number of validators, which will compete
to validate and store in the Blockchain ledger the received
event to maximize the profit by deciding at each state the
action to be taken (bet or not bet) depending on the conditions
and the context. Our model will have the advantage of not
consuming too much energy and helping validators to make
the right decisions at each time, by using ML and MDP for
better resource optimization and fair benefit sharing.

E. OUR ARCHITECTURE
In this work, we will focus on the equipment installed in
private homes because, in our opinion, they pose the most FIGURE 4. Flow Chart – Inserting an event.
security risks due to several reasons, such as willingness to
falsify bills, the types of equipment are not being updated or If the event is legitimate, the IoTSSP broadcasts the event
are not compliant with the standard, the local area network to all the validators to launch the competition between several
(LAN) is not protected, and other such factors. In our Fig.3 nodes validators (NV) of the network whose objective is to
architecture, we consider that the customer (e.g., a house) has maximize their profit by winning the reward (Gas) offered
mainly several IoTs and a smart meter connected to its LAN by the equipment which generated the event. The winner will
that uses a home router to communicate with the electricity validate the event and ask the other nodes to store this event
provider via the Internet. in a decentralized way based on BT.
In order to make the competition between the nodes sig-
nificantly less energy-intensive, we will use an improved
version of the PoW consensus that we will call GPoW (Green
Proof of Work). This consensus will use ML to classify nodes
based on a list of futures and history. A node before deciding
to compete and thus consuming energy, It will estimate its
chances to win using ML, the node will be classified as
potential winner (PW) or potential loser (PL).
We will produce two versions of the GPoW consensus,
GPoW 1.0, where the ML layer is added to the classical PoW
consensus, and GPoW 2.0, which is an improved version of
the GPoW 1.0 because we add the Markov Decision Process.

F. GPOW 1.0 : ADDING THE MACHINE LEARNING

LAYER
When an event is broadcasted, the nodes receive as informa-
tion: EventSource, Gas and TTL. Where:
FIGURE 3. Proposed architecture to secure smart grids. • EventSource : The ID of the device that sent the event
4 VOLUME 4, 2016

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3370239

Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

• Gas : The gas offered by the device for the validation of – p : The probability that a validator V who has bet to
the event be chosen from among all the validators who have
• TTL : The TTL field will have a given value on an bet
IP packet when the device sends it. This value will Based on the probability tree Fig.5, we can deduce the
be decremented each time the packet passes through probability p.
a router on the Internet network. Once arrived at the
destination, it will allow for estimating the time elapsed
before the packet’s arrival.
The features that will allow us to apply machine learning
will be the information received and listed above, plus the
following three features:
• TimeFrameImpact : The day will be divided into twelve
time slots of two hours. Initially, the value of the Time-
FrameImpact variable will be fifty for all nodes and
on all slots. After each processing request (Addition
of an event in the Blockchain), the value of the Time-
FrameImpact variable will be re-evaluated for all nodes
on the slot corresponding to the processing time. We
will remove points (one for example) for the losing
nodes, and we will add points (one for example) for the
winning node, with zero as the minimum and a hundred
as the maximum value of the TimeFrameImpact.
• CPU : The available CPU of the node
• Memory : The available memory of the node
• Bandwidth : The available bandwidth of the node
FIGURE 5. Probability tree – The validators are identical.
Each validator will have his features (Gas, TTL, Time-
FrameImpact, CPU, Memory, Bandwidth) to be able to apply The probability p can be calculated through the formula
the prediction model locally (which will be updated regularly (1), below:
based on previous results) for a self-classification (PW or PL)
to decide whether to participate or not. This classification Xn
1 C m−1
would normally eliminate numerous potential losers, which p= × Pnn−1 k (1)
m k=0 Cn
would have a very significant impact on reducing energy m=1
consumption. The competition between validators could be schematized
through the Markov Decision Process (Fig. 6).
G. GPOW 2.0 : APPLYING THE MARKOV DECISION
PROCESS
In order to make it clear, we will apply our approach on 2 use
cases, in the first one we will consider that all the validators
are identical and in the second one, we will have 2 kinds of
validators.
Model 1 : The Blockchain validators are identical
Scenario description:
We assume that we have n validators/players.
• Each player has 2 possible actions :

– 1 : to bet
– 2 : wait (do not bet)
• The player can switch between 2 states : FIGURE 6. Markov Decision Process – The validators are identical.
– F : Free
– B : Busy Our model is based on the formulas 1, 2, 3, 4 and 5.
• We will focus on one validator V . Because the scenario
 
1−p p
is the same for the other validators. P (.; .; 1) = (2)
1 0
• We will consider that:

– b : The bet value

 
1 0
– r : The reward value P (.; .; 2) = (3)
1 0
VOLUME 4, 2016 5

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3370239

Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

  (F;rf ) or (F;rs ), depending on the probabilities p(rf ) and

0 p(rs ).
U (.; 1) = (4)
0 The competition between validators could be schematized
 −b
 through the Markov Decision Process (Fig. 7).
U (.; 2) = n (5)
r−b
Where:
• P (.; .; 1) defines the transition matrix to go from a state
to another one if the player decides to bet (1). These
transitions are well schematized in Fig.6.
• P (.; .; 2) defines the transition matrix to go from a state
to another one if the player decides not to bet (2).
• U (.; 1) defines the reward matrix if the player decides to
bet (1) in the different possible states (F and B). FIGURE 7. Markov Decision Process – 2 categories of validators.
• U (.; 2) defines the reward matrix if the player decides
not to bet (2) in the different possible states (F and B). Our model is based on the formulas below.
Model 2 : two kinds of Blockchain validators n
X 1 C m−1
Scenario description: pf = × Pnn−1 k (6)
• We assume that we have n validators/players and that m=1
m k=0 Cn
n = ns + nf ns
1 C m−1
× Pnsns −1 k
X
• ns players from n players could bet bs and bf , such as: ps = (7)
bs > bf . m=1
m k=0 Cns
• nf players from the n players could bet only bf , such
 
(1 − pf ) × p(rf ) (1 − pf ) × p(rs ) pf × p(rf ) pf × p(rs ) 0 0
as: bs > bf .  (1 − ps ) × p(rf )
 (1 − ps ) × p(rs ) 0 0 ps × p(rf ) ps × p(rs )

• The player who has bet in a state will not be able to bet

P (.; .; 1) = 


p(rf )
p(rf )
p(rs )
p(rs )
0
0
0
0
0
0
0
0




(8)
p(rf ) p(rs ) 0 0 0 0
in the next state. 
p(rf ) p(rs ) 0 0 0 0


• We will focus on the ns validators, because for the nf  

validators the decision will be simple for them: They p(rf ) p(rs ) 0 0 0 0
p(rf ) p(rs ) 0 0 0 0
have to bet bf each time.  
• Each player can be Free (F) or Busy (B) or Saturated (S) p(rf )
P (.; .; 2) =  p(rs ) 0 0 0 0 (9)
p(rf ) p(rs ) 0 0 0 0
– Free: all the resources of the player are available
p(rf ) p(rs ) 0 0 0 0
– Busy : The player bet a part of his resources bf ,
 
such as (bs > bf ) 0
– Saturated : The player bet all his resources bs , such 0
 
as (bs > bf ) 0
U (.; 1) = 0
 (10)
• A reward will be randomly : rs or rf and we assume  
that rf < rs 0
• The possible states are: (F;rf ); (F;rs ); (B;rf ); (B;rs ); 0
(S;rf ); (S;rs )  −bf 
• p(rf ) is the probability that rf occurs and p(rs ) is the n
 −bs 
probability that rs occurs.  ns 
r − b 
• We consider that when the reward is rf and a player bet, U (.; 2) =  f f (11)
r − b 
the value of his bet will be bf . When the reward is rs  f f
r − b 
and a player bet, the value of his bet will be bs . s s
• Actions are : 1 : bet; 2 : Wait rs − bs
• For the winner: Where:
– The payoff will be (rf -bf ) or (rs -bs ) • P (.; .; 1) defines the transition matrix to go from a state
– And we assume that (rf -bf ) < (rs -bs ) [(F;rf ); (F;rs ); (B;rf ); (B;rs ); (S;rf ); (S;rs )] to another
• pf is the probability for a player who bet bf to be one if the player decides to bet (1).
chosen. And ps is the probability that one of the ns • P (.; .; 2) defines the transition matrix to go from a state
players who bet bs to be chosen. to another one if the player decides not to bet (2).
• Our objective is to find the best policy for the ns players • U (.; 1) defines the reward matrix if the player decides to
to maximize their total payoffs. bet (1) in the different possible states.
Whatever the action chosen in the states (B;rf ), (B;rs ), (S;rf ) • U (.; 2) defines the reward matrix if the player decides
and (S;rs ) the user will return to the starting box either to not to bet (2) in the different possible states.
6 VOLUME 4, 2016

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3370239

Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

V. RESULTS AND DISCUSSIONS

A. GPOW 1.0 : SIMULATION RESULTS:
We apply our approach to 10 nodes, taking into account the
time factor. When starting our simulation, we consider that
the impact of time slots is identical (=50) for the different
nodes. In each loop, we add one point (+1) to the time slot
impact if the node won, and we subtract one point (-1) from
the time slot impact if the node loses. After a while, we found
ourselves with the Table.2 as time slot impacts.

TABLE 2. Slot time impacts on the potential gain or loss of nodes

Node1 Node2 Node3 Node4 Node5 Node6 Node7 Node8 Node9 Node10
00h-02h 94 40 31 89 36 76 39 53 46 53
02h-04h 11 33 74 59 25 50 43 6 75 14
04h-06h 30 29 70 5 78 77 17 66 62 54
06h-08h 52 14 26 7 19 47 89 31 87 39
08h-10h 48 93 89 41 25 94 66 86 29 43
10h-12h 66 48 68 24 29 71 20 72 75 31
12h-14h 34 10 30 46 48 91 67 58 52 25
14h-16h 83 58 57 75 68 10 74 80 15 86
16h-18h 87 76 95 63 43 57 27 10 7 94
18h-20h 68 46 87 86 9 9 49 61 9 25
20h-22h 45 31 88 18 76 75 83 16 7 41 FIGURE 8. Model 2 Simulation Results.
22h-24h 92 39 44 30 17 46 93 51 40 60

We consider that we are launching the competition be- In the second simulation, we set all the parameters and we
tween these 10 nodes in the different time slots for different vary the number of validators with the higher performance:
Gas values (integer numbers ranging from 10 to 100). These ns ∈ [1..10]; n=10 ; rf =4; rs =10 ; bf =2 ; bs =6 p(rf ) =2/3 ;
nodes will have different performance in terms of memory, p(rs )=1/3
processor and bandwidth. The TTL (Time To Live) will be The results are visible in the graph below (Fig. 9).
influenced by the path traced by the event packet. This simu-
lation will generate a dataset of more than 10000 lines, with
as features : Gas, TTL, TimeFrameImpact, CPU, Memory,
Bandwidth. The label of these features will be 1 (if the node
win) and 0 (if the node lose)
Based on this dataset and applying ML, we can predict for
a given node whether it is a potential winner or a potential
loser. We use the KNeighbors classifier and the GaussianNB
model, which have accuracies exceeding 0.93. The competi-
tion will be between the limited number of potential winners.

B. GPOW 2.0 : MODEL 1 - SIMULATION RESULTS:

Through the application of the Reinforcement Learning Al-
gorithm, and by varying the values of n (the number of
validators), r (the value of the reward) and b (the value of
the bet), we always find that the optimal policy is :
• If the node is in state F (Free) it is recommended to bet
• If the node is in state B (Busy) it is recommended not to
FIGURE 9. Model 2 Simulation Results.
bet
With the increase in the number of the validators with
C. GPOW 2.0 : MODEL 2 - SIMULATION RESULTS: higher performance, we notice that the occupancy rate of
We set all the parameters and we vary the value of the higher stages (S;rf ) and (S;rs ) decreases and that it increases for
bet : bs ∈ [2..10]; n=7 ; ns =3 ; rf =3; rs =10 ; bf =2 ; p(rf ) stages (F ;rf ) and (F ;rs ). On the other hand the rate of
=2/3 ; p(rs ) =1/3 stages (B;rf ) and (B;rs ) remains stable.
The results are visible in the graph below (Fig. 8).
We notice that in this case the rate of the occupancy In the third simulation, we set all the parameters and we
of the different states by the validators with the higher vary the Probability that rs (a superior reward) occurs :
performance remains fixed. This will only impact the payoff p(rs ) ∈ [0, 1 : 0, 1 : 0, 9]; n=10; ns =4; rf =4; rs =10; bf =2;
of the validator. bs =6; p(rf )=1-p(rs )
The results are visible in the graph below (Fig. 10).
VOLUME 4, 2016 7

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3370239

Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

TABLE 3. Example of 10 nodes at a given time

Node TimeFrameImpact CPU Memory Bandwidth TTL Gas

Node1 94 1 8 54 82 10
Node2 85 1 8 36 41 10
Node3 92 1 8 40 28 10
Node4 89 1 8 40 87 10
Node5 36 1 8 23 59 10
Node6 76 1 8 53 39 10
Node7 39 1 8 50 6 10
Node8 53 3 32 33 71 10
Node9 82 2 16 68 15 10
Node10 53 3 32 22 11 10

TABLE 4. Results using GPoW 1.0 Consensus

Node Result
Node1 Potential loser
Node2 Potential loser
Node3 Potential winner
Node4 Potential loser
Node5 Potential loser
Node6 Potential winner
FIGURE 10. Model 2 Simulation Results. Node7 Potential loser
Node8 Potential loser
Node9 Potential winner
Node10 Potential winner
With the increase of this probability, we notice that the
occupancy rate of stages (S;rs ) increases. The aim of these
simulations is to see the effect of the conditions (value of the
bet, number of nodes, frequency of the different rewards) to
increase the chances of the nodes (potential winners selected
in the first round using GPoW 1.0) to be in a saturated
state, i.e. to win the competition to further reduce power
consumption through our GPoW 2.0 consensus proposal.

D. A USE CASE :
Let us assume that we have a list of 10 nodes as shown in the
table 3.
In the classical PoW, all these nodes must compete to mine
the event received.

Using our approach, in the first step (GPoW 1.0 consensus)

we apply the ML (Reinforcement Learning) to find the results
in the table 4.
FIGURE 11. Model 2 Simulation Results.
We can see that the number of competitors is reduced to 4
nodes.
VI. CONCLUSION
Now let’s use MDP to upgrade to version 2 of our GPoW Opening smart grids to the Internet requires significant in-
consensus. novation to counter the rapid evolution of attack techniques.
We assume that : n=4; ns =2; p(rs )=0.3; p(rf )=0.7; rf =10; Our approach reliably stores the events generated by the vari-
rs =20; bf =5; bs =12. ous devices in the smart grid. We decentralise the information
Through the policy iteration of MDP algorithm, we find that by involving the network nodes in the event validation and
the optimal policy is that the node must always bet when it’s storage process using an enhanced version of the Proof of
in the Free states ((F;rf ) or (F;rs )). Work (PoW) consensus, which we call GPoW 1.0, a proposal
that uses machine learning based on historical transactions to
Using the stationary distribution of MDP algorithm, we can allow nodes to predict their potential to win or lose, reducing
see from the Fig. 11 that the occupancy rate of the free states the number of nodes competing and thus the total amount
(the first and second bars) does not exceed 80%, so the node of energy consumed. Our second version GPoW 2.0, is an
will not bet in 20% of the cases. Which will again reduce improved version of GPoW 1.0, because it will apply on
energy consumption. the nodes resulting from this first phase (GPoW 1.0), the
algorithms of the MDP to define the best policy for this
8 VOLUME 4, 2016

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3370239

Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

reduced number of nodes to keep in competition only the [15] Zoya Pourmirza, Aditya Srivastava, “Cybersecurity Analysis for the Com-
nodes with maximum of chance to win, thus making our munication Protocol in Smart Grids”, 2020 IEEE 8th International Confer-
ence on Smart Energy Grid Engineering (SEGE), 2020
consensus greener. [16] I. S. Stoyanov, T. B. Iliev, G. Y. Mihaylov, B. I. Evstatiev, S. A. Sokolov,
As a perspective, we plan to work on the use of events coming “Analysis of the Cybersecurity Threats in Smart Grid”, 2018 IEEE 24th
from smart grid network devices and stored in a reliable and International Symposium for Design and Technology in Electronic Pack-
aging (SIITME), 2018
eco-friendly way, using GPoW 2.0 to use machine learning [17] Pallab Ganguly, Mita Nasipuri, Sourav Dutta, “Challenges of the Existing
to detect known and unknown cyber-attacks. Security Measures Deployed in the Smart Grid Framework”, 2019 IEEE
Combining our approach with prior research work 7th International Conference on Smart Energy Grid Engineering (SEGE),
2019
[27][28][29] rooted in BT and ML opens up the possibil- [18] Drd. Irene Ionita, “Cybersecurity concerns on real time monitoring in
ity for exploring alternative research avenues, addressing electrical transmission and distribution systems (SMART GRIDS)”, 2019
diverse security aspects of SG, or delving into topics like 54th International Universities Power Engineering Conference (UPEC),
2019
IoT[30][31][32] and the cognitive radio[38]. [19] Zakaria ElMrabet, Naima Kaabouch, Hassan El Ghazi, Hamid El Ghazi,
“Cyber-security in smart grid: Survey and challenges”, Elsevier, May 2018
REFERENCES [20] Meng Fanlin, Ye Wei, “Summary of Research on Security and Privacy of
[1] A. T. Yee Chong, M. A. Mahmoud, F. -C. Lim and H. Kasim, Smart Grid”, 2020 International Conference on Computer Communication
"A review of Smart Grid Technology, Components, and Implementa- and Network Security (CCNS), August 2020
tion," 2020 8th International Conference on Information Technology [21] Ahmet Aytuğ Ayrancı, Hacı İlhan, “Decentral Smart Grid Control System
and Multimedia (ICIMU), Selangor, Malaysia, 2020, pp. 166-169, doi: Stability Analysis Using Machine Learning”, 2022 2nd International Con-
10.1109/ICIMU49871.2020.9243430. ference on Computing and Machine Intelligence (ICMI), July 2022
[2] F. Ayadi, I. Colak, I. Garip and H. I. Bulbul, "Impacts of Renewable [22] Qinglei Guo, Shuai Chen, Junsheng Wang, Xiaoting Pan, “Research and
Energy Resources in Smart Grid," 2020 8th International Conference Design of Electric Power Engineering Project Management System Based
on Smart Grid (icSmartGrid), Paris, France, 2020, pp. 183-188, doi: on Blockchain Technology”, 2022 International Conference on Blockchain
10.1109/icSmartGrid49881.2020.9144695. Technology and Information Security (ICBCTIS), August 2022
[3] S. Tripathi, P. K. Verma and G. Goswami, "A Review on SMART GRID [23] Ahmed S. Musleh, Guo Chen, Zhao Yang Dong, Chen Wang, Shiping
Power System Network," 2020 9th International Conference System Mod- Chen, “Vulnerabilities, Threats, and Impacts of False Data Injection At-
eling and Advancement in Research Trends (SMART), Moradabad, India, tacks in Smart Grids: An Overview ”, 2020 International Conference on
2020, pp. 55-59, doi: 10.1109/SMART50582.2020.9337067. Smart Grids and Energy Systems (SGES), November 2020
[4] A. I. Kawoosa and D. Prashar, "A Review of Cyber Securities [24] El Mrabet, Z., Ezzari, M., Elghazi, H., & El Majd, B. A. (2019, March).
in Smart Grid Technology," 2021 2nd International Conference on Deep learning-based intrusion detection system for advanced metering
Computation, Automation and Knowledge Management (ICCAKM), infrastructure. In Proceedings of the 2nd international conference on
Dubai, United Arab Emirates, 2021, pp. 151-156, doi: 10.1109/IC- networking, information systems & security (pp. 1-7).
CAKM50778.2021.9357698. [25] El Mrabet, Z., El Ghazi, H., & Kaabouch, N. (2019, May). A performance
[5] R. Marah, I. E. Gabassi, S. Larioui and H. Yatimi, "Security of Smart comparison of data mining algorithms based intrusion detection system for
Grid Management of Smart Meter Protection," 2020 1st International smart grid. In 2019 IEEE International Conference on Electro Information
Conference on Innovative Research in Applied Science, Engineering Technology (EIT) (pp. 298-303). IEEE.
and Technology (IRASET), Meknes, Morocco, 2020, pp. 1-5, doi: [26] W. F. Fihri, H. E. Ghazi, B. A. E. Majd and F. E. Bouanani, "A Ma-
10.1109/IRASET48871.2020.9092048. chine Learning Approach for Backoff Manipulation Attack Detection in
[6] M. B. Mollah et al., "Blockchain for Future Smart Grid: A Comprehensive Cognitive Radio," in IEEE Access, vol. 8, pp. 227349-227359, 2020, doi:
Survey," in IEEE Internet of Things Journal, vol. 8, no. 1, pp. 18-43, 1 10.1109/ACCESS.2020.3046637.
Jan.1, 2021, doi: 10.1109/JIOT.2020.2993601. [27] Ayub Khan, Abdullah and Laghari, Asif and Rashid, Mamoon and Li,
[7] D. Sikeridis, A. Bidram, M. Devetsikiotis and M. J. Reno, "A blockchain- Hang and Javed, Abdul Rehman and Gadekallu, Thippa. (2023). Artificial
based mechanism for secure data exchange in smart grid protection intelligence and blockchain technology for secure smart grid and power
systems," 2020 IEEE 17th Annual Consumer Communications and Net- distribution Automation: A State-of-the-Art Review. Sustainable Energy
working Conference (CCNC), Las Vegas, NV, USA, 2020, pp. 1-6, doi: Technologies and Assessments. 57. 103282.
10.1109/CCNC46108.2020.9045368. [28] Mazhar, Tehseen and Irfan, Hafiz and Khan, Sunawar and Haq, Inayatul
[8] Q. Xu, Z. He, Z. Li and M. Xiao, "Building an Ethereum-Based Decentral- and Ullah, Inam and Iqbal, Muhammad and Hamam, Habib. (2023).
ized Smart Home System," 2018 IEEE 24th International Conference on Analysis of Cyber Security Attacks and Its Solutions for the Smart Grid
Parallel and Distributed Systems (ICPADS), Singapore, 2018, pp. 1004- Using Machine Learning and Blockchain Methods. Future Internet. 15.
1009, doi: 10.1109/PADSW.2018.8644880. 10.3390/fi15020083.
[9] R. Akhras, W. El-Hajj, M. Majdalani, H. Hajj, R. Jabr and K. Sha- [29] Kandasamy, Manivel and Anto, S. and Baranitharan, K. and Rastogi, Ravi
ban, "Securing Smart Grid Communication using Ethereum Smart and Satwik, Gunda and Sampathkumar, A.. (2023). Smart Grid Security
Contracts," 2020 International Wireless Communications and Mobile Based on Blockchain with Industrial Fault Detection Using Wireless
Computing (IWCMC), Limassol, Cyprus, 2020, pp. 1672-1678, doi: Sensor Network and Deep Learning Techniques. Journal of Sensors. 2023.
10.1109/IWCMC48107.2020.9148345. 10.1155/2023/3806121.
[10] J. V. Botello, A. P. Mesa, F. A. Rodríguez, D. Díaz-López, P. Nespoli, [30] Gupta, Rajesh and Jadav, Nilesh and Nair, Anuja and Tanwar, Sudeep and
et F. G. Mármol, “BlockSIEM: Protecting Smart City Services through a Shahinzadeh, Hossein. (2022). Blockchain and AI-based Secure Onion
Blockchain-based and Distributed SIEM“, Sensors, vol. 20, no 16, Art. no Routing Framework for Data Dissemination in IoT Environment Under-
16, janv. 2020, doi: 10.3390/s20164636. lying 6G Networks. 10.1109/SCIoT56583.2022.9953671.
[11] Y. T. Aklilu et J. Ding, “Survey on Blockchain for Smart Grid Manage- [31] Shahinzadeh, Hossein and Zanjani, s. Mohammadali and Moradi, Jalal and
ment, Control, and Operation“, Energies, vol. 15, no 1, Art. no 1, janv. Fayaz-Dastgerdi, Mohammad-Hossein and Yaïci, Wahiba and Benbouzid,
2022, doi: 10.3390/en15010193. Mohamed. (2022). The Transition Toward Merging Big Data Analytics,
[12] P. Zheng, Z. Zheng, J. Wu, et H.-N. Dai, “XBlock-ETH: Extracting and IoT, and Artificial Intelligence with Blockchain in Transactive Energy
Exploring Blockchain Data From Ethereum“, IEEE Open J. Comput. Soc., Markets. 10.1109/GEC55014.2022.9986604.
vol. 1, p. 95-106, 2020, doi: 10.1109/OJCS.2020.2990458. [32] Mazhar, Tehseen and Irfan, Hafiz and Haq, Inayatul and Ullah, Inam and
[13] W. T. Scherer, S. Adams, et P. A. Beling, “On the Practical Art of State Ashraf, Madiha and Al Shloul, Tamara and Ghadi, Yazeed, Imran and
Definitions for Markov Decision Process Construction“, IEEE Access, vol. Elkamchouchi, Dalia. (2023). Analysis of Challenges and Solutions of IoT
6, p. 21115-21128, 2018, doi: 10.1109/ACCESS.2018.2819940. in Smart Grids Using AI and Machine Learning Techniques: A Review.
[14] Avi Gopstein, Cuong Nguyen, Cheyney O’Fallon, David Wollman, Nelson Electronics. 12. 242. 10.3390/electronics12010242.
Hasting, “ NIST Framework and Roadmap for Smart Grid Interoperability [33] I. Hammouti, A. Addaim and Z. Guennoun, "Proposed Architecture of
Standards Release 4.0”, NIST Special Publication, 2020 Cyber Security in Smart Grids, Blockchain as Solution," 2022 IEEE

VOLUME 4, 2016 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3370239

Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

Information Technologies and Smart Industrial Systems (ITSIS), Paris, HASSAN EL GHAZI is currently an Asso-
France, 2022, pp. 1-4, doi: 10.1109/ITSIS56166.2022.10118374. ciate Professor in Communications Systems De-
[34] G. Chen, M. He, J. Gao, C. Liu, Y. Yin and Q. Li, "Blockchain-Based partment at the National Institute of Posts and
Cyber Security and Advanced Distribution in Smart Grid," 2021 IEEE Telecommunications (INPT), Morocco. He re-
4th International Conference on Electronics Technology (ICET), Chengdu, ceived his M.S. degree in Wireless Communi-
China, 2021, pp. 1077-1080, doi: 10.1109/ICET51757.2021.9451130. cations and his Ph.D degree in Electrical En-
[35] A. S. Sani, D. Yuan and Z. Y. Dong, "SDAG: Blockchain- gineering from the University Of Polytechnique
enabled Model for Secure Data Awareness in Smart Grids," 2023
Hautsde-France France, in 2004 and 2008, respec-
IEEE Power & Energy Society Innovative Smart Grid Technolo-
tively. He advised many Ph.D. and graduate stu-
gies Conference (ISGT), Washington, DC, USA, 2023, pp. 1-5, doi:
10.1109/ISGT51731.2023.10066338. dents at both INPT and Mohammed V University,
[36] S. Samy, M. Azab and M. Rizk, "Towards a Secured Blockchain-based Rabat, Morocco. So far, his research contributions have culminated in more
Smart Grid," 2021 IEEE 11th Annual Computing and Communication than 50 papers in a wide variety of international journals and conferences.
Workshop and Conference (CCWC), NV, USA, 2021, pp. 1066-1069, doi: He served as reviewer for IEEE access, Elsevier, Springer. His main research
10.1109/CCWC51732.2021.9376089. interests are related to cyber-physical security, smart grid systems, and
[37] E. Esenogho, K. Djouani and A. M. Kurien, "Integrating Artificial Intelli- cognitive radio networks. Dr. El Ghazi served as the General chair of the
gence Internet of Things and 5G for Next-Generation Smartgrid: A Survey IWTSC’18 conference and conference Chair of NISS’19 conference.
of Trends Challenges and Prospect," in IEEE Access, vol. 10, pp. 4794-
4831, 2022, doi: 10.1109/ACCESS.2022.3140595.
[38] E. Ebenezer, Theo. G. Swart, and T. Shongwe, "Leveraging on the Cog-
nitive Radio Channel Aggregation Strategy for Next Generation Utility
Networks," MDPI Energies Journal, Special Issue on Smart Energy of,
vol. 12, no.14, July 2019.
[39] Nyangaresi, V.O., Abduljabbar, Z.A., Refish, S.H.A., Al Sibahee, M.A.,
Abood, E.W., Lu, S. (2022). Anonymous Key Agreement and Mutual
Authentication Protocol for Smart Grids. In: Jin, H., Liu, C., Pathan,
AS.K., Fadlullah, Z.M., Choudhury, S. (eds) Cognitive Radio Oriented WASSIM FASSI FIHRI obtained a M.Sc. in
Wireless Networks and Wireless Internet. CROWNCOM WiCON 2021 Computer Science and Telecommunications from
2021. Lecture Notes of the Institute for Computer Sciences, Social Infor- Ibn Tofail University, Morocco, in 2008 and a
matics and Telecommunications Engineering, vol 427. Springer, Cham. Ph.D. in Computer Science and Telecommuni-
[40] Nyangaresi, V.O., Abduljabbar, Z.A., Al Sibahee, M.A., Abood, E.W., cations from the Institut National des Postes et
Abduljaleel, I.Q. (2022). Dynamic Ephemeral and Session Key Generation Télécommunications (INPT), Morocco, in 2021.
Protocol for Next Generation Smart Grids. In: Bao, W., Yuan, X., Gao, He is also a cybersecurity architect and a certi-
L., Luan, T.H., Choi, D.B.J. (eds) Ad Hoc Networks and Tools for IT. fied Project Manager Professional (PMP). His re-
ADHOCNETS TridentCom 2021 2021. Lecture Notes of the Institute search interests include cybersecurity, networking,
for Computer Sciences, Social Informatics and Telecommunications En- cognitive radio networks, machine learning, and
gineering, vol 428. Springer, Cham. blockchain. He is the author of more than six publications in renowned
[41] Aziz, I.T.; Jin, H.; Abdulqadder, I.H.; Hussien, Z.A.; Abduljabbar, conferences and journals, such as IEEE Access, Springer and Wiley.
Z.A.; Flaih, F.M.F. A Lightweight Scheme to Authenticate and Se-
cure the Communication in Smart Grids. Appl. Sci. 2018, 8, 1508.
https://doi.org/10.3390/app8091508
[42] Hussien, Z.A.; Abdulmalik, H.A.; Hussain, M.A.; Nyangaresi, V.O.; Ma,
J.; Abduljabbar, Z.A.; Abduljaleel, I.Q. Lightweight Integrity Preserving
Scheme for Secure Data Exchange in Cloud-Based IoT Systems. Appl.
Sci. 2023, 13, 691. https://doi.org/10.3390/app13020691
[43] M. A. Al Sibahee et al., "Lightweight Secure Message Delivery for E2E
S2S Communication in the IoT-Cloud System," in IEEE Access, vol. 8,
pp. 218331-218347, 2020, doi: 10.1109/ACCESS.2020.3041809. OMAR AIT OUALHAJ is a Professor Assistant
[44] S. M. Umran, S. Lu, Z. Ameen Abduljabbar and X. Tang, "A blockchain- at the National Institute of Posts and Telecom-
based architecture for securing industrial iots data in electric smart grid," munications (INPT) in Rabat. He holds a Ph.D.
Computers, Materials and Continua, vol. 74, no.3, pp. 5389–5416, 2023. degree in Computer Science from the National
School of Computer Science and Systems Anal-
ysis (ENSIAS), Mohammed V University of Ra-
bat, Morocco, which he obtained in March 2019.
In 2012, he received his research MS degree in
Computer Science and Telecommunication from
the Faculty of Sciences, Mohammed V University
of Rabat. Ait oualhaj’s research activities mainly focus on the Internet of
NABIL TAZI CHIBI was born in Casablanca Things (IoT), sensor networks, ad-hoc networks, Delay Tolerant Networks
Morocco in 1977, he is a Ph.D. student in the (DTNs), vehicular networks (routing and energy efficiency), mobility, and
Communications Systems Department at the Na- the performance evaluation of mobile networks using game theory and MDP,
tional Institute of Posts and Telecommunications learning approach, and machine learning techniques, he has served as a
(INPT), Morocco. He is currently a Director of reviewer for numerous renowned international journals and conferences,
the information systems. He received his engineer- such as TNSESI, IEEE GLOBECOM, IEEE ICC, IEEE WCNC, ICT,
ing degree in Computer Science and telecommu- IWCMC, amongst others. He is actively involved in research and continues
nication in 2002, National School of Computer to contribute to diverse scientific communities.
Science and Systems Analysis (ENSIAS), Rabat,
Morocco. His research interests include Machine
Learning, Blockchain, cybersecurity and smart grid.

10 VOLUME 4, 2016

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4