Content uploaded by Deepika Agrawal
Author content
All content in this area was uploaded by Deepika Agrawal on Mar 16, 2020
Content may be subject to copyright.
RESEARCH ARTICLE
GWO-C: Grey wolf optimizer-based clustering scheme for
WSNs
Deepika Agrawal | Muhammad Huzaif Wasim Qureshi | Pooja Pincha |
Prateet Srivastava | Sourabh Agarwal | Vikram Tiwari | Sudhakar Pandey
Department of Information Technology,
National Institute of Technology, Raipur,
Raipur, India
Correspondence
Deepika Agrawal, Department of
Information Technology, National
Institute of Technology, Raipur, Raipur,
India.
Email: deepika721@gmail.com
Summary
A wireless sensor network (WSN) is a prominent technology that could assist
in the fourth industrial revolution. Sensor nodes present in the WSNs are func-
tioned by a battery. It is impossible to recharge or replace the battery, hence
energy is the most important resource of WSNs. Many techniques have been
devised and used over the years to conserve this scarce resource of WSNs.
Clustering has turned out to be one of the most efficient methods for this pur-
pose. This paper intends to propose an efficient technique for election of clus-
ter heads in WSNs to increase the network lifespan. For the achievement of
this task, grey wolf optimizer (GWO) has been employed. In this paper, the
general GWO has been modified to cater to the specific purpose of cluster head
selection in WSNs. The objective function for the proposed formulation con-
siders average intra-cluster distance, sink distance, residual energy, and CH
balancing factor. The simulations are carried out in diverse conditions. On
comparison of the proposed protocol, ie, GWO-C protocol with some well-
known clustering protocols, the obtained results prove that the proposed proto-
col outperforms with respect to the consumption of the energy, throughput,
and the lifespan of the network. The proposed protocol forms energy-efficient
and scalable clusters.
KEYWORDS
clustering, cluster head, grey wolf optimization, optimization, wireless sensor networks
1|INTRODUCTION
Wireless sensor networks (WSNs) are proving out to be one of the most promising mechanisms for a reception,
processing, and transmitting data from distant environments to the central data processing stations (base station [BS]).
1
They find their usage in a variety of tasks including broadcasting, routing, and forwarding. A typical WSN consists of
thousands of tiny and autonomous sensor nodes with limited battery capacity. These sensor nodes are generally
deployed in remote locations and therefore batteries cannot be replaced easily. Thus, it becomes an important issue to
replenish their energy. Before going for the replenishment of their energy, it is more important to sustain the energy of
the nodes. This is where energy conservation in WSNs comes into play. In order to sustain a WSN, it is important to
save on energy losses. Clustering provides an efficient and sufficiently simple method to realize this.
Received: 18 August 2018 Revised: 4 October 2019 Accepted: 8 January 2020
DOI: 10.1002/dac.4344
Int J Commun Syst. 2020;e4344. wileyonlinelibrary.com/journal/dac © 2020 John Wiley & Sons, Ltd. 1of15
https://doi.org/10.1002/dac.4344
Clustering in its essence is essentially the grouping of several nodes of a WSN and establish communication
between the group and the BS through the group leader known as cluster head (CH).
2,3
The selection of clusters and
corresponding CHs is a challenging and daunting task on its own. Various techniques have been applied for years to
achieve an optimal selection of CHs.
Over the years, many optimization techniques
4
have been applied to find the excellent group of CHs in the network.
These techniques were fuzzy logic,
5
ant colony optimization,
6
genetic algorithms,
7
particle swarm intelligence,
8
har-
mony search algorithm,
9
etc.
In this paper, the grey wolf optimization (GWO) technique
10
is used to solve the CH selection problem. This is done
by simulating the grey wolves to be the solutions themselves. The alphas being the best solutions and the rest of the
subsequent grades of solutions.
The remaining of the paper is arranged as follows:
Section 2 gives an outline of the related work done so far in the area of clustering. Section 3 presents a thorough
description of the GWO technique. The energy model and the assumptions made about the network is depicted in
Section 4. Section 5 deals with the explanation of the proposed protocol in detail. Section 6 analyzes the proposed proto-
col. Section 7 presents the experimental results accompanied by the conclusion.
2|RELATED WORK
A well-known and standard clustering protocol is proposed, ie, low-energy adaptive clustering hierarchy (LEACH). This
probabilistic model is used in this protocol. Each sensor node present in the network picks an arbitrary number
between 0 to 1 and then the comparison is done with the predefined threshold. If the chosen number is less than the
predefined threshold, then the sensor node becomes a CH. Non-CH nodes will then join the nearest CH. A TDMA
schedule is generated by the CH to schedule the transmission time of its cluster members. It has several disadvantages,
and performance is poor. Also, it may possible that the node having less energy will become CH.
11
To resolve the issue
of LEACH protocol, several other protocols were proposed. They are low-energy adaptive clustering hierarchy central-
ized (LEACH-centralized),
12
stable election protocol (SEP),
13
and energy-efficient heterogeneous clustering (EEHC).
14
LEACH-centralized takes residual energy into consideration while selecting CHs. While SEP and EEHC protocols are
based on heterogeneous networks. Some nodes in the network are made advanced nodes having higher energy than
other nodes.
The disadvantages of these protocols have led to the development of many successive clustering protocols. They are
hybrid energy-efficient distributed clustering (HEED),
15
power-efficient gathering in sensor information systems,
16
dis-
tributed energy-efficient clustering (DEEC),
17
and threshold sensitive energy-efficient sensor network.
18
HEED protocol is unsuitable for heterogeneous networks, hence to resolve this issue, a new protocol is
implemented, namely, hybrid distributed energy-efficient heterogeneous clustering protocol (HDEEHC). Three parame-
ters are taken into consideration for CH election, namely, residual energy, node degree, and the type of node. It has bet-
ter performance in case of the heterogeneous environment.
19
Balancing the energy consumption in the network is a crucial issue. So to resolve this issue, several protocols are
developed. Energy-efficient layered clustering (EEHC) is a protocol in which clustering is fixed. The residual energy of
nodes elects the CHs. It prolongs the network lifespan.
20
Several optimization techniques are also used to resolve the issue of WSN. Particle swarm optimization (PSO) is one
of them which has the advantages of providing a better solution with higher efficiency in computation. A PSO-
clustering is proposed that chooses the sensor node as CH which is nearer to the BS in that cluster.
21
PSO-C is another protocol that considers residual energy and distance between nodes to elect the CHs.
22
PSO
shows better performance with comparison to LEACH-C and LEACH with respect to throughput and network life-
time.
23
Another protocol is proposed based on graph theory and PSO. A weighted function is computed based on
the distance and remaining energy.
24
PSO can also be applied in maximizing the coverage in mobile WSNs. Sensor
nodes are deployed using PSO.
25
Intra-cluster distance acts as an important criterion in the consumption of the net-
work's energy. A PSO-based CH selection technique is used to select the best CHs that has less intracluster
distance.
26
Another technique is proposed based on novel chemical reaction optimization (nCRO).
27
It prolongs the network's
lifespan. Another protocol is proposed by the same author based on PSO for applications that are time-sensitive. It takes
energy into consideration and extends the lifespan of the network.
28
A protocol is proposed known as PSO-ECHS which
2of15 AGRAWAL ET AL.
considers sink distance, intracluster distance, and remaining energy. PSO is applied. However, it doesn't balance the
load by forming equal clusters, hence suffers from large energy consumption.
29
A new technique known as social group optimization (SGO) is also proposed in WSNs.
30
This technique decreases
the distance to transmit the data and also permit the nodes to spend less energy.
A protocol based on GWO is proposed known as fitness value-based improved grey wolf optimization
(FIGWO).
31
A fitness function is calculated to select the CHs. This fitness function ensures that the node having the
highest energy as well as the node located near the BS have higher chance of selection as CH. Also, the distance to
transmit data is also recalculated whenever a new CH is elected. However, it suffers from balancing the load among
the CHs.
A routing protocol is proposed for the selection of optimal routes with the help of Tabu PSO is proposed.
32
The pro-
tocol enhances the number of clusters formed, percentage of nodes alive and shows the reduction of average packet loss
rate and average end-to-end delay. A hierarchical clustering method using type 2 fuzzy logic is proposed for mobile
WSNs.
33
This enhances the network lifetime and minimizes the packet loss. A genetic algorithm-based optimized clus-
tering (GAOC) protocol is proposed for heterogeneous WSN.
34
Multiple data sinks are used.
The proposed protocol in this paper uses GWO for the selection of CH. The reason for choosing GWO over other
meta-heuristic techniques is that the GWO has a faster convergence rate. Moreover, GWO leads to the continuous
reduction of search space as well as decision variables are less. It also avoids local optima.
3|GREY WOLF OPTIMIZER
The GWO is a new meta-heuristic algorithm to solve many problems of optimization. The GWO draws its inspiration
from the naturally existent hierarchy of leadership in the hunting mechanism of the pack of grey wolves. It is a rela-
tively new method for optimization and also the least researched algorithm.
10
The algorithm is similar to genetic algo-
rithms when it comes to implementation and usage. The mathematical equations for this algorithm are derived from
the observed patterns of the hunting mechanisms of the pack.
10
The equations are then conformed to the problem at
hand to find a feasible optimized solution.
The candidate solutions can also be classified on the basis of the hierarchical structure of the social dominance of
the grey wolves. As such, the best and most optimized solution is described as alpha (α) with beta (β) and delta (δ) for-
ming the second- and third-best solutions, respectively.
10
All the solutions apart from these are taken to be omega (Ω)
solutions and are the least fit solutions. The process of optimization is based on the process of hunting and it is guided
by the α,β,δ, and Ωparameters.
During hunting, the grey wolves enclose in and surround the prey.
10
This enclosing behavior of the wolves can be
depicted mathematically as follows:
D
!=C
!
:Xp
!tðÞ−X
!tðÞ
,ð1Þ
X
!t+1ðÞ=Xp
!tðÞ−A
!
:D
!,ð2Þ
where trepresents current iteration, A
!and C
!are coefficient-vectors, Xp
!is the vector's position of the prey, and X
!is the
vector's position of a grey wolf.
The vectors A
!and C
!can be computed as follows:
A
!=2a
!
:r1
!−a
!,ð3Þ
C
!=2:r2
!,ð4Þ
where r1
!and r2
!are random vectors in [0,1] and the components of a
!are decreased linearly from 2 to 0 over repeated
iterations.
In this process, it is assumed that the location of the prey is unknown. The hunting process is lead by the best candi-
date solutions alpha and beta and the least important members, ie, omegas update their positions according to the
AGRAWAL ET AL.3of15
information made available by the best search agents viz. alphas and betas.
10
The mathematical equations for this pur-
pose are modeled as follows:
Dα
!=C1
!
:Xα
!−X
!
,ð5Þ
Dβ
!=C2
!
:Xβ
!−X
!
,ð6Þ
Dδ
!=C3
!
:Xδ
!−X
!
,ð7Þ
X1
!=Xα
!−A1
!
:Dα
!
,ð8Þ
X2
!=Xβ
!−A2
!
:Dβ
!
,ð9Þ
X3
!=Xδ
!−A3
!
:Dδ
!
,ð10Þ
X
!t+1ðÞ=X1
!+X2
!+X3
!
3,ð11Þ
The final step in the process of hunting is the attack. The process of attacking can be mathematically defined using
the operators stated above. This is done by decreasing the value of d
!and also reducing the range of variation of A
!in
the range of [−2a, 2a] while a is reduced from 2 to 0 over the iterations. If the values of A
!lie in [−1,1], the position of
the search agent will be between the current position and the prey's position. If|A|< 1, the wolves attack the prey. Thus,
it can be seen that according to the GWO algorithm, the search agents update their positions according to the positions
of the alpha, beta, and delta members.
The search for prey starts when the wolves diverge from each other to find the prey. This search is also dependent
on the positions of the alpha, beta, and delta members. The conditions for search or attack are dictated by the values of
A
!as follows:
•|A|>1=> diverge and search
•|A|<1=> converge and attack
4|SYSTEM MODEL AND ASSUMPTIONS
4.1 |Energy model
In this model, when the threshold distance (d
0
) is greater than the propagation distance (d) then the consumption of
energy of a node is directly proportional to d
2
. The overall consumption of energy of each node to transmit the l-bit
packet of data is specified by the following equations:
ETX l,dðÞ=l×Eelec +l×εfs ×d2,if d <do
l×Eelec +l×εmp ×d4,if d≥do
()
,ð12Þ
where E
TX
is the sum of overall energy required to transmit, E
elec
is the dissipation of energy per bit to run the
circuit, ie, transmitter or receiver, ε
fs
is the energy used for amplification in free space model and ε
mp
in multi-
path model and it is highly dependent on the transmitter amplifier model, and d
0
is the threshold transmission
distance.
4of15 AGRAWAL ET AL.
In the same way, to receive l-bit of data, the consumption of energy by the receiver circuit is given by
ERX lðÞ=l*Eelec,ð13Þ
where E
RX
is the consumption of the energy required to receive data, E
elec
is the dissipation of energy per bit to run the
circuit, ie, transmitter or receiver, and depends on several factors such as modulation, digital coding, signal spreading,
and filtering. In general, propagation of the radio wave is highly variable and is very complex to model.
Etotal =E
TX +E
RX,ð14Þ
where E
total
is the total energy loss for the system.
4.2 |Assumptions
The following are the list of the assumptions that are made about the network:
1. Once deployed, all nodes are stationary.
2. The nodes are homogeneous in nature.
3. The BS is also stationary and having sufficient energy.
4. The nodes are left unattended after deployment.
5|PROPOSED PROTOCOL
In this section, the detailed description of the proposed protocol is given. The process of clustering is split into two seg-
ments: the first segment is CHs selection phase and the second segment is cluster formation phase.
5.1 |Cluster selection
The election of CHs in a network is done by the GWO technique. Generally, the CHs are selected based on the sensor
network's parameters like distance, residual energy, and degree. In our proposed protocol, ie, GWO-C, a new parameter
is included to elect the CHs, ie, CH balancing factor. After the initialization of the network, all the sensor nodes trans-
mit their exact location and remaining energy to the BS. The BS runs the GWO-based clustering protocol. A fitness
function is nascent grounded on various parameters like residual energy, sink distance, intracluster distance, and CH
balancing factor.
The main intention of the GWO-C is to elect the CHs in the network so that the lifespan of the network is pro-
longed. To efficiently elect the CHs, four parameters are considered. They are the average intracluster distance, the dis-
tance of the CHs from the sink, the average remaining energy, and CH balancing factor. The following
section describes the terminologies used in the proposed protocol and after that the next section described the fitness
function used in the proposed protocol.
5.1.1 |Terminology
To implement the GWO-C protocol, the following terminologies are used:
1. n: Total number of alive sensor nodes
2. m: Total number of CHs
3. S: The collection of all sensor nodes, ie, S = {s
1,
s
2
,s
3
,…..,s
n
}
4. C: The collection of all CHs i.e., C = {CH
1,
CH
2….,
CH
m
}
5. l
j
: The numbers of sensor nodes in the cluster j
AGRAWAL ET AL.5of15
6. D
max
: Sensor node's range.
7. R
max
: The maximum communication range of the CH
8. T
H
: The threshold energy for being a CH
9. D
0
: The threshold distance
10. E
sk
: Sensor node's s
k
initial energy, 1≤k≤n
11. E
CHj
: CH's current energy CH
j
,1≤j≤m
12. Comm (s
k
): All the nodes which are which the communication range
13. Dis (s
k
,s
j
): The distance between two sensor nodes s
k
and s
j
14. Dis (s
k
,CH
j
): The distance between sensor node s
k
and CH CH
j
15. Dis (CH
j
, BS): The distance between CH CH
j
and BS
5.1.2 |Nascent of the fitness function
The following are the parameters responsible for the derivation of fitness function:
Average intra-cluster distance (F
1
)
The distance between all the sensor nodes with their respective CH is calculated and the sum of all these is known as
intra-cluster distance. To minimize energy consumption in the network, this intra-cluster distance needs to be mini-
mized. Since sensor nodes consume some energy during the communication to their respective CH, it is given as
F1 = Xm
j=1
1
ljXlj
k=1dis sk,CHjðÞ
:ð15Þ
Average sink distance (F
2
)
The average of the sink distance is calculated as the ratio of the distance between the BS and the CH to the total sensor
nodes present in the respective CH. This parameter is taken into consideration because distance plays an important role
in energy consumption. Hence, there is a need to minimize this distance to reduce the consumption of energy.
It is given as
F2 = Xm
j=1
1
lj
dis CHj,BS
:ð16Þ
Residual energy (F
3
)
Since the lifespan of the network depends on the utilization of the energy, hence it is highly needed to minimize energy
consumption. Hence, this parameter takes into account. It is calculated as the total current energy of all the elected
CHs. Total energy needs to be maximized; hence, inverse of this is taken into consideration to balance each objective
function.
F3 = 1
Pm
j=1 ECHj
:ð17Þ
CH balancing factor (F
4
)
There is a need to balance the cluster. Due to the random organization of sensor nodes, there is a chance that some big
clusters are formed and some small clusters. Hence, this parameter is taken into consideration to balance the consump-
tion of energy.
F4 = Xm
j=1
n
m−lj:ð18Þ
Instead of separately minimizing each fitness function, it is better to minimize the combination of the above fitness
function as shown in Equation (19). The above fitness functions are strongly in harmony of each other.
6of15 AGRAWAL ET AL.
The following fitness function is used:
Fitness function = a*f1+b*f2+c*f3+1−a+b+cðÞ*f4,ð19Þ
where a,b, and crepresent constant value and a + b + c = 1
5.1.3 |Algorithm
In the GWO technique, a pack signifies the number of CHs. The flowchart of the proposed algorithm is given as
follows:
For the selection of CHs, the first 10% of the alive sensor nodes whose residual energy is greater than the average
residual energy are selected as a CH for each pack. Then, fitness function value is calculated for each CH present in a
pack. The node whose fitness function value is less in a pack then that node is selected as a new and final CH.
After the application of GWO, nodes in a pack are the final CHs as explained in the flowchart of the proposed algo-
rithm depicted in Figure 1. After this phase, a cluster is formed as explained in next section.
The algorithm for the proposed protocol is given as follows:
Input: Set of alive sensor nodes in a round
Number of Packs N
p
Number of cluster heads in a pack: 10% of the total number of alive sensor nodes whose residual energy is greater
than the average residual energy
Output: Optimal choice of cluster heads CH= {CH
1
,CH
2,
CH
3,……,
CH
m
}
a. Step 1: Initialize packs P
i
,choose 10% sensor nodes as cluster heads in a pack from locAlive whose residual
energy is greater than the average residual energy. /*The selection of specific sensor nodes as cluster heads */
b. Step 2: for i= 1 to N
p
do
1. Calculate Fitness (P
i
)/
*Using equation 19*/
FIGURE 1 Flowchart of the proposed
algorithm
AGRAWAL ET AL.7of15
2. Pbest
i
=P
i
Endfor
c. Step 3: Gbest = {Pbesti|Fitness (Pbest
i
) = min (Fitness (Pbest
i
), i,1<i≤N
p
)}
i. Z=P with the minimum Fitness value
d. Step 4: for t=0toT
R
/*T
R
= Maximum number of iterations */
i. fori= 1 to N
p
do
ii. for g = 1 to the size of N
p
do
1. Calculate Fitness (P
i
)/
*Using equation 19*/
2. Select the leader node α,β, and δaccording to the best three fitness value
3. Update the position of the prey using eqns 8, 9 and 10
4. Take an average of the 3 solutions and update position for the cluster head
5. Find the nearest sensor node to be selected as new cluster head
iii. Endfor
iv. if Fitness (P
i
)<Fitness (Pbest
j
)then
1. Pbest
i
=P
i
v. Endif
vi.if Fitness (P
i
)<Fitness (Gbest)then
1. Gbest = Pi
2. z = i
vii. Endif
viii. End for
ix. Endfor
e. Step 5: CH = Set of sensor nodes in P
i
f. Step 6: Stop
5.2 |Cluster formation
Once the election of CHs is done, non-CH nodes join the nearest CHs. Non-CH nodes transmit a message to request to
the CHs to join the cluster. CHs then send accept message to the non-CH nodes. Nodes having minimum distance will
join the CH. In this way, cluster formation is done.
6|PROTOCOL ANALYSIS
6.1 |Convergence rate
The number of iterations considered by the evolutionary optimization technique in a given maximum number of itera-
tions to obtain the best solution is defined as the convergence rate. From Table 1, it is clear that the proposed protocol
converges in a minimum number of iterations to obtain the best solution as compared to the other clustering protocols
used for comparison. The convergence rate of the proposed protocol is less because GWO leads to the continuous reduc-
tion of search space as well as decision variables are less. It also avoids local optima.
TABLE 1 Convergence rate analysis
Technique
Convergence Rate Analysis
Mean Standard Deviation
GWO-C 80.34 25.65
FIGWO 134.66 34.98
PSO-ECHS 185.45 185.45
Bolded data gives the convergence rate analysis of all algorithms.
8of15 AGRAWAL ET AL.
6.2 |Computational complexity
The main contribution in calculating computational complexity of GWO-C is of initialization and iterations. From
Algorithm 1, in step 1, initialization is done. So, the iteration complexity is O (N
p
). Now, one for loop started at step
2 and executed individually and is up to the number of packs in the worst case. Hence, the complexity is O (N
p
). In step
4, there are three for loops. In the worst case, outer for loop executed till the maximum number of iterations reached.
First, for loop started at line number d(i) executed till number of pack, ie, N
p
in the worst case. Second, for loop started
at line number d (ii) executed till size of pack, ie, m. So, the complexity for step 4 is O (T
r
N
p
m) in the worst case. Hence,
the overall complexity for the proposed algorithm is O (N
p
)+O(N
p
)+O(T
r
N
p
m).
6.3 |Scalability
To prove that the proposed protocol is scalable, simulations are carried out in a large environment with a network size
of 500 *500 and 1000 *1000 with 500 nodes. In both the simulation setups, the proposed protocol outperforms with its
counterparts as shown in Table 2.
6.4 |Comparison of clustering algorithms
Table 3 presents a detailed description of the comparison made between several clustering-based algorithms and pro-
posed GWO-C algorithm. The proposed algorithm uses four parameters in the objective function. The objective function
is designed in such a way that it increases the network lifetime by balancing the load among the nodes.
7|RESULTS
The GWO-C was implemented in MATLAB. The protocol was tested in three different scenarios. These three different
scenarios are created in which the BS's position is changed to study the effect of BS position.
In scenario 1, the BS's position is taken at the center of the region of the interest (ROI).
In scenario 2, the BS's position is taken at the corner of the ROI.
In scenario 3, the BS's position is taken outside the ROI.
The comparison here is done for 100 wireless sensor nodes which are present in a “100 ×100”unit square area in
each scenario. In every round for each scenario, approximately 10% of the wireless sensor nodes were elected to act as
the CH for a cluster.
The proposed protocol is compared with PSO-based clustering protocol
29
and fitness value-based improved GWO
(FIGWO).
31
The results are compared on the basis of number of dead nodes, throughput, and the remaining residual
energy of the nodes. Table 4 shows the simulation parameters used for the proposed protocol.
7.1 |Evaluation of the proposed protocol on the basis of a number of dead nodes
The proposed protocol is first evaluated on the basis of network lifetime, ie, the number of dead nodes in a particular
round. The results are shown in Figures 2–4. The network lifetime is one of the important measures in WSNs.
TABLE 2 First node death round
Protocol
First Node Death Round
500 * 500 1000 * 1000
PSO-ECHS 2 1
FIGWO 3 2
GWO-C 5 4
AGRAWAL ET AL.9of15
It can be seen for scenario 1, from Figure 2, that the first node death round in PSO-ECHS protocol is 523, in FIGWO
it is 697, while the first node death round in proposed protocol is 756.
For scenario 2 when BS position is at the corner of the region of interest, the first node in PSO-ECHS protocol gets
dead around round number 376, in FIGWO, it gets dead around 595th round, and in case of GWO-C, it gets dead
around round number 795.
TABLE 3 Comparison of several cluster based algorithms and proposed algorithm
Protocol Type of Node
Inter-cluster
Topology
Energy
Awareness
CH
Selection
Using
Heuristic
Algorithm
Number of
Parameters in
Objective
Function
Load in
Terms of
Number of
Sensor Nodes
LEACH
11
Homogeneous Direct Not required Not
required
No Not applicable No
LEACH-C
12
Homogeneous Direct Not required Not
required
No Not applicable No
SEP
13
Heterogeneous Direct Not required Not
required
No Not applicable No
EEHC
14
Heterogeneous Direct Not required Not
required
No Not applicable No
HEED
15
Heterogeneous Direct Not required Required No Not applicable No
PSO-C
22
Homogeneous Direct Required Required Yes Two No
PSO-ECHS
29
Homogeneous Direct Required Required Yes Three No
SGO
30
Homogeneous Direct Required Required Yes Two No
FIGWO
31
Homogeneous Direct Required Required Yes Two No
GWO-C
(Proposed)
Homogeneous Direct Required Required Yes Four No
TABLE 4 Parameters for simulation
Parameter Value
Target area 100 ×100 m
2
Number of sensor nodes 100
Initial energy of sensor nodes 0.5 J
Base station locations Center, corner, outside
Percentage of CHs 10%
FIGURE 2 Comparison of dead nodes when BS is at the center
of the ROI
10 of 15 AGRAWAL ET AL.
For scenario 3, the first node in PSO-ECHS gets dead around round number 64, and in FIGWO, it is 175, while in
case of GWO-C protocol, the first node gets dead around round number 454.
It is clear from the Table 5 that the proposed protocol, ie, GWO-C, has increased the lifetime of the network by
30.08% and 7.8% in comparison to the PSO-ECHS and FIGWO, respectively, when BS is at the center of ROI. When BS
is at the corner of ROI, the GWO-C protocol prolongs the lifetime of the network by 25.15% and 52.7% in comparison to
the PSO-ECHS and FIGWO, respectively, and while when BS is outside the range of ROI, the GWO-C protocol prolongs
the lifetime of the network by 61.45% and 85.9% in comparison to the PSO-ECHS and FIGWO, respectively.
The proposed protocol, ie, GWO-C protocol performs better as it considers residual energy, CH balancing factor dur-
ing the election of CHs. The CH is heavily loaded and consumes more energy than other sensor nodes. Hence, it is to
be properly balanced. The GWO-C protocol effectively balances the load present in the network. The GWO-C protocol
has more alive nodes than other algorithms used for comparison. It also specifies that GWO-C prolongs the lifetime of
the network and ensures better transmission of observed data.
FIGURE 3 Comparison of dead nodes when BS is at the corner
of the ROI
FIGURE 4 Comparison of dead nodes when BS is at the outside of the ROI
TABLE 5 First node death round for all protocols
Protocols
First Node Dead Round
When BS is at the Center of ROI When BS is at the Corner of ROI When BS is at the Outside of ROI
PSO-ECHS 523 376 64
FIGWO 697 595 175
GWO-C 756 795 454
AGRAWAL ET AL.11 of 15
7.2 |Evaluation of the proposed protocol with respect to remaining residual energy
Next, the protocol is evaluated for the remaining residual energy. The remaining residual energy is also high in pro-
posed protocol as compared to the PSO-based clustering.
Since the CH is elected based on different parameters like residual energy, sink distance, intracluster distance, and
CH balancing factor, the distance acts as an important role in the consumption of energy. Hence, it takes in to account
FIGURE 6 Comparison of residual energy when BS is at the corner of the ROI
FIGURE 7 Comparison of residual energy when BS is at the outside of the ROI
FIGURE 5 Comparison of residual energy when BS is at the center of the ROI
12 of 15 AGRAWAL ET AL.
at the time of CH election in the proposed protocol. Moreover, a new parameter is introduced, ie, CH balancing factor,
which balances the load among the CHs by creating equal clusters. Hence, the remaining energy in GWO-C is high as
compared to the PSO-ECHS and FIGWO, as shown in Figures 5–7. GWO-C has high total residual energy in each
round as compared to other protocols. Since the GWO-C protocol balances the load among the network, no sensor node
in the network dies too early due to the excessive consumption of the energy.
7.3 |Evaluation of the proposed protocol on the basis of throughput
After that, the proposed protocol is evaluated for throughput, ie, the number of data packets received by the BS. The
number of data packets received is highly dependent on the network lifespan and the remaining residual energy. Since
lifespan of the network and remaining residual energy is high in GWO based clustering, hence it received more number
of packets, as shown in Figure 8. A proper fitness function is designed to elect the CHs; hence, GWO-based clustering
performs better.
8|CONCLUSION
The GWO is a relatively new technique with a vast array of options available for its enhancement. In this paper, it is
used to elect the CHs. A proper fitness function is designed which takes important parameters of the network into con-
sideration. The results were compared for the three standard scenarios of the situation of the BS, and the GWO was
found to deliver consistently better performance compared to the PSO-ECHS and FIGWO. The results were tested
against three standard metrics of throughput, residual energy, and first dead node. The GWO was able to deliver better
results in all the scenarios and in all the metrics. The proposed work was idealized, implemented, and tested for the
WSNs with static sensor nodes. The work can further be expanded to the networks with mobile sensor nodes, ie, the
sensors capable of changing their real-time positions.
ORCID
Deepika Agrawal https://orcid.org/0000-0003-0252-1777
REFERENCES
1. Akyildiz IF, Su W, Sankarasubramaniam Y, Cayirci E. Wireless sensor networks: a survey. Computer networks. 2002;38(4):393-422.
2. Younis O, Krunz M, Ramasubramanian S. Node clustering in wireless sensor networks: recent developments and deployment chal-
lenges. IEEE network. 2006;20(3):20-25.
FIGURE 8 Comparison in terms of packets received by the base
station
AGRAWAL ET AL.13 of 15
3. Abbasi AA, Younis M. A survey on clustering algorithms for wireless sensor networks. Computer communications. 2007;30(14-15):2826-
2841.
4. Kulkarni RV, Forster A, Venayagamoorthy GK. Computational intelligence in wireless sensor networks: A survey. IEEE communications
surveys & tutorials. 2011;13(1):68-96.
5. Agrawal D, Pandey S. FUCA: fuzzy-based unequal clustering algorithm to prolong the lifetime of wireless sensor networks. International
Journal of Communication Systems. 2018;31(2):e3448.
6. Ye Z, Mohamadian H. Adaptive clustering based dynamic routing of wireless sensor networks via generalized ant colony optimization.
Ieri Procedia. 2014;10:2-10.
7. Bara'a AA, Khalil EA. A new evolutionary based routing protocol for clustered heterogeneous wireless sensor networks. Appl Soft Com-
put. 2012;12(7):1950-1957.
8. Singh B, Lobiyal DK. A novel energy-aware cluster head selection based on particle swarm optimization for wireless sensor networks.
Human-Centric Computing and Information Sciences. 2012;2(1):13.
9. Agrawal, D. and Pandey, S., 2017, November. FLIHSBC: fuzzy logic and improved harmony search based clustering algorithm for wire-
less sensor networks to prolong the network lifetime. In International Conference on Ubiquitous Computing and Ambient Intelligence
(pp. 570-578). Springer, Cham.
10. Mirjalili S, Mirjalili SM, Lewis A. Grey wolf optimizer. Advances in engineering software. 2014;69:46-61.
11. Heinzelman, W.R., Chandrakasan, A. and Balakrishnan, H., 2000, January. Energy-efficient communication protocol for wireless micro-
sensor networks. In System sciences, 2000. Proceedings of the 33rd annual Hawaii international conference on (pp. 10-pp). IEEE.
12. Heinzelman WB, Chandrakasan AP, Balakrishnan H. An application-specific protocol architecture for wireless microsensor networks.
IEEE Transactions on wireless communications. 2002;1(4):660-670.
13. Smaragdakis G, Matta I, Bestavros A. SEP: A stable election protocol for clustered heterogeneous wireless sensor networks. Boston Uni-
versity Computer Science Department; 2004.
14. Kumar D, Aseri TC, Patel RB. EEHC: energy efficient heterogeneous clustered scheme for wireless sensor networks. computer communi-
cations. 2009;32(4):662-667.
15. Younis O, Fahmy S. HEED: a hybrid, energy-efficient, distributed clustering approach for ad hoc sensor networks. IEEE Transactions on
mobile computing. 2004;3(4):366-379.
16. Lindsey S, Raghavendra C, Sivalingam KM. Data gathering algorithms in sensor networks using energy metrics. IEEE Transactions on
parallel and distributed systems. 2002;13(9):924-935.
17. Qing L, Zhu Q, Wang M. Design of a distributed energy-efficient clustering algorithm for heterogeneous wireless sensor networks. Com-
puter communications. 2006;29(12):2230-2237.
18. Manjeshwar, A. and Agrawal, D.P., 2001, April. TEEN: a routing protocol for enhanced efficiency in wireless sensor networks. In null
(p. 30189a). IEEE.
19. Wang J, Zhu X, Cheng Y, Zhu Y. A distributed, hybrid energy-efficient clustering protocol for heterogeneous wireless sensor network.
International Journal of Grid and Distributed Computing. 2013;6(4):39-50.
20. Gu, Y., Jing, D. and Guo, J., 2012, December. Energy efficient layered clustering approach for WSN. In Control Engineering and Commu-
nication Technology (ICCECT), 2012 International Conference on (pp. 628-631). IEEE.
21. Kulkarni RV, Venayagamoorthy GK. Particle swarm optimization in wireless-sensor networks: A brief survey. IEEE Transactions on
Systems, Man, and Cybernetics, Part C (Applications and Reviews). 2011;41(2):262-267.
22. Guru, S.M., Halgamuge, S.K. and Fernando, S., 2005, December. Particle swarm optimisers for cluster formation in wireless sensor
networks. In Intelligent Sensors, Sensor Networks and Information Processing Conference, 2005. Proceedings of the 2005 International
Conference on (pp. 319-324). IEEE.
23. Latiff, N.A., Tsimenidis, C.C. and Sharif, B.S., 2007, September. Energy-aware clustering for wireless sensor networks using particle
swarm optimization. In Personal, Indoor and Mobile Radio Communications, 2007. PIMRC 2007. IEEE 18th International Symposium on
(pp. 1-5). IEEE.
24. Ab Aziz, N.A.B., Mohemmed, A.W. and Sagar, B.D., 2007, November. Particle swarm optimization and Voronoi diagram for wireless
sensor networks coverage optimization. In Intelligent and Advanced Systems, 2007. ICIAS 2007. International Conference on
(pp. 961-965). IEEE.
25. Gopakumar, A. and Jacob, L., 2008. Localization in wireless sensor networks using particle swarm optimization.
26. Saravanan, M. and Madheswaran, M., 2014. A hybrid optimized weighted minimum spanning tree for the shortest intrapath selection in
wireless sensor network. Mathematical Problems in Engineering, 2014.
27. Rao PS, Banka H. Energy efficient clustering algorithms for wireless sensor networks: novel chemical reaction optimization approach.
Wireless Networks. 2017;23(2):433-452.
28. Banka, H. and Jana, P.K., 2016. PSO-based multiple-sink placement algorithm for protracting the lifetime of wireless sensor networks.
In Proceedings of the second international conference on computer and communication technologies (pp. 605-616). Springer, New Delhi.
29. Rao PS, Jana PK, Banka H. A particle swarm optimization based energy efficient cluster head selection algorithm for wireless sensor net-
works. Wireless networks. 2017;23(7):2005-2020.
30. Parwekar P. SGO a new approach for energy efficient clustering in WSN. International Journal of Natural Computing Research (IJNCR).
2018;7(3):54-72.
14 of 15 AGRAWAL ET AL.
31. Zhao X, Zhu H, Aleksic S, Gao Q. Energy-efficient routing protocol for wireless sensor networks based on improved grey wolf optimizer.
KSII Transactions on Internet & Information Systems. 2018;12(6):4.
32. Vijayalakshmi K, Anandan P. A multi objective Tabu particle swarm optimization for effective cluster head selection in WSN. Cluster
Computing. 2019;22(5):12275-12282.
33. Kousar, A., Mittal, N. and Singh, P., 2020. An improved hierarchical clustering method for mobile wireless sensor network using type-2
fuzzy logic. In Proceedings of ICETIT 2019 (pp. 128-140). Springer, Cham.
34. Verma S, Sood N, Sharma AK. Genetic algorithm-based optimized cluster head selection for single and multiple data sinks in heteroge-
neous wireless sensor network. Applied Soft Computing. 2019;85:105788.
How to cite this article: Agrawal D, Wasim Qureshi MH, Pincha P, et al. GWO-C: Grey wolf optimizer-based
clustering scheme for WSNs. Int J Commun Syst. 2020;e4344. https://doi.org/10.1002/dac.4344
AGRAWAL ET AL.15 of 15
A preview of this full-text is provided by Wiley.
Content available from International Journal of Communication Systems
This content is subject to copyright. Terms and conditions apply.
