Content uploaded by Khaled Elleithy
Author content
All content in this area was uploaded by Khaled Elleithy on Sep 06, 2015
Content may be subject to copyright.
Intern ational Journ al of Scientific & E ngineering Research, Volume 6, Issue 8, August-2015 1243
ISSN 2229-5518
IJSER © 2015
http://www.ijser.org
Dynamic Clustering of Heterogeneous Wireless
Sensor Networks using a Genetic Algorithm,
Towards Balancing Energy Exhaustion
Mohamed Elhoseny1, Khaled Elleithy2, Hamdi Elminir 3, Xiaohui Yuan4, and Alaa Riad 5
Abstract— Placing few heterogeneous nodes in Wireless Sensor network (WSN), such as nodes with more computing powers, is an
effective way to increase network availability in terms of lifetime. Despite the success of various clustering strategies of heterogeneous
WSN, the numerous possible sensor clusters make searching for an optimal network structure an open challenge. In this paper, we
propose a heterogeneous sensor node clustering method using a Genetic Algorithm to optimize the energy exhaustion namely Dynamic
Clustering of Heterogeneous WSNs using Genetic Algorithm ’DCHGA’. In DCHGA, the structure of the network is dynamically decided after
each message transmission round. Compared with state-of-the-art methods, DCHGA greatly extended the network life and the average
improvement with respect to the second best performance (using stable nodes) based on the first-node-die and the last-node-die were
33.8% and 13%, respectively. While in case of mobility heterogeneity of sensors, the improvement was between 12.6% and 9.8%. The
balanced energy consumption greatly improved the network lifetime and allowed the sensor’s energy to evenly deplete. The computational
efficiency of DCHGA is comparable to the others and the overall average time across all experiments was 0.6 seconds with a standard
deviation of 0.06.
Index Terms— Dynamic Clustering, Wireless Sensor Network, Heterogeneous Sensors
—————————— ——————————
1 INTRODUCTION
he heterogeneous clustering model has been used in Wire-
less Sensor Networks (WSNs) to improve its performance
in terms of network availability [1]. Although there are
great works in the process of forming clusters, the dynamic
nature of WSN and numerous possible cluster configurations
make searching for an optimal network structure a complicat-
ed defy [2]. The heterogeneous model is an adapted model of
homogeneous clustering model, i.e., LEACH [3]. This modifi-
cation can be achieved by placing few heterogeneous nodes in
network [4]–[6], such as nodes with more computing power.
In a heterogeneous WSN, in addition to the network structur-
ing factors, e.g., distance to the base-station, and distance
among nodes, factors such as initial energy, data processing
capability, ability to serve as a cluster head, and node mobility
greatly influence the network lifespan [7]–[9]. Moreover, the
lifetime of the network is maximized when the remaining en-
ergy of nodes in the network remains the same during the
network lifetime. This is, however, difficult to achieve in a
real-world WSN due to different roles of sensor nodes and
various signal transmission distance. The nodes serving as
cluster head consume additional energy to fulfill tasks such as
receiving messages from member nodes and relaying the ag-
gregated messages to the base station. Balancing node energy
consumption and extending the overall network lifespan are
non-trivial given many factors that could affect the energy
expenditure of each node [10], [11].
To extend the network lifetime in a heterogeneous network,
several methods have been proposed that account for one or
more of these factors. Stable Election Protocol (SEP) [4] used
weighted probabilities to elect cluster heads depending on the
remaining energy in the sensor nodes. In addition, Developed
Distributed Energy-Efficient Clustering (DDEEC) [5] method
improved upon SEP by categorizing sensor nodes based on
their energy level. The nodes with higher energy were the
“advanced nodes” and the cluster head were selected from
these group of nodes. Threshold Sensitive Stable Election Pro-
tocol (TSEP) [6] extended SEP method by grouping sensor
nodes into three energy levels, and the cluster heads were se-
lected based on thresholds. Similarly, Energy Efficient Hetero-
geneous Clustered scheme (EEHC) [12] and Efficient Three
Level Energy algorithm (ETLE) [13] selected cluster heads
based on the probability proportional to the residual energy.
In Hybrid Energy Efficient Reactive protocol (HEER) [14], the
cluster head selection is based on the ratio of residual energy
of nodes and the average energy of the network. Both of Ener-
gy efficient heterogeneous clustered scheme (EEHC) [12] and
Efficient Three Level Energy algorithm (ETLE) [13] assume
three levels of heterogeneity and nodes are randomly distrib-
uted and are stationary. In EEHC, the cluster heads were se-
lected based on weighted election probabilities of each node
according to the residual energy. While in ETLE, each node
T
————————————————
• 1,4 Department of Computer Science and Engineering, University of North
Texas, Denton, TX, U.S.A.; E-Mails: Mohamed.elhoseny@unt.edu, xiao-
hui.yuan@ unt.edu
• 2Department of Computer Science and Engineering, University of Bridge-
port, CT, U.S.A.; E-Mail: elleithy@bridgeport.edu
• 4Department of Electrical Engineering, Kafr El-Sheikh University, Kafr El-
Sheikh, Egypt; E-mail: hamdy elminir@eng.kfs.edu.eg
• 5Department of Information Systems, Mansoura University, Manso ura,
Egypt; E-mail: amriad_2000@mans.edu.eg
IJSER
Intern ational Journ al of Scientific & E ngineering Research, Volume 6, Issue 8, August-2015 1244
ISSN 2229-5518
IJSER © 2015
http://www.ijser.org
chose a random number between 0 and 1. If the value of this
random number was less than a threshold value , i.e., T , the
node will be selected to serve as a cluster head. In Hybrid En-
ergy Efficient Reactive protocol (HEER) [14], the CH selection
is based on the ratio of residual energy of nodes and average
energy of the network. All of these methods were proposed
for WSN with initial energy as the only heterogeneity factor.
Although most of current research concentrated on energy as
the only heterogeneity factor, many types of heterogeneous
resources, e.g., communication capability, data processing
power, and efficiency, were introduced to WSN for improved
performance. Providing sensor node with more processing
capabilities aims to prevent it from exhausting its energy
quickly in case of acting as a cluster head. Allowing mobile
sensors in heterogeneous model increases the number of WSN
applications compared with stationary sensors, i.e., tracking
animal movements applications [15]. On the other hand, deny-
ing some nodes to serve as a cluster head, e.g., nodes with low
energy, increases its chance to stay alive. Searching for a bal-
ance among many factors is an involved and complex process.
Heuristic optimization methods, such as Genetic Algorithms
(GA), have been employed in the routing protocols of WSN
[16]–[19]. When GA is used, a key objective is to define an ap-
propriate fitness function that encodes the network structure.
However most of GA-based work reported in literature was
developed for homogeneous model, i.e., HCR [16], [20], while
the remaining was concerned with heterogeneous WSN in
which the difference between sensors in the initial energy is
the dominant factor of heterogeneity. The Evolutionary Based
Clustered Routing Protocol (ERP) [18] overcames the limita-
tions of clustering-algorithm-based GAs by uniting the cluster-
ing aspects of cohesion and separation error, and proposed a
new fitness function based on these two aspects.
In this article, we propose a sensor clustering method for dy-
namically organizing heterogeneous WSN using GA called
’DCHGA’. DCHGA provides a framework to integrate mul-
tiple heterogeneity and clustering factors, which employs
remaining energy, expected energy expenditure, network lo-
cality, and distance to the base station in search for an optimal,
dynamic network structure for heterogeneous WSN. Hetero-
geneity factors are integrated as constraints to chromosomes
and validation is performed to ensure network integrity. To
avoid high energy consumption of sensor nodes, the base sta-
tion will run the GA after each round to dynamically forming
the structure of the network based on the new characteristics
of the sensors, i.e., remaining energy. Genetic algorithm uses
random search to suggest the best appropriate design. We use
this algorithm in order to obtain the most efficient clustering
structure. The reason for choosing GA is its convergence and
its flexibility in solving multi-objective optimization problems
like dynamic clustering of WSN [16]
The contribution of this work includes: First, a GA-based
method is proposed to provide a dynamic clustering method
for WSN. In addition, it provides flexibility of optimizing mul-
tiple factors concurrently. The GA chromosomes encode the
selection of cluster heads and the dynamic clusters are formed
accordingly. The separation of cluster head selection and net-
work structure makes this method versatile for integrating
diverse factors. Second, the expected energy expenditure is
derived, together with other energy and spatial metrics, to
achieve balanced energy consumption across all nodes and
improve the network’s longevity.
In the reminder of this article, section 2 presents the related
work of constructing heterogeneous WSN to extend its life-
time. Then, section 3 describes our proposed method for het-
erogeneous WSNs construction. Section 4 discusses our exper-
imental results including a comparison study with state-of-
the-art methods and analysis of energy consumption. Section 5
provides conclusions of this paper.
2 RELATED WORK
To extend the network lifetime in a heterogeneous network,
Stable Election Protocol (SEP) [4] used weighted probabilities
to elect cluster heads depending on the remaining energy in
sensor nodes. In addition, Developed Distributed Energy-
Efficient Clustering (DDEEC) [5] method improved upon SEP
by categorizing sensor nodes based on their energy level. The
nodes with higher energy were the “advanced nodes” and the
cluster head were selected from these group of nodes. Thresh-
old Sensitive Stable Election Protocol (TSEP) [6] extended SEP
method by grouping sensor nodes into three energy levels,
and the cluster heads were selected based on thresholds.
Similarly, Energy Efficient Heterogeneous Clustered scheme
(EEHC) [12] and Efficient Three Level Energy algorithm
(ETLE) [13] select cluster heads based on the probability pro-
portional to the residual energy. In Hybrid Energy Efficient
Reactive protocol (HEER) [14], the cluster head selection is
based on the ratio of residual energy of nodes and the average
energy of the network. Both of Energy efficient heterogeneous
clustered scheme (EEHC) [12] and Efficient Three Level Ener-
gy algorithm (ETLE) [13] assume three levels of heterogeneity
and nodes are randomly distributed and are stationary. In
EEHC, the cluster heads are selected based on weighted elec-
tion probabilities of each node according to the residual ener-
gy. While in ETLE, each node chooses a random number be-
tween 0 and 1. If the value of this random number was less
than a threshold value, i.e., T , the node will be selected to
serve as a cluster head. In Hybrid Energy Efficient Reactive
protocol (HEER) [14], the CH selection is based on the ratio of
residual energy of nodes and average energy of the network.
All of these methods are proposed for WSN with initial energy
as the heterogeneity factor.
In [21], the Degree of connectivity is the main factor of select-
ing a CH. The degree of connectivity of a node, i.e. the number
of its neighbors, is also a criterion that seems interesting to
study. Intuitively, the more neighbors a sensor has, the more it
seems to be an appropriate candidate as cluster head, since a
sensor with a low degree of connectivity might have little in-
formation, from its neighborhood, to aggregate and to forward
to the BS. In the initial phase, each sensor is involved in the
neighborhood information exchanges (hello protocol), which
allows it to determine its degree of connectivity and the loca-
tion of BS. In EEUC [22], the distance between the node and
the BS is the main parameter for selecting the CH. The EEUC
IJSER
Intern ational Journ al of Scientific & E ngineering Research, Volume 6, Issue 8, August-2015 1245
ISSN 2229-5518
IJSER © 2015
http://www.ijser.org
resulted in a network that is partitioned into clusters of une-
qual size, and that the clusters closer to BS have smaller sizes
than those farther from the BS.
Many intelligent algorithms which provide adaptive methods
that present intelligent behavior in complex and dynamic en-
vironments like WSNs are exist [23]. Various works reported
in literature [23]–[27] debated the routing protocols in Cluster-
based WSN based on intelligent algorithms as reinforcement
learning, ant colony optimization, fuzzy logic, genetic algo-
rithm, and neural networks. Furthermore, a lot of clustering
mechanisms have been proposed. For example, Local Negoti-
ated Clustering Algorithm presents a novel clustering method,
which uses the similarity of nodes readings as an important
feature during the process of creating cluster. ACE constructs
the WSN clusters in a fixed number of iterations using the
node degree as the main factor. In GA-WCA, load balanced
factor with a sum of distance from all neighbor nodes to CHs
represents the main factor in network construction. on the
other hand, LA2D-GA depends only on the distance as the
main parameter to calculate fitness function that is used to
evaluate the chance of the node to be a CH. LA2D-GA repre-
sent the chromosome in a two-dimensional grid which eluci-
dates valid statistics of a WSN [28].
In [29], a two level fuzzy logic approach is used to Cluster
Head (CH) election based on four parameters namely- number
of neighbor nodes, remaining energy, energy dispersion and
distance from the base station. The authors supported their
idea for number of neighbors by stating that The number of
neighbor nodes has been considered to be one determining
parameter because CH must be chosen from an area where
sufficient neighbor nodes are available LELE [30] protocol se-
lects CH on basis of remaining energy and the distance be-
tween a node and its neighbors, and the node with maximum
energy and suitable position is chosen as the CH. LELE is pro-
posed to improve load balancing in LEACH protocol Leader
Election with Load balancing Energy . So when the network is
operating, the probability of the nodes becoming leader de-
creases or increases depending on the difference of the energy
level of one node and neighbors, the distance of the node from
neighbors, as well as the number of neighbors, and the proba-
bility of the nodes’ to become a leader.
In [21], the Degree of connectivity is the main factor of select-
ing a CH. The degree of connectivity of a node, i.e. the number
of its neighbors, is also a criterion that seems interesting to
study. Intuitively, the more neighbors a sensor has, the more it
seems to be an appropriate candidate as a cluster head, since a
sensor with a low degree of connectivity might have little in-
formation, from its neighborhood, to aggregate and to forward
to the BS. In the initial phase, each sensor is involved in the
neighborhood information exchanges (hello protocol), which
allows it to determine its degree of connectivity and the loca-
tion of BS. In EEUC [22], the distance between the node and
the BS is the main parameter for selecting the CH. The EEUC
resulted in a network that is partitioned into clusters of une-
qual size, and that the clusters closer to BS have smaller sizes
than those farther from the BS.
Searching for an optimal balance among many factors is non-
trivial. Genetic Algorithm (GA) has been applied in the rout-
ing protocol of WSN [17]–[19]. When GA is used, a key objec-
tive is to define an appropriate fitness function that encodes
the network structure and its goodness. However, most of
GA-based work was developed for homogeneous model, i.e.,
HCR [20], while the remaining was concerned with static het-
erogeneous WSN. There are no additional efforts reported for
the mobile heterogeneous model. The Evolutionary Based
Clustered Routing Protocol (ERP) [18] overcomes the limita-
tions of clustering-algorithm-based genetic algorithms by unit-
ing the clustering aspects of cohesion and separation error,
and proposed a new fitness function based on these two as-
pects.
3 HETEROGENEOUS WSN CLUSTERING USING GENEIC
ALGORITHM
3.1 Energy Model and Clustering Factors
As we deal with two levels of heterogeneity, our model has
two types of sensors: normal and advanced sensor nodes. The
advanced sensor has additional initial energy and lower ener-
gy consumption for data processing, i.e., receiving and trans-
mitting messages. Based on that, we adopt the first order radio
model to describe the sensor’s energy [31] as shown in Figure
1. The consumed energy E of a normal sensor node s is the
summation of energy used to:
1) Acquire l bits of data (
)(lE
A
S
)
2) Receive l’ bits of data (
)'(lE R
S
)
3) Process l’’ bits of data (
)''(lE P
S
)
4) Transmit l’’ bits of data over distance d (
),''( dlE T
S
),
and
5) Move from location x to location y.
),(),''()''()'()( yxEdlElElElEE
M
S
T
S
P
S
R
S
A
Ss
++++=
(1)
where
*
'ElEE i
R
S+=
and
i
E
is the idle energy expendi-
ture.
n
i
T
SdlEE ''+=
, and n = 4 for long distance transmis-
sion and n = 2 for short distance transmission, and
*
E
repre-
sents the cost of beam forming approach for energy reduction.
The long and short transmission distance is determined by the
distance threshold as we will explain later.
To compute the expected consumed energy
^
E
of a non-CH
sensor node
'
s
and a CH sensor node s, assume l bits of data
are collected by each sensor node in a round. Given Ns sen-
sors in a cluster, the expected consumed energy
^
E
are com-
puted as follows:
IJSER
Intern ational Journ al of Scientific & E ngineering Research, Volume 6, Issue 8, August-2015 1246
ISSN 2229-5518
IJSER © 2015
http://www.ijser.org
),'(
2
^
'sslDEES+=
, (2)
),,'()1(
4*
^
BslDNlENEE
ss
S
+++=
(3)
where E is the constant energy consumption including the
energy of data acquisition, processing, idle and moving. Func-
tions D(s’, B) and D(s, B) use Euclidean distance to give the
distance between sensor nodes inside the cluster and from the
cluster head to the base station, respectively. In addition, the
local sensor density is proportional to the number of sensors
within the δ-vicinity as follows:
Gs
(
δ
)
∝
||Ss||,
and
Ss
=
{s
i
;
D
(
s, si
)
≤ δ}
(4)
where Ss is the set of sensor nodes in the δ-vicinity of s and
function || · || gives the set size.
3.2 Network Structure Building using Genetic
Algorithm
In our proposed framework, a binary chromosome is used to
specify the CHs in the network, in which a one represents a
CH and a zero represents a member node to a cluster as
shown at Figure 2. When a sensor becomes inactive, i.e., out of
power, its corresponding gene value is set to -1, which ex-
empts this sensor from further GA operations.
The mapping to sensor clusters from a chromosome is to min-
imize the network communication distance D as follows:
∑∑
= =
=
C
i
N
jj
i
s
ssDD
1 1
),(
(5)
where C is the number of clusters in a network and Nsi is the
number of member nodes in a cluster headed by node si. In
practice, minimizing D is equivalent to assigning sensor nodes
to clusters following the nearest neighbor rule.
The fitness function integrates energy factors, spatial distanc-
es, and the local sensor density:
∑∑
′
′
+++=
ss
ss
sG
N
DE
E
EtE
f),(
1
ˆ
1
ˆ
~
)0( )(
δ
(6)
where Es(t) is the remaining energy of sensor node s at round t
and Es(0) is the initial energy of sensor node s.
E
~
is the total
energy cost if the messages are transmitted directly from all
sensor nodes to the BS.
D
ˆ
is the total distance between the
CHs and the BS:
∑
=
=
C
i
BSsDD
1
),(
ˆ
(7)
where each si is a sensor node that serves as a CH. Including
sensor density favors the choice of CHs with more close
neighbors. In cases where it is clear one or more factors play
more vital role, uneven weights can be employed in the fitness
function.
3.3 Network Structure Validation and Evaluation
In a heterogeneous WSN, functions and capabilities of sensors
significantly vary. Some sensors are unable to serve as a clus-
ter head and some are preferred to take the role due to their
superior processing power and available energy. However,
classical optimization method such as GA provides no inte-
grated mechanism for ensuring alignment of different roles of
the sensors. In addition, the random initialization and GA op-
erations could introduce chromosomes that completely violate
the current sensor properties. In DCHGA, heterogeneity is
presented as constraints and hence a validation process is
needed before evaluating chromosomes’ fitness to ensure
network integrity.
Figure 2 illustrates also the validation process that leverages
static and dynamic sensor properties. In the process of GA
optimization, a new chromosome represents the proposed
structure for the WSN. Each gene in the chromosome defines
Fig. 1. First order radio model of a node. Each component has
an ener
gy consumption model that is a function of message
length.
Fig. 2. Framework for Heterogeneous WSN Clustering using Genetic
Algorithm.
IJSER
Intern ational Journ al of Scientific & E ngineering Research, Volume 6, Issue 8, August-2015 1247
ISSN 2229-5518
IJSER © 2015
http://www.ijser.org
the expected role of the corresponding sensor node, i.e.,
whether it serves as a cluster head or a member node. The
process consults ‘the ability to serve as a CH’, and ‘the Suffi-
cient Energy’ tables. The role of ‘The ability to serve as a CH’
table is to determine whether the node can serve as a cluster
head (one represents serving as cluster head; otherwise, mem-
ber node). While, the ‘Sufficient Energy’ table is used to show
the current energy status of the node, i.e., zero for disabled
node and one for available node. The validation process de-
termines if a chromosome complies with the constraints and
hence retained in the offspring pool; otherwise, the chromo-
some is abandoned.
GA generates new chromosomes through crossover and muta-
tion operations and evaluates their fitness. The crossover op-
eration is performed with two randomly selected chromo-
somes determined by a crossover probability to regulate the
operation. When crossover is excluded, the parent chromo-
somes are duplicated to the offspring without change. Varying
the crossover probability alters the evolution speed of the
search process. In practice, the value of crossover is close to 1.
The mutation operation involves altering the value at a ran-
domly selected gene within the chromosome. Similarly, a mu-
tation probability is used to regulate the performance of muta-
tion. Different from the crossover probability, the mutation
probability is usually fairly small. Essentially mutation opera-
tion could create completely new species, i.e., an arbitrary lo-
cus in the fitness landscape. Hence, it is a means to get out of a
local optimum. Recall that when a sensor node becomes inac-
tive, its corresponding gene is set to -1 to exempt it from muta-
tion operations.
After the validation process is executed, Eq. (6) is used to
evaluate the fitness of chromosomes. An intermediate pool of
chromosomes is created to hold the individuals created in a
generation, and depending on the needs, the user can specify
any intermediate population size that is greater than the initial
population size.
The evolution terminates when one of the following criteria is
satisfied: 1) the maximum number of generations is reached;
or 2) the fitness function converges. Upon completion of the
GA evolution, the chromosome that gives the best fitness val-
ue is used to restructuring the nodes.
3.4 Heterogeneous WSN clustering Algorithm
Algorithm 1 presents DCHGA method. In this algorithm,
],1[ Qq∈
denotes the number of generations, and the popula-
tion size is P. The pool of chromosome, denoted by U, is ini-
tialized with randomly generated individuals.
In crossover operation, two chromosomes are randomly se-
lected from U and, according to the crossover probability α,
two new chromosomes are created by switching consecutive
genes. In mutation operations, the value of a randomly picked
gene is altered between 0 and 1 according to the mutation
probability β.
4 RESULTS AND DISCUSSION
In our evaluation, we assumed that each sensor node can di-
rectly reach the base station if it is provided with sufficient
energy. The simulated sensor network was in an area of 100
meters by 100 meters (m) with 50 sensors randomly placed in
the field and the data packet size was 400 bits. The network
parameters used in our experiments are listed in Table I. The
heterogeneity includes different initial power, data processing
efficiency, capability of serving as cluster head, and node mo-
bility. For the sensors with greater data processing efficiency,
the energy used is 50% of that used by a regular sensor. 10% of
sensor nodes possessed greater initial energy and data pro-
cessing efficiency and 10% of sensor nodes were unable to
serve as cluster heads. The heterogeneous sensors were chosen
randomly in each experiment. Regarding GA running parame-
ters, we used the population size of 30 for 30 generations. The
crossover probability and mutation probability are 0.8 and
0.006, respectively. As we mentioned before, the mutation
process should be close to 1 while the mutation should be
close to zero. The neighborhood distance δ was 20 meters (m)
throughout our experiments.
To evaluate DCHGA in different environments, we created
two cases with low and high sensors density, i.e., 50 node and
100 node, respectively. For each case, two scenarios of hetero-
geneity are designed: 1) sensors may differs in their initial en-
ergy , and 2) sensors may differs in initial energy, data pro-
IJSER
Intern ational Journ al of Scientific & E ngineering Research, Volume 6, Issue 8, August-2015 1248
ISSN 2229-5518
IJSER © 2015
http://www.ijser.org
cessing capability, and the ability to serve as a cluster head.
Comparison studies were conducted with five state-of-the-art
methods including HEER [14], TSEP [6], DDEEC [5], ETLE
[13], and ERP [18].
Scenario 1: This scenario aims to evaluate the impact of
heterogeneity in terms of initial energy of the sensor nodes.
Table VI compares network life of DCHGA with five state-of-
the-art methods, which include HEER [14], TSEP [6], DDEEC
[5], ETLE [13], and ERP [18]. The average number of rounds
when the first node died (FND) and last node died (LND) are
reported; and 10 experiments were conducted for the analysis.
DCHGA exhibited the longest average network life. The aver-
age improvement with respect to the second best performance
based on FND and LND are 33.8% and 13%, respectively. Fig.
3 depicts the number of live nodes throughout the network
life, which presents a progressive view. The dash line with
solid dot shows the results of DCHGA. The balanced energy
consumption greatly improved the network life and allowed
the sensor energy to deplete evenly. This means the stability
[18] of DCHGA is the best one compared with the five other
methods. It is clear that DCHGA greatly extended the network
life.
Fig. 3 depicts the change of the percentage of live sensor
nodes throughout the entire network life. It is evident that the
improvement of DCHGA method is significant.
Scenario 2: The purpose of this scenario is to evaluate the
impact of heterogeneity in terms of initial energy, processing
capability, and the ability of the sensor to act as a cluster head.
Table III presents the percentage of live sensor nodes through-
out the life span of the WSN using the two cases for each sce-
nario. The number of round is the average of 10 experiments
with random sensor node placement. Compared with initial
energy heterogeneity scenario, using the mentioned three fac-
tors of heterogeneity together yielded the largest number of
rounds when the first sensor node dies. Depending on the
sensor density, the improvement of using these three factors of
heterogeneity was in the ranges of 20.8% to 38.4%. This means
the more heterogeneity capabilities assigned to the advanced
nodes, the more network lifetime. It is clear that our proposed
heterogeneity factors greatly extended the network lifespan.
Fig. 3 (a). Network lifetime f or high density field with 100 sensors
Fig. 3 (b). Network lifetime f or low density field with 50 sensors
TABLE III. NETWORK LIFE SPAN WITH DIFFERENT HETEROGE-
NEITY FACTORS IN THE TWO PROPOSED CASES. S-CH: SOME
NODES CAN NOT SERVE AS A CLUSTERS HEAD.
TABLE II. NETWORK LIFE SPAN WITH DIFFERENT METHODS IN
THE TWO PROPOSED CASES. FND: ROUND AT WHICH FIRST
NODE DIE. LND: ROUND AT WHICH LAST NODE DIE.
IJSER
Intern ational Journ al of Scientific & E ngineering Research, Volume 6, Issue 8, August-2015 1249
ISSN 2229-5518
IJSER © 2015
http://www.ijser.org
Figure 4 depicts the change of the percentage of live sensor
nodes throughout the entire network life. It is evident that the
improvement of using different heterogeneity factors is signif-
icant. Our experiments also showed that the average number
of clusters before the first node die was 7% and 8% in high and
low density fields respectively.
Figure 5 illustrates an example of the remaining energy at
various transmission rounds of all sensors. At round 0, i.e., the
initialization, 5 nodes (highlighted with green bars) were
fueled with greater energy at 1J. The red bars mark sensors
unable to serve as cluster head. As transmission continued, the
remaining energy of sensors gradually reduced mostly evenly.
Table IV lists the average remaining energy of the low-
initial-energy sensors and its standard deviation at various
transmission rounds. Due to unequal distances to the cluster
head of the member nodes, energy expenditure for sensors
varied, and it is inevitable that STDs continued to increase.
However, the small STDs indicate balanced energy consump-
tion among sensors.
Figure 6 illustrates the spatial and frequency view of sensor
nodes serving as cluster heads throughout the life of the net-
work. The size of sphere is proportional to the number of
times a sensor served as cluster head. It is clear that the ones
with higher initial energy served as cluster head most times.
The placement of higher energy sensors is randomized, which
is unfortunately uneven in the field. Despite that the high-
initial-energy sensors dominated the choice of cluster head,
their spatial disadvantage, i.e., closely located with each other,
made some low-initial- energy sensors to act as cluster head to
serve nearby sensors. The average number of clusters in all
rounds of our 10 experiments is 6, among which 97% of times
high-initial-energy nodes served as cluster head. The forming
of clusters was greatly influenced by the spatial location of
sensor nodes. It is interesting to see that the low-initial- energy
nodes that served as cluster head are usually far away from
the high-initial-energy ones, which justifies their role as cluster
head.
Fig. 4 (a). Network lifetime for different terms of heterogeneity using
high density field with 100 sensors
Fig. 4 (b). Network lifetime for different terms of heterogeneity using
low density field with 50 sensors
Fig. 5. The remaining energy of sensor nodes at various transmission
TABLE IV. REMAINING ENERGY (J) VARIANCE OF SENSOR
NODES.
Fig. 6. Spatial and frequency view of sensor nodes serving as cluster
head.
IJSER
Intern ational Journ al of Scientific & E ngineering Research, Volume 6, Issue 8, August-2015 1250
ISSN 2229-5518
IJSER © 2015
http://www.ijser.org
Efficiency is an important factor in real-world applications.
Our experiments were conducted in a computer with Intel
core i5 2.6GHz CPU, 4GB memory, and Windows 7 operating
system. The algorithms were implemented with C# program-
ming language. Table VII lists the average time used to struc-
ture clusters in each transmission round. The time reported is
before the first node became unavailable due to energy ex-
haustion. The number within parenthesis is the standard devi-
ation. In addition to 50 sensors in the field, we also experi-
mented with 100 randomly placed sensors with the other pa-
rameters remain the same. The average time used by GAHN
was comparable to the other methods.
To evaluate the proposed method in terms of mobility het-
erogeneity, we will use M-DCHGA to indicate to our method
with node mobility. Table VI presents the percentage of live
sensor nodes throughout the life span of the WSN using dif-
ferent cases, i.e, static and mobile sensors. The number of
round is the average of 10 experiments with random sensor
node placement. Depending on the sensor density, the im-
provement was in the ranges of 27.3% to 33.44% using static
sensors. While in case of sensors mobility, it was between
12.6% and 9.8%. This means the stability [18] of DCHGA is the
best one compared with the five other methods. It is clear that
DCHGA method greatly extended the network life.
Figure 7 depicts the change of the percentage of live sensor
nodes throughout the entire network life. The experiments
showed that the average number of clusters before the first
node die was 8%, and 6% for high density scenario in case of
static and mobile sensors respectively. While it was 9%, and
8% for low density scenario in case of static and mobile sen-
sors respectively.
To evaluate M-DCHGA efficiency, table VII lists the aver-
age time (in seconds) and standard deviation used to form
clusters in each transmission round. The time reported is be-
fore the first node became unavailable due to energy exhaus-
tion. Despite the standard deviation increased when the num-
ber of sensor nodes was doubled, the average time was very
close for all cases. It is evident that the efficiency of DCHGA is
mostly independent from sensor mobility and number of sen-
sors. The overall average time across all experiments is 0.6
seconds with a standard deviation of 0.06. The efficiency of M-
DCHGA method is also satisfactory.
TABLE V. AVERAGE TIME (IN SECONDS) FOR NETWORK STRUC-
TURING.
TABLE VI. NETWORK LIFE SPAN USING DCHGA METHOD WITH
MOBILE NODES HETEROGENEITY. FND: ROUND AT WHICH FIRST
NODE DIE. LND: ROUND AT WHICH LAST NODE DIE.
Fig. 7 (a). Network lifetime for high density scenario
Fig. 7 (b). Network lifetime f or low density scenario
TABLE VII. AVERAGE TIME (IN SECONDS) USED TO IDENTIFY
OPTIMAL NETWORK STRUCTURE IN EACH ROUND USING
DCHGA AND M-DCHGA COMPARED WITH THE FIVE STATE-OF-
ART METHODS.
IJSER
Intern ational Journ al of Scientific & E ngineering Research, Volume 6, Issue 8, August-2015 1251
ISSN 2229-5518
IJSER © 2015
http://www.ijser.org
5 CONCLUSIONS
In homogeneous WSN, clustering protocols assumed that
all the sensor nodes are supplied with the same characteristics
i.e., initial energy. However, placing few heterogeneous nodes
in WSN, such as nodes with more computing powers, is an
effective way to increase network lifetime and reliability. In a
heterogeneous WSN, in addition to the network structuring
factors, e.g., distance to the base-station, and distance among
nodes, factors such as initial energy, data processing capabil-
ity, ability to serve as cluster head, node mobility greatly in-
fluence the network lifespan. In addition, the lifetime of the
network is maximized when the remaining energy of nodes in
the network remains the same during the network lifetime.
This is, however, difficult to achieve in a real-world WSN due
to different roles of sensor nodes and various signal transmis-
sion distance.
In this paper, we propose a heterogeneous sensor node
clustering method based on Genetic Algorithm called DCHGA
to optimize the energy exhaustion. In DCHGA, the structure
of the network is dynamically decided after each message
transmission round. In addition, it provides a framework to
integrate multiple heterogeneity factors, i.e., initial energy,
data processing capability, ability to serve as cluster head,
node mobility. Compared with state-of- the-art methods,
DCHGA greatly extended the network life and the average
improvement with respect to the second best performance
based on the first-node-die and the last-node-die were 33.8%
and 13%, respectively. While in case of sensors mobility, the
improvement was between 12.6% and 9.8%. The computation-
al efficiency of DCHGA is comparable to other algorithms and
the overall average time across all experiments was 0.6 se-
conds with a standard deviation of 0.06.
REFERENCES
[1] D. Xie, Q. Zhou, X. You, B. Li, and X. Yuan. A novel energy-efficient
cluster formation strategy: From the perspective of cluster members.
IEEE Communications Letters, 17(11):2044–2047, NOVEMBER 2013.
[2] M. Elhoseny, X. Yuan, H. ElMinir, a nd A. Riad. Extending self-
organizing network availability using genetic algorithm. In ICCCNT,
Hefei, China, July 2014.
[3] W. Heinzelman, A. Chandrakasan, and H. Balakrishnan. Energy-
efficient communication protocol for wireless microsensor networks.
In The Hawaii International Conference on System Sciences, Maui,
Hawaii, Jan 2000.
[4] G. Smaragdakis, I. Matta, and A. Bestavros. SEP: a stable election
protocol for clustered heterogeneous wireless sensor network. In Se-
cond International Workshop on Sensor and Actor Network Proto-
cols and Applications, 2004.
[5] B. Elbhiri, S. Rachid, and S. Elfkihi. Developed distributed energy-
effecient clustering (DDEEC) for heterogeneous wireless sensor. In
Communications and Mobile Network, pages 1–4, Rabat, 2010.
[6] A. Kashaf, N. Javaid, Z . Khan, and I. Khan. TSEP: Threshold-
sensitive stable election protocol for WSNs. In Conference on Fron-
tiers of Information Technology, pages 164–168, 2012.
[7] A. Iqbal, M. Akbar, N. Javaid, S. Bouk, M. Ilahi, and R. Khan. Ad-
vanced LEACH: A static clustering-based heterogeneous routing
protocol for WSNs. Journal of Basic and Applied Scientific Research,
3(5):864–872, 2013.
[8] M. Elhoseny, X. Yuan, Z. Yu, C. Mao, H. El-Minir, and A. Riad. Bal-
ancing energy consumption in heterogeneo us wireless sensor net-
works using genetic algorithm. IEEE Communications Letters,
pp(99):1, 2014.
[9] T. Sudeep, N. Kumar, and J. Niu. EEMHR: energy-efficient multi-
level heterogeneous routing protocol for wireless sensor networks.
International Journal of Communication Systems, 27(9):1289–1318,
2014.
[10] K. Tripathi, N. Singh, and K. Verma. Two-tiered wireless sensor net-
works - base station optimal positioning case study. IET Wireless
Sensor Systems, 2(4):351–360, December 2012.
[11] L. Wang, C. Wang, and C. Liu. Optimal number of clusters in dense
wireless sensor networks: A cross-layer approach. IEEE Transac-
tions On Vehicular Technology, 58(2):966–976, February 2009.
[12] D. Kumar, T. Aseri, and R. Patel. EEHC: energy efficient heterogene-
ous clustered scheme for wireless sensor network. Computer Com-
munications, 32(4):662–667, 2009.
[13] N. Tuah, M. Ismail, and K. Jumari. Energy efficient algorithm for
heterogeneous wireless sensor network. In IEEE International Con-
ference on Control System and Computing a nd Engineering, pages
92–96, Penang, November 2011.
[14] N. Javaid, N. Mohammad, K. Latif, U. Qasim, A. Khan, and M. Khan.
HEER: hybrid energy efficient reactive protocol for wireless sensor
networks. In Saudi International Electronics and Communications
and Photonics Conference, pages 1–4, Riyadh, April 2013.
[15] S. Ehsan, K. Bradford, M. Brugger, B. Hamdaoui, Y. Kovchegov, D .
Johnson, and M. Louhaichi. Design and analysis of delay-tolerant
sensor networks for monitoring and tracking free-roaming animals.
IEEE Transactions on Wireless Communications, 11(3):1220–1227,
2012.
[16] P. Ali, H. Mashhadi, and S. Javadi. An optimal energy efficient clus-
tering method in wireless sensor networks using multiobjective ge-
netic algorithm. International Journal of Communication Systems,
26(1):114–126, 2013.
[17] S. Bayrakl and S. Erdogan. Genetic algorithm based energy efficient
clusters in wireless sensor networks. Procedia Computer Science,
10:247–254, 2012.
[18] B. A. Attea and E. A. Khalil. A new evolutionary based routing pro-
tocol for clustered heterogeneous wireless sensor networks. Applied
Soft Computing, 12(7):1950–1957, 2012.
[19] Y. Wu and W. Liu. Routing protocol based on genetic algorithm for
energy harvesting-wireless sensor networks. IET Wireless Sensor
Systems, 3(2):112–118, June 2013.
[20] S. Hussain, A. Matin, and O. Islam. Genetic algorithm for energy
efficient clusters in wireless sensor networks. In The 4th International
Conference on Information Technology ITNG, pages 147–154. IEEE,
April 2007.
[21] C. Diallo, M. Marot, and M. Becker. Single-node cluster reduction in
WSN and energy-efficiency during cluster formation. In the 9th IFIP
Annual Mediterranean Ad Hoc Networking Conference, France,
2010.
[22] L. Chengfa, Y. Mao, C. Guihai, and W. Lie. An energy-efficient une-
qual clustering mechanism for wireless sensor networks. In IEEE In-
ternational Conference on Mobile Adhoc and Sensor Systems, Wash-
ington, DC, Nov 2005.
[23] W. Guo and W. Zhang. A survey on intelligent routing protocols in
wireless sensor networks. Journal of Network and Computer Appli-
cations, 38:185–201, 2014.
IJSER
Intern ational Journ al of Scientific & E ngineering Research, Volume 6, Issue 8, August-2015 1252
ISSN 2229-5518
IJSER © 2015
http://www.ijser.org
[24] G. Ahmed, N. Khan, and R. Ramer. Cluster head selection using
evolutionary computing in wireless sensor networks. In Progress in
Electromagnetics Research Symposium, pages 883–886, 2008.
[25] N. Bhaskar, B. Su bhabrata, and P. Soumen. Genetic algorithm based
optimization of clustering in ad-hoc networks. International Journal
of Computer Science and Information Security, 7(1):165–169, 2010.
[26] M. Asim and V. Mathur. Genetic algorithm based dynamic approach
for routing protocols in mobile ad hoc networks. Journal of Academ-
ia and Industrial Research, 2(7):437–441, 2013.
[27] A. Karimi, S. Abedini, F. Zarafshan, and S. Al-Haddad. Cluster head
selection using fuzzy logic and chaotic based genetic algorithm in
wireless sensor network. Journal of Basic and Applied Scientific Re-
search, 3(4):694–703, 2013.
[28] K. Rana and M. Zaveri. Synthesized cluster head selection and rout-
ing for two tier wireless sensor network. Journal of Computer Net-
works and Communications, 13(3), 2013.
[29] K. Kannammal, T. Purusothaman, and M. Manjusha. An efficient
cluster based routing in wireless sensor networks. Journal of Theo-
retical and Applied Information Technology, 59(3), 2014.
[30] M. Shirmohammadi, K. Faez, and M. Chhardoli. LELE: leader elec-
tion with load balancing energy. In International Conference on
Communications and Mobile Computing, pages 106–110, 2009.
[31] M. Elhoseny, X. Yuan, H. ElMinir, and A . Riad. Extending self-
organizing network availability using genetic algorithm. In ICCCNT,
Hefei, China, July 2014.
IJSER