Journal of Ambient Intelligence and Humanized Computing

https://doi.org/10.1007/s12652-020-02002-1

ORIGINAL RESEARCH

Implementation of thermal aware wireless sensor network clustering

algorithm based on fuzzy and spider optimized cluster head selection
R. Karpaga Priya1 · S. Venkatanarayanan2

Received: 4 February 2020 / Accepted: 17 April 2020

© Springer-Verlag GmbH Germany, part of Springer Nature 2020

Abstract
In a modern approach, a smart grid requires an innovation on various fronts such as a wireless sensor network which is an
important component of smart grid applicability. To fulfill the Quality of Service (QoS) needs in smart grid network, a
structure of network and topology should be optimized especially in the urban areas. The clustering methodologies are useful
technique for optimizing network topologies. In clustering, the clustering process consists of cluster head node selection and
rotation which is based on Residual Energy, Distance of node from base station etc. However, the impact of temperature rise
has not been considered so far. Based on the previous related works, this paper proposes a Thermal Aware solution based
on combining Eigen Centrality Fuzzy Cluster size Control and Spider Optimization Algorithm. Furthermore, an influence of
temperature can be realized through Received Signal Strength (RSS) and the number of packets received. The proposed algo-
rithm is simulated in MATLAB and implemented in hardware test bed using Zigbee and PIC microcontroller. Consequently,
the result confirms the impact of thermal heat on Cluster Head selection control and also the prediction of number of rounds.

1 Introduction Clustering is one of the essential methods for increasing

the network lifetime in Wireless Sensor Networks (WSNs).
Wireless Sensor Networks (WSNs) have been an excellent A major challenge in WSN is to select appropriate cluster
candidate for online monitoring and replacement for manual heads and it also includes grouping of sensor nodes into
diagnosis. Because of their unique feature and the advan- clusters and electing Cluster Heads (CHs) for all the clusters.
tages, wireless sensor network based smart grid monitoring Each CH collects the data from the respective cluster nodes
has favoured like self configuration, low cost, low power and forwards the aggregated data to the base station. Hence,
consumption and rapid deployment (Ebrahimi et al. 2011; it consumes more power than other nodes in a cluster. A
Al-Anbagi et al. 2013). However, many of these deploy- tremendous challenging task in that clustering is to select
ments are heavily exposed to temperature raised by larger suitable cluster heads.
packet transferring and temperature changes within a day In a cluster, a cluster rotation is a common technique for
over seasons. Temperature has considerable impact on WSN balancing energy dissipation and to improve network life-
and hence, the network has to be designed to tolerate all the time. A very first well known hierarchical routing protocol
possible temperature changes (Monowar and Bajaber 2015; has been introduced by Heinzelman et al. 2000 and it is
Boanoy et al. 2013; Bag and Bassiouni 2008). Temperature called as LEACH. In LEACH, the clusters are constructed
changes can lead to loss of link quality and synchronization based on the strength of signal received by sensor nodes. By
or early battery exhaustion. Further, it can also influence considering leach as a basic protocol, more number of clus-
key network parameters like throughput, lifetime and delay tering protocol has been developed. Most standard clustering
(Bannister et al. 2008). algorithms include LEACH (Heinzelman et al. 2000), HEED
(Younis and Fahmy 2004), PEGASIS (Lindsey and Raghav-
endra 2002), EECS (Ye et al. 2005), EEMC (Jin et al. 2008),
TEEN (Manjeshwar and Agrawal 2001), PANEL (Buttyan
* R. Karpaga Priya and Schaffer 2007) and TL-LEACH (Hong et al. 2008).
[email protected] In the existing works, the intelligent techniques such
1
ECE, VCE, Madurai, India as Genetic algorithm (GA), Fuzzy logic (FL), Ant Colony
2 Optimization (ACO) (Taheri et al. 2012; Singh and Soni
EEE, KLNCE, Madurai, India

13
Vol.:(0123456789)
 R. K. Priya, S. Venkatanarayanan

2019), Particle Swarm Optimization (PSO), Artificial Bee centrality and it is named as Eigenvector centrality based
Colony (ABC) Optimization etc. are used for clustering. A Cluster Size Control (Ev-CSC). The cluster-head alters the
genetic algorithm based method (GABEEC) has been pro- communication radius of the respective cluster on the basis
posed to optimize the lifetime of wireless sensor networks of Eigenvector centrality and α-cut sets.
(Set-up and Steady-state phase). By using a GA, the method In some literatures, a clustering algorithm based on
is able to find an applicable number of cluster-heads and Swarm Intelligence has been proposed and in which, clus-
their locations. tering is to be considered as an optimization framework (Ni
The computational and swarm intelligence algorithms can et al. 2015). Ho and Ewe (2005) have used Ant colony opti-
be used to overcome various challenges in WSN (Nokhanji mization (ACO) to construct load balanced clusters with
et al. 2015; Ho and Ewe 2005; Arora et al. 2019). The the consideration of node mobility. Jiang et al. (2010) have
Computational Intelligence (CI) has been introduced for proposed Energy-Aware Unequal Clustering (EBUC) pro-
CH selection by considering various factors like Residual tocol using Particle Swarm Optimization (PSO) in order to
Energy (RE) of node, node degree, Mobility, and Distance overcome a hotspot problem. AbdelSalam and Olariu (2012)
from sink, Reliability and other combining relevant param- have proposed a Bio-inspired backbone Selection by Tiling
eters. To the best of the researcher’s knowledge, this is the the deployment area around sink nodes using identical regu-
very first work which has been proposed to design a node lar hexagons to mitigate the challenges in WSN. Elhabyan
clustering method based on thermal heat or temperature rise and Yagoub (2014) have proposed a PSO protocol for hier-
in WSN nodes. archical clustering in WSN. This enhances the energy effi-
In this approach, a Hybrid method of fuzzy logic and the ciency of network by setting an upper bound on the number
Heuristic Spider Optimization Algorithm (SOA) has been of CHs and also minimizes the number of CHs compared to
proposed. This proposed method works with two processes. that upper bound using PSO. Cai et al. (2015) have proposed
Firstly, the fuzzy algorithm deals with the initialization part a Bee-Sensor-C. It is a Bee inspired routing protocol which
of dividing the nodes to form a cluster and cluster size con- offers parallel data transmission and multipath construction
trol. Then, a Spider Optimization Algorithm (SOA) has been method.
proposed to select an efficient CH with the consideration
of Thermal Heat value and the distance between the nodes
in each clustering group. The algorithm aims at forming
well-balanced clusters so that; the load is balanced in the 3 Problem statement
wireless network.
The rest of paper has been structured as follows. Sec- The Traditional methods increase the system performances
tion 2 describes the works which are related to clustering by increasing the clock frequencies and hit a bottleneck,
algorithms. Section 3 provides the problem statements and due to high levels of integration and consequent increase in
network model. Section 4 presents the proposed fuzzy and power dissipation of chip.
spider optimization. Simulation results and the implemen- The wireless link between the sensor network and higher
tation are given in Sect. 5. Sect. 6 depicts the conclusion. network traffic densities leads to temperature hotspots in net-
work. Therefore, the wireless links result in performance
degradation in terms of reduced number of packets received
2 Related works and RSS. Consequently, the location and the utilization of
sensor nodes are considerable factors for thermal optimiza-
Several algorithms have been proposed to accomplish the tion in WSN.
clustering task. Lee and Cheng (2012) have proposed a fuzzy Bannister et al. (2008) have expressed a relation between
based clustering approach with energy prediction to increase RSSI and temperature as follows:
the lifetime and fuzzy rules are applied for evenly distribut- ( [ ] [ ])
ing the work load to cluster nodes. Taheri et al. (2012) have RSSIloss[dBm] = 0.1996 T ◦ C −25 ◦ C (1)
proposed an energy aware clustering protocol using fuzzy
where Received Signal Strength (RSS) is inversely propor-
logic called ECPF and Fuzzy cost is computed by both node
tional to temperature and it can vary up to 8 dBm. As such
degree and centrality of node. Then, the fuzzy cost is applied
signal strength threshold satisfies necessary Packet Recep-
to the node to become a CH. Nokhanji et al. (2015) have pre-
tion Rate (PRR) and it can be varied up to 8 dBm.
sented a fuzzy logic by considering two factors for CH selec-
An architectural innovation in connection with suitable
tion. Firstly, RE of nodes has been considered for selection.
power and thermal management strategies is the key for
Then, node degree and centrality are utilized to choose clus-
designing high performance yet energy-efficient massive
ter head. Jain and Ramana Reddy (2015) have proposed a
multicore chips.
novel approach for cluster size control based on Eigenvector

13
Implementation of thermal aware wireless sensor network clustering algorithm based on fuzzy…

From the survey, it is clear that many wireless sen- 3.2 Eigenvalue centrality
sor network design techniques have been proposed so
far, and many of them consider distance between nodes, Eigenvalue centrality is an extended version of degree cen-
deployment, energy consumption, delay, throughput and trality. It finds dominant eigen vector of matrix linked with
number of dead nodes. Some works are related to thermal adjacent matrix of target graph. By calculating the number
aware routing in wireless body area sensor network. To of IN and OUT connections of a node in graph, centrality
the best of the researcher’s knowledge, no technique is values are assigned and the relevance of the node is found
available for a Network Thermal Aware based on cluster- based on Markov chain model.
ing. Hence, in the proposed work, it is attempted to apply A PageRank value of the respective node PR (­ vi) is given
clustering techniques by considering Thermal Heat rise in Eq. (5):
of clusters as a parameter and it is implemented in test ( )
(1 − 𝜕) ∑ PR(vj)
bed for validation. PR(vi ) =
|V|
+𝜕
outdegreev(vj) (5)
(vi,vj)∈E

3.1 Network model where V is the total number of nodes and 𝜕 is the Damping
factor which is responsible for attaining a periodic irreduc-
In this approach, a non-uniform operation of n sensor ible chain.
nodes over 2-dimensional region has been considered.
The WSN model (Jain and Ramana Reddy 2015) has been
considered as a Fuzzy Edge Graph G (V,E,R) where ‘V’ 3.3 α‑cut set
indicates set of nodes. Next, ‘E’ is the set of edges which
is connected to the nodes vi and vj such that ­vj ∊ V, vj ≠ vi. Consider X as fuzzy set and { X𝛼 as a crisp}set which is also
Where vj represents the nodes that lie within the transmis- defined as α-cut, i.e., X𝛼 = X|𝜇x (x) >= 𝛼 . Any element of
sion range r(vi) ∊ R of node vi. x ∊ X𝛼 with a degree of membership is greater than or equal
(i.e.)., to α. For a given fuzzy sets of X and Y, α-cut sets of X𝛼 and
{( ) ( )} Y𝛼 follow the below properties:
E = vi , vj ∈ V∗v ∶ dvi , vj <= r vi (2) ( )
X̃ ∪ Ỹ ∝ = X𝛼 ∪ Y𝛼
where, ‘R’ is the communication range value sets that are
assigned to every node in V and they also depend upon ( )
X̃ ∩ Ỹ ∝ = X𝛼 ∩ Y𝛼
the set of different power levels (PП) which are available
with the transmitter of a node such that PΠ → r.R(vi). ( )
These are bound by upper and lower limit i.e. rmin and rmax, X̃ ≠ X𝛼 except ∝= 0.5
respectively. Both rmax and rmin depend upon the available ∝

maximum and minimum transmission of power levels with

∼
respect to transmitter of node ­(rcr). Hence, ­rcr is defined as X0 = X
the maximum of the nearest node distances and is given
below: Depending on the fuzzy membership value 𝜇(vi , vj ) , a
{ node vi can classify its neighbor nodes i.e., vi∊Ni into vari-
{ }}
rcr (V) = max min
vi ∈V vj ∈V,vj ≠vi
dvi , vj (3) ous groups using α-cut sets, respectively.

By considering Communication Losses above the

Communication Distance (­ d crossover), Maximum cluster
radius is defined for a cluster-head node as: 4 Proposed method
√
max
4𝜋 Lhr ht
(4) In brief, a technique for initial cluster formation and size
rc ≤ dcrossover
𝜆 controlling has been proposed by dividing sensor networks
into several fuzzy subsets according to node’s location
where L is the loss factor of system, hr is the receiving
before the communication process as used in (Jain and
antenna’s height than above ground, ht is the transmitting
Ramana Reddy 2015). At each round, each node in clusters
antenna’s height than the above ground, and λ is the carrier
is assigned to subset. Then, SOA is used to CH selection
signal wavelength.
and message is sent to base station which reduces computa-
tion time.

13
 R. K. Priya, S. Venkatanarayanan

4.1 Fuzzy clustering and size control using Eigen The algorithm of initialization based on Fuzzy Eigen cen-
centrality and α cut set trality clustering is given as Algorithm 1 in detail. Accord-
ing to algorithm 1, an initialise set up phase of cluster Head
The most important characteristic of Eigen centrality based formation is done, at iteration k, and every node vi creates
cluster is that it offers all assistances by correlated cluster- its own neighbour table N(vi)at the communication rangercmax.
ing methods while dealing with their restrictions in respect The neighbour table has the list of nearby nodes along with
of prefixing node exploitation, sink arrangement and spatial all the individual membership values µ(vi, ­vj). Each and every
preparation of network. In other hand, this technique improves node of vi attains ∝-cut sets and degree of nodes C∝ (vi) with
the applicability of linked clustering selection control methods different values of ∝ and also computes the corresponding
by making them free from node exploitation, location of sink PageRank value PR(vi). Consequently, depending upon the
arrangement and spatial preparation of network group. Unlike clustering method, prospective cluster-heads are selected by
the solutions that are presented in literature, the proposed using the proposed Spider Optimisation Algorithm (SOA).
approach considers both the problems related to non-uniform Depending on No and the PageRank value PR(vi), these SOA
node level and relay load. Also, this method does not need a chosen cluster-heads calculate the desired number of nodes Ni.
huge amount of initial communication at various power levels The desired node degree Ni provides the least value of ∝ for
for updating neighbour tables. It might have been achieved which the number of nearby nodes is similar to or better than
by including fuzzy logic approach with a -cut set, which pro- the desired node degree Ni. The value of ∝ is mapped with
vides the additional advantage of simpler and faster result. its own individual power level ( P∝) opted by SOA cluster-
This fuzzy method is used in the initial stage to partition the heads for the upcoming communications. The normal nodes
sensor nodes into a multiple groups by fuzzy clustering model are joined towards the nearest CH and the nodes, which do not
according to node arrangement before the process to commu- drop under communication range of any cluster-head, set their
nicate. In this Eigen centrality based cluster, the clustering communication range to its critical range.
communication range capacity is controlled with lower jump
and upper jump.

13
Implementation of thermal aware wireless sensor network clustering algorithm based on fuzzy…

4.2 Design of objective function The total threshold is found by giving equal weights to
the two components.
The objective function is mainly used to control the total
T = 0.5*t1 + 0.5*t2
energy consumption and the total temperature rise which are
caused by communication between nodes. Therefore, four ∑k ( ∑N )
main factors have been taken into consideration. Ttotal =
i=1=0
i
TCH +
j=1
ij
Tmem (10)

4.3 Average distance from member node to CH All the data given above have difference in dimension and
magnitude. For synthesis, a normalization function is used
The entire network is divided into a group of K clusters with to convert the above values to comparable level.
­Ni member nodes inside, i = 1,2,…,K, and for K cluster head 2
nodes, the max average distance is defined as Eq. (6) y= tanh(x) (11)
𝜋
⎧ ∑ Ni ⎫ The objective function is defined as Eq. (12)
(CMij , CHij ) ⎪
max ⎪ j=1
dist1 = i=1,2,…,K ⎨ ⎬ (6) F = 𝛼dist1 + 𝛽dist2 + Etotal + 𝜎Ttotal (12)
⎪ Ni ⎪
⎩ ⎭
The above factors are normalized to the maximum aver-
Maximum distance from CH nodes to the BS age values. The positive factor values of α, β, γ and σ deter-
mine the priority weight of distance from node to CH, dis-
dist2 = i=1,2,…,K max {d(CH i , BS)} tance from CH to BS, total energy consumption of network
and total temperature rise, respectively with 𝛼 + 𝛽 + +𝜎 = 1.

4.4 Overall network energy consumption

4.6 CH Selection method using spider optimisation
The whole network energy consumption is defined as in Eq.
algorithm (SOA)
(7)
Several swarm intelligence algorithms have been used in the
∑
K
∑
Ni previous decades. Among all the algorithms, PSO and ACO
(7)
i ij
Esum = (ECH + Emem ) are the two algorithms widely studied. The Spider Optimisa-
i=1 j=1 tion Algorithm (SOA) may also be considered as a swarm
where K denotes the number of cluster head nodes, ECH i
is intelligence algorithm, but it is not like PSO and ACO and
the energy consumption of cluster head i during a round and it is elaborated below. The PSO, like SOA, has been initially
Emem is the energy consumption of a member node in cluster
ij proposed for solving general optimization problems. In this
i during a round. paper, Spider Optimisation Algorithm (SOA) has been used
to select an efficient CH with the consideration of Temperature
and the distance between the nodes in each clustering group.
4.5 Overall network temperature rise
The algorithm aims at forming well-balanced clusters so that,
the load is balanced in the wireless network.
The temperature of node is defined as the measure of total
number of packets routed in communication and hence, load
4.7 Problem model for CH selection
is handled. The temperature threshold depends on two com-
ponents (i.e.)., component ‘t1’ which, represents a function
The energy consumption and temperature rise of CH node are
of average temperature of neighbouring nodes and compo-
greater than other nodes in a cluster. Therefore, a cautious con-
nent ‘t2’ defines a function of nodes with own temperature.
sideration is given to every node in the network before select-
√ ing CHs to avoid dead nodes. In this paper, SOA has been
t1 = k1* (avg − n) (8)
applied with inertia weight for the selection of CH nodes. By
√ solving the issues of clustering problem, SOA has been used to
t2 = k2* (temp − n) (9) find a CH by solving fitness function by updating vibration and
best solution. The process of SOA is presented in algorithm 2.
where k1 and k2 are the constant values that are set by
experiments, (temp-n) represents the temperature of node
and (avg-n) represents an average temperature of neighbour-
ing nodes.

13
 R. K. Priya, S. Venkatanarayanan

4.8 Basic parameter of SOA Table 1  Simulation parameters

Parameter Value
The spiders are the main agents of the proposed SOA and they
perform this optimization algorithm in this section. At the very Number of nodes 100
beginning of the algorithm, pre-defined numbers of spiders Distributed area 500*500
are positioned on the web. Every spider holds a memory, and Initial energy of sensor node (Ei) 2J
stores the individual information in the following: Transmission energy (Eelec) 50 nJ/bit
Amp energy (Eamp) 10 Pj/bit/m2
• The position of ‘sp’ on the web. Sensing energy (Es) 5 Nj
• The fitness of the present position of ‘sp’. Packet length (l) 4000 bit
• The target vibration of sp in the previous iterative pro- Max transmission range (r) 150 m
cess. Power level 5
• The number of iterations from the time, when ‘sp’ has t1 0.05
last changed its target vibration. t2 0.05
• The progress that ‘sp’ performed in the past iteration. T (initial) 1 unit
• The dimension mask that ‘sp’ is employed to guide pro- α (factors in objective function) 0.1
gress in the past iteration. β (factors in objective function) 0.1
γ (factors in objective function) 0.3
In this SOA, an algorithm starts with the initialization σ (factors in objective function) 0.5
phase, and then performs a searching in an iterative manner.
Finally, it terminates the algorithm when the corresponding
output solutions are found. The SOA initially evaluates the
fitness values of each and every artificial spider with their position of all the vibrations and also its attenuation. In this
different positions on the web. Then, it updates the global algorithm, a ‘Vib’ is used to represent a |pop| vibration.
optimum value, if possible. The fitness values are calculated Upon the receipt of ‘Vib’, ‘sp’ will choose the strongest and
once for each and every spider, during each iterative process. the best vibration (­ VibB) from ‘Vib’. It also compares its best
Then, all these spiders will generate a vibration at their posi- intensity with the intensity of the target vibration (­ VibT)
tions by using a formula stored in its memory. Therefore, ‘sp’ will store V ­ ibB as V­ ibT
i.e., where ‘sp’, if the intensity of V­ ibB is greater, and C ­ sp is the
( ) number of iterations, since ‘sp’ has last changed its targeted
( ) 1 vibration, is reset to null; else, the original ­VibT is retained
I Psp , Psp,t = log ( ) +1 (13)
f Psp − C and ­Csp is incremented by one. Here, P ­ svib is used to repre-
sent the source positions of Vib and PsT to represent the
After all the vibrations are achieved by using the above source positions of ­VibT, respectively. After the dimension
equation, the algorithm simulates the propagation steps of mask ­(msp) is found, a final following position of spider Psp
fo

these vibrations using the below equation is evaluated based on the mask for sp. The value of i-th
dimension of the following position Psp,i, is generated by the
fo
( ( ))
( ) ( ) D Pa , Pb
given Eq. (15).
I Pa , Pb,t = I Pa , Pa,t × exp (14)
𝜎Xra {
fo
( ) fo Psp,i ms,i=0
where D Pa , Pb = |||Pa − Pb || | Psp,i = (15)
Prsp,i ms,i=1
From the above, every spider ‘sp’ will receive |pop| dif-
ferent kinds of vibrations generated by alternate spiders,
where pop is the population of spider. The output of the
received information of all the vibrations includes the source

13
Implementation of thermal aware wireless sensor network clustering algorithm based on fuzzy…

When it reaches the set number of iteration, a WSN runs where l denotes number of bit per packet and d represents
into the steady-state operation stage until, the network is dead. distance from sender to receiver node. εfs and εamp denote
energy consumption factor of amplification for free space
5 Simulation and multi path models and ­Eelec denotes per bit energy
dissipation
To validate the clustering, fuzzy and spider optimized cluster The temperature of node is defined as the measure of
head selection are simulated in MATLAB. The simulation total number of packets routed in communication and hence,
parameters used in the experiments are listed in Table 1. For load is handled. In common, the node is considered as dead
simulation, an energy model is used as transmission and the node, when its remaining energy reaches zero. When the
receiving cost is characterized by the following Eq. 16. number of dead node rises above threshold, the network is
considered as dead.
l∗Etx +l∗𝜀 ∗d2 ,d<do
ETx (l, d) = {l∗Eelec fs
} For simulation, it has been assumed that for each
tx
transmission and reception of packets, the temperature is
+l∗𝜀 4 ,d≤d
elec famp ∗d o

increased by 0.1 units. The heat generated in every node

Rx
ERx (l) = l ∗ Eelec (16) gets dissipated with time and it leads to drop in tempera-
ture. For stable operation of wireless sensor network, the

13
 R. K. Priya, S. Venkatanarayanan

Fig. 1  Node mortality versus X axis-Mortality %

rounds with BS at center of y axis-Rounds
network
500

400

300

200

100

0
1% 10% 20% 30% 40% 50%

LEACH FCPSO PROPOSED

Fig. 2  Node mortality versus X axis-Mortality %

rounds with BS at edges of y axis-Rounds
network
140
120
100
80
60
40
20
0
1% 10% 20% 30% 40% 50%

LEACH FCPSO PROPOSED

Fig. 3  Number of node alive X axis-Rounds

versus rounds BS at centre of Y axis-Number of node alive
network
120

100

80

60

40

20

0
0 200 400 600 800 1000 1200 1400

LEACH FCPSO PROPOSED

13
Implementation of thermal aware wireless sensor network clustering algorithm based on fuzzy…

Fig. 4  Number of node alive X axis-Rounds

versus rounds BS at edge of Y axis-Number of node alive
network
120
100
80
60
40
20
0
0 100 200 300 400 500 600

LEACH FCPSO PROPOSED

Fig. 5  Total data message x axis-Rounds

received versus rounds BS at y axis-Total data message received/bit x 10^7
center of network
25

20

15

10

0
0 200 400 600 800 1000 1200

LEACH FCPSO PROPOSED

Fig. 6  Total data message x axis-Rounds

received versus rounds BS at y axis-Total data message received/bit x 10^7
edge of network
10

0
0r 100r 200r 300r 400r 500r 600r

LEACH FCPSO PROPOSED

cooling rate of the system is also assumed to 0.1 units per with standard LEACH and Fuzzy Clustering Particle Swarm
simulation interval. In this experiment, k1 and k2 are set to Optimization algorithm (FCPSO).
be 0.05. To verify the effectiveness of the proposed tech- Because of the significant importance of BS location,
nique, parameters like number of nodes alive, total data the BS location is varied by placing centre or edges of the
message received and mortality percentage are compared network area.

13
 R. K. Priya, S. Venkatanarayanan

5.1 Test bed results

Figure 7 shows the experimental setup. The hardware is

implemented using MATLAB, Zigbee and PIC micro-
controller with temperature measuring unit, respectively.
WSN consists of 10 member nodes, 1 cluster head and
1 static base station in an outdoor environment with the
area of 100 m2. The nodes are placed randomly with equal
initial energy of 750 J at specific interval communication
established and continued for the entire period. Using
XCTU, the RSSI of the CH along with its energy con-
sumption and temperature is measured for every half an
hour at each round.
Table 2 shows the number of rounds, corresponding CH
node temperature, RSSI values for randomly varying node
distance and reclustering with and without temperature
Fig. 7  Test bed implementation consideration. From the table, it is observed that the imple-
mented algorithm chooses CH with the consideration of both
remaining energy and temperature rise of the node. It will
From Figs. 1 and 2, it is clear that the death time of the also create a impact by selecting higher RSSI node as CH.
node is prolonged with the proposed algorithm. The main
reason for this is that the proposed method selects cluster
head, which reduces energy consumption of node communi-
6 Conclusion
cation and at the same time, nodes with the highest remain-
ing energy and the lowest temperature unit have the prior-
In this work, a temperature aware clustering algorithm based
ity to be a cluster head, Balancing energy consumption and
on eigen centraily clustering and spider optimization has
temperature profile improve a nodes life time. The Figs. 3
been proposed for smart grid wireless sensor networks.
and 4 show very mild dying process of node because of the
In fuzzy clustering, a Eigenvector centrality (PageRank)
consideration of energy and distance from the base station.
and α-cut set are used for computing communication load
Figures 5 and 6 show total message received at base sta-
and controlling cluster size. In SOA, the fitness function is
tion. The results clearly show that the proposed method
designed for the factors like temperature rise in clusters,
effectively selects cluster head and delivers more messages
distance from the node within the cluster to the CH node
than LEACH and FCPSO. The proposed algorithm chooses
as well as the distance from the CH node to the base sta-
heads considering energy, distance and temperature of the
tion. The proposed work has been compared with the well-
node also. Minimum temperature node possesses good
established CH selection schemes. The results show that the
received signal strength and lower energy dissipation which
proposed algorithm can significantly reduce the mortality
ensure more data delivered to base station.
rate of nodes and extend the network lifetime. In future, the
The Fig. 3 and 6 show better improvement in total number
temperature impact on node, due to environmental factors
of data packets received and number of nodes alive because
may also be considered
of neglecting high temperature unit node as a cluster head
above the threshold.

Table 2  RSSI and Temperature values for selected CH with and without temperature consideration
Rounds Energy remaining Energy remaining Temperature of Temperature of corre- RSSI of correspond- RSSI of cor-
of CH (proposed) of CH (Fuzzy) in J corresponding CH sponding CH (fuzzy) ing CH (proposed) responding CH
in J (proposed) in ◦C in ◦C in dB (fuzzy) in dB

0 750 750 30 30 − 56 − 56
200 745.1 744.2 32.5 33.1 − 55 − 58
400 735.9 736.6 32.8 34 − 56 − 61
600 730.3 728.1 32 36.7 − 57 − 64
800 718.2 715.9 34 39.33 − 59 − 67

13
Implementation of thermal aware wireless sensor network clustering algorithm based on fuzzy…

References Jain A, Ramana Reddy BV (2015) Eigenvector centrality based clus-

ter size control in randomly deployed wireless sensor networks.
Expert Syst Appl 42(5):2657–2669
AbdelSalam AS, Olariu S (2012) BEES bioinspired backbone E selec-
Jiang C, Shi W, Xiang M (2010) Energy-balanced unequal clustering
tion in wireless sensor networks. IEEE Transact Parallel Distrib
routing protocol for wireless sensor networks. J China Univ Posts
Syst 23(1):44–51
Telecommun 17(4):94–99
Al-Anbagi I, Erol-Kantarci M, Mouftah HT (2013) QoS-aware
Jin Y, Wang L, Kim Y, Yang X (2008) Eemc: an energy-efficient
inter-cluster head scheduling in WSNs for high data rate smart
multi-level clustering algorithm for large-scale wireless sensor
grid applications. In: IEEE global communications conference
networks. Comput Netw 52(3):542–562
(GLOBECOM)
Lee J, Cheng W (2012) Fuzzy-logic-based clustering approach for
Arora VK, Sharma V, Sachdeva M (2019) ACO optimized self-organ-
wireless sensor networks using energy predication. IEEE Sens
ized tree-based energy balance algorithm for wireless sensor net-
J 12(9):2891–2897
work. J Ambient Intell Human Comput 10:4963–4975
Lindsey S, Raghavendra C (2002) Pegasis: power-efficient gathering
Bag A, Bassiouni MA (2008) Hotspot preventing routing algorithm
in sensor information systems. In: Aerospace conference proceed-
for delay-sensitive applications of in vivo biomedical sensor net-
ings, IEEE, vol 3, 2. pp 3–1125–3–1130
works. Inf Fusion 9:389–398
Manjeshwar A, Agrawal DP (2001) Teen: a routing protocol for
Bannister K, Giorgetti G, Gupta SK (2008) Wireless sensor networking
enhanced efficiency in wireless sensor networks, in parallel and
for “hot” applications: effects of temperature on signal strength,
distributed processing symposium. In: Proceedings 15th Interna-
data collection and localization. In: Proceedings of the ACM 15th
tional, pp 2009–2015
workshop on embedded networked sensors, Charlottesville, 2–3
Monowar MM, Bajaber F (2015) On designing thermal-aware localized
June
QoS routing protocol for in-vivo sensor nodes in wireless body
Boanoy CA, Wennerströmx H, Zúñiga MA (2013) Hot packets: a sys-
area networks. Sensors 5:14036–14044
tematic evaluation of the effect of temperature on low power wire-
Ni Q, Pan Q, Du H (2015) A novel cluster head selection algo-
less transceivers. In: 5th extreme conference on communication
rithm based on fuzzy clustering and particle swarm optimiza-
(ExtremeCom).
tion. In: IEEE/ACM transactions on computational biology and
Buttyan L, Schaffer P (2007) Panel: Position-based aggregato rnode
bioinformatics
election in wireless sensor networks. In: 2007 IEEE Inter-national
Nokhanji N, Hanapi ZM, Subramaniam S, Mohamed MA (2015) An
conference on mobile adhoc and sensor systems, pp 1–9
energy aware distributed clustering algorithm using fuzzy logic
Cai X, Duan Y, He Y (2015) Bee-Sensor-C: an energy-efficient and
for wireless sensor networks with non-uniform node distribution.
scalable multipath routing protocol for wireless sensor networks.
Wirel Personal Commun 84:395–419
Int J Distrib Sens Netw
Singh M, Soni SK (2019) Fuzzy based novel clustering technique by
Ebrahimi MS, Daraei MH, Behzadan V, Khajooeizadeh A (2011) A
exploiting spatial correlation in wireless sensor network. J Ambi-
novel utilization of cluster-tree wireless sensor networks for situ-
ent Intell Human Comput 10:1361–1378
ation awareness in smart grids. In: IEEE PES Innovative Smart
Taheri H, Neamatollahi P, Younis O, Naghibzadeh S, Yaghmaee M
Grid Technologies, pp 13–16
(2012) An energy-aware distributed clustering protocol in wireless
Elhabyan RS, Yagoub MCE (2014) PSO-HC: particle swarm optimi-
sensor networks using fuzzy logic. Ad Hoc Netw 10:1469–1481
zation protocol for hierarchical clustering in wireless sensor net-
Ye M, Li C, Chen G, Wu J (2005) Eecs: an energy efficient cluster-
works. In: 10th IEEE international conference on collaborative
ing scheme in wireless sensor networks in PCCC 2005. In: 24th
computing: networking, applications and worksharing
IEEE international performance, computing, and communications
Heinzelman W, Chandrakasan A, Balakrishnan H (2000) Energy-effi-
conference, April 2005, pp 535–540
cient communication protocol for wireless microsensor networks,
Younis O, Fahmy S (2004) Heed: a hybrid, energy-efficient, distributed
in System Sciences, 2000. In: Proceedings of the 3­ 3rd Annual
clustering approach for ad hoc sensor networks. IEEE Transact
Hawaii International Conference on, vol 2. p 10
Mob Comput 3(4):366–379
Ho CK, Ewe HT (2005) Performance of an ant colony optimization
(ACO) algorithm on the dynamic load-balanced clustering prob-
Publisher’s Note Springer Nature remains neutral with regard to
lem in ad hoc networks. In: International conference on compu-
jurisdictional claims in published maps and institutional affiliations.
tational and information science CIS: computational intelligence
and security, pp 622–629
Hong J, KookJ LS, Kwon D, Yi S (2008) T-leach: the method of thresh-
old-based cluster head replacement for wireless sensor networks.
Inf Syst Front 11(5):513–521

13