Accepted Manuscript

Title: An optimal clustering mechanism based on Fuzzy-C

means for wireless sensor networks

Authors: Shengchao Su, Shuguang Zhao

PII: S2210-5379(17)30170-1
DOI: http://dx.doi.org/10.1016/j.suscom.2017.08.001
Reference: SUSCOM 178

To appear in:

Received date: 8-5-2017

Revised date: 18-7-2017
Accepted date: 8-8-2017

Please cite this article as: Shengchao Su, Shuguang Zhao, An optimal clustering
mechanism based on Fuzzy-C means for wireless sensor networks, Sustainable
Computing: Informatics and Systemshttp://dx.doi.org/10.1016/j.suscom.2017.08.001

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
 An optimal clustering mechanism based on Fuzzy-C means for
wireless sensor networks
Shengchao Su1,2,*, Shuguang Zhao1
1 College of Information Science and Technology, Donghua University, Shanghai 201620, China
2
Laboratory of Intelligent Control and Robotics, Shanghai University of Engineering Science,
Shanghai 201620, China

Highlights
• An optimal clustering mechanism has been proposed to balance the node’s energy
consumption and prolong the lifetime under energy-constrained wireless sensor
networks.
• The optimal number of cluster heads is estimated based on node’s density.
• Definition of the objective function based on the distance and the weight presented.
• An improved Fuzzy-C means clustering algorithm is proposed to divide the sensor
nodes into a specified number of clusters.
• Simulation and performance evaluation provided for the proposed mechanism.

Abstract: In order to balance the node’s energy consumption and extend the lifetime under
energy-constrained wireless sensor networks, an energy-efficient clustering algorithm based on
Fuzzy-C means for wireless sensor networks is proposed. Taking into account the uneven
distribution of the sensor nodes and the uncertainty of the radio channel, the cluster formation
process of nodes is modeled as a fuzzy partition of sample space in this paper. Firstly, the overall
energy consumption of the networks is analyzed, and the optimal number of cluster heads is
estimated based on node’s density. Secondly, in the design of the objective function, the distance
from the node to the cluster head and the weight of the membership values are considered. Then,
the improved Fuzzy-C means clustering algorithm is proposed to divide the sensor nodes into a
specified number of clusters. Finally, a single hop communication mode is used for intra cluster
communication, and inter cluster communication adopts a multi-hop communication mode. The
simulation results show that the proposed algorithm can obtain uniform spatial distribution of
cluster heads and balance the energy consumption of network effectively.
Keywords: wireless sensor networks; Fuzzy-C means; multi-hop; routing transmission
mechanism; energy-efficient

1. Introduction
By deployment of various micro sensors, wireless sensor networks (WSNs) can perceive,
collect and process the information of the area, and then send it to the observer. As a new type of
autonomous self-organized network, it forwards the sensed information to the destination terminal
in single or multi hop manner, which connects the logical world and the physical world effectively
[1]. In general, the number of nodes in WSNs is huge and the resources are severely limited. In
order to reduce the energy consumption and extend the life of the network, it is very important to

1
design routing algorithm in network layer [2]. Due to advantages of scalability and high energy
efficiency, clustering route schemes become a hot spot in the fields of communication mode in
WSNs. Typical clustering routing algorithms, including LEACH[3], HEED[4], Ring Routing[5],
PHC[6] and EBCS[7], etc. solve the problem of energy consumption from different aspects based
on the characteristics of WSNs.
Since the WSNs are composed of plenty of sensor nodes, the random mode exploited in the
deployment will lead to the different density of nodes in the monitoring area inevitably. The above
algorithms seldom consider the topological structure of the practical application of WSNs. Mostly,
the logical structure of clusters is inconsistent with the actual distribution of nodes, and it results
in the increase of energy consumption of network communication [8]. In addition, WSNs are easy
to be affected by the environmental factors, such as signal attenuation and noise interference in
wireless communication. However, most of the proposed algorithms do not take into account the
uncertainty of optimal clustering and lack of robustness owing to strict and hard partition method
[9].
To solve the above problems, an optimal clustering mechanism based on Fuzzy-C means for
WSNs (OCM-FCM) is proposed in this paper. The clustering process of nodes is modeled as a
fuzzy clustering problem in the sample space, and the improved Fuzzy-C means clustering
algorithm is designed to divide the sensor nodes into different clusters.
The remainder of the paper is organized as follows: Section 2 conducts a brief survey of
some of the clustering algorithms with their pros and cons. In Section 3, the system model is
introduces. Section 4 provides the details of the proposed algorithm. Then, Section 5 presents the
simulation results, and Section 6 concludes the paper.

2. Related work
Due to the limited energy of the sensor nodes, it is necessary to reduce the frequent remote
transmission, so as to prolong the network lifetime. Therefore, direct communication between
nodes and base stations is not encouraged. An effective method is to divide the network nodes into
different clusters, where the cluster head will be selected in each cluster and is responsible to
transmit the data of the entire cluster to the base station [10]. Therefore, only a small number of
nodes need to transmit data remotely, and other nodes will carry out short distance transmission to
save energy consumption.
A Game-theoretic approach for efficient clustering in wireless sensor networks (GAEC) is
proposed [11], and the basic idea is that the Selection of cluster heads is based on game theory and
set a slavery cluster header to avoid the impact of failure of the original cluster header. In the
process of clustering, the round-robin mode is not applied. Once the cluster head does not work
normally, candidate cluster heads need to be reset and it will cause the excessive energy
consumption of cluster heads participating in the redeployment.
A inter-cluster multi-hop routing protocol improved based on LEACH (CMRAOL) is
presented in [12]. The residual energy and the distance between the node and the base station are
considered to apply to the threshold function of cluster head election, and the multi-hop routing
transmission mechanism is applied for inter-cluster communication. In [13], an energy efficient
distributed uneven clustering routing algorithm for WSNs is proposed. The cluster heads are
elected by the timing broadcast mechanism to reduce the consumption of control message
exchange, and the cluster head can transmit data to the base station according to the social welfare

2
function. In [14], the compressed sensing algorithm is introduced to reduce the amount of data
transmitted by the node, and the data collected from the cluster head is employed to reduce the
redundant information. In [15], a kind of multi-level heterogeneous network topology control
algorithm is proposed, which design a method to construct a reverse connected dominating set tree
to optimize the network topology. In [16], a stable election protocol for clustered heterogeneous
wireless sensor networks (SEP) in presented, where Two kinds of cluster head election
probabilities are put forward to construct bio-level heterogeneous clusters in WSNs. It can
postpone the death time of the first node, but does not apply in a heterogeneous WSNs
environment. In [17], a distributed clustering algorithm for heterogeneous networks is proposed.
Since the initial energy of the sensors is different, which results in the diverse in terms of the
competition probability of the nodes being cluster head; also, the probability will vary with the
node’s residual energy over time.
In [18], a novel data transmission mechanism based on unequal clustering (DTUC) is
presented to solve the hot spot problem, which arises from the multi-hop data transmission mode.
Multiple layers are divided and some clusters are formed base on the distance and the probability
of nodes closer to the sink node or not. In [19], a clustering algorithm based on Parameter
Optimization of the network is proposed. According to the different distance from the base station,
the cluster size varies adaptively to ensure the balance of energy consumption among different
clusters. By controlling the relevant optimization parameters and the cluster scale, the cluster head
serves as the local control center continuously formulated by its remaining energy and location
information to reduce the frequency of updating cluster head, which can reduce the energy
consumption of the communication in the cluster effectively.
Aiming at the problem of energy saving and QoS perception, an energy-efficient and
perceived QoS aware video routing over wireless multimedia sensor networks (PEMuR) is
proposed[20], which combines the hierarchical routing protocol and intelligent energy aware
packet scheduling algorithm to achieve the network load balance. In [21], a clustering strategy
based on genetic algorithm is proposed, and based on the maximum inter cluster communication
distance and the maximum number of inter cluster communication, a routing strategy based on
genetic algorithm is proposed, and the key factors of maximum inter cluster communication
distance and the maximum number of inter cluster communication are taken into account of the
process of crossover and mutation. In [22], a routing algorithm is proposed by a combination of
efficient particle encoding scheme with multi-objective fitness function to resolve the optimization
problems about energy efficient clustering and routing. Although the above algorithm can
optimize the network topology structure and prolong the working time of the whole network to a
certain extent, the optimization performance will be affected by the selection of the objective
function or other factors with the increase of the number of nodes in the network.
Clustering wireless sensor network is a hierarchical structure [23], in which the cluster heads
collect the measurement information from the member nodes and then send the aggregated data
directly to the base station. Most of routing algorithms in WSNs focus on the hierarchical
clustering structure, and the clustering mechanism is very effective to reduce energy loss [24]. In
this paper, the cluster formation process of nodes is modeled as a fuzzy partition of sample space.
The optimal number of cluster heads is estimated based on node’s density, and the objective
function focuses on the distance from each member node to the CH and weighted sum of
membership value. As with other algorithms, it is impossible for the proposed approach to succeed

3
in all scenarios. However, it can at least provide an alternative and complementary solution to
some problems in which other available clustering techniques may fail.

3. System model

3.1 Network model

The principle of WSNs clustering routing mechanism is that the sensor nodes are divided into
several clusters, where the elected cluster head is responsible for managing the member nodes to
work collaboratively and then forward the fusion result to other cluster heads or base station.
In our analysis, the N sensor nodes are randomly and uniformly deployed to the circular
area with the radius R , where there exists only one base station situated in the central point.
Furthermore, the assumptions about the sensor nodes are as following:
(1) The node can obtain the position information of the system by using the GPS positioning
system or the positioning algorithm;
(2) After deployment, all sensor nodes are static, and the base station is far from the
monitoring area;
(3) The initial energy of the nodes is equal;
(4) The wireless communication between nodes is symmetrical.
The Clustering routing method was originally put forward by LEACH, and it affects most
clustering routing algorithms later. To prolong the network lifetime, random cluster head selection
is introduced in LEACH. Obviously, due to the neglect of the actual geographical location of the
node distribution, this randomness can easily lead to two problems: First, the number of cluster
heads (CHs) generated per round is uncertain, and the rounds as the number of CHs being selected
is equal to the optimum value is less than 20%; Secondly, the position of CHs generated in each
round is not uniform, CHs are too dense in some regions, and few in some areas conversely. Those
defects can bring about that the node’s energy consumption is not balanced and some nodes will
soon consume energy, thus speeding up the whole network of death.
The clustering process in WSNs and pattern clustering analysis has a great similarity in some
respects. If the WSNs are viewed as a collection of a large number of two-dimensional samples
and each sensor node is one of the samples, the fuzzy subset obtained by fuzzy clustering
corresponds to the cluster, and then the cluster center is equivalent to the CH node. Therefore, the
idea of fuzzy clustering can be applied to the research of clustering algorithm for WSNs. By
taking into account of uncertainty degree of each node being assigned to the corresponding cluster,
the optimized result is more consistent with the real application than the traditional classification
methods.

3.2 Radio energy consumption model

According to the distance between the transmitter and the receiver, the energy consumption
will be measured by radio energy dissipation model [5]. And according to the distance from the
target, the free space and the multi-path fading channel models can be used. The energy
consumption of the transmitting node includes the energy consumption of the transmitting circuit
and the power amplification. Therefore, to transmit a k bits message over a distance d , the
energy consumption can be formulated as:

4
  Eelec * k   fs * k * d , d  d0
 2

ETx (k , d )  ETx elec (k )  TTx amp (k , d )   (1)

 Eelect * k   mp * k * d , d  d0
4


To receive k bits message, the energy consumption will be:

ERx (k )  ERx elec (k )  Eelec * k (2)

where ETx elec , TTx  amp is the energy consumption by receiving and transmitting the message of

each node.  fs ,  mp is power consumption of the free space propagation and power consumption

of multipath propagation respectively. Eelec is the energy/bit consumed by the

transmitter/receiver electronics. d 0 is the threshold value of distance.

4. Proposed clustering mechanism based on Fuzzy-C means

4.1 Estimation of optimum number of CHs

Suppose S is a circular bounded region with a radius R , and the base station is situated in
the center point. There are N sensor nodes evenly distributed in the area. Let k denote the
number of clusters, the average number of sensor nodes is N / k for each cluster, including: one
CH and N / k  1 member nodes. Let ( x, y) represent the coordinate of a CH and dtoBS as the
distance to the base station, the mathematical expectation of the distance can be given as:
1
E[dtoBS ] 
 R2 
S
x 2  y 2 dxdy (3)

Let Ech be the energy consumed by CH, and it consists of three parts as following:

Ech  Ech _ recv  Ech _ merg  Ech _ frwd (4)

Where Ech _ recv is the energy consumption for receiving message from member nodes can be

given by:
N
Ech _ recv  lEelec (  1) (5)
k

In which, Ech _ merg denotes the energy consumed for data fusion. Let EDA be the energy

consumption per bit of data fusion and CH being demanded to collect monitoring data. Hence,
N
assuming that the data are fused to perfect data fusion, then Ech _ merg  lEDA ( ) .
k

In addition, Ech _ frwd is the energy consumption of CH node for data forwarding to the base

station. Since the base station is far away from the cluster, the multipath attenuation model is used
and the energy consumed can be expressed as:

Ech _ frwd  lEelec  l mp dtoBS

4
(6)

To sum up, the energy consumption of each CH can be obtained as:

5
 N N
Ech  lEelec (  1)  lEDA ( )  lEelec  l mp dtoBS
4

k k
(7)
N N
 l *{Eelec (  1)  EDA ( )  Eelec   mp E[dtoBS ]4 }
k k
Let dtoCH be the distance from member node to its CH. Suppose that the CH node is located
at the center of the cluster, mathematical expectation of the distance from the member nodes to the
CH should be obtained:
E[dtoCH ]   x 2  y 2  ( x, y )dxdy (8)
Sch

For the sake of convenience, cluster member nodes are evenly distributed and the node’s

density  ( x, y)  N . Let Enon _ CH be energy consumption for cluster members per frame. Due
 R2

to the short distance between the member nodes and the CH, Enon _ CH can be calculated as:

Enon _ CH  l *{Eelec   fs [ E[dtoCH ]2 ]} (9)

Thus, the energy consumption of a cluster can be expressed as:

N
ESCH  Ech  (  1) Enon _ CH (10)
k
The total energy consumption of k clusters in the network is:

Etotal  k * ESCH  k *{(2N  k ) Eelec  ( N  k ) fs E[dtoCH ]2  NEDA  k mp E[dtoBS ]4 } (11)

By taking the derivative with respect to k in Formula (11), the optimum value of k can be
obtained as:

2 N  fs R 2
kopt  (12)
 mp N  4 Eelec

4.2 Initialization of the centroid of the clusters

Fuzzy C-means (FCM) clustering is a kind of clustering method similar to unsupervised
learning [25]. By using fuzzy theory, the samples are divided into different categories according to
classification method [26]. In FCM, the data set X  {x1 , x2 , , xn } with n samples can be
sorted into k fuzzy sets. Each sample is assigned to a certain class according to certain fuzzy
membership degree, and the center position of each class is calculated [27]. By modifying the
location of the cluster center repeatedly, the function value with a certain similarity index is
minimized. The final classification result is to ensure that the similarity between the samples in the
same class is the largest, and the similarity is to achieve the minimum among the different classes.
During the formation phase of cluster, OCM-FCM adopts Fuzzy C-means method to select

kopt CHs according the analysis result in section 4.1. The whole sensor network is divided into

kopt clusters, and each node will join in a certain cluster. The selection of CH node is depending

on the node’s current energy, and the node with more residual energy will be of higher probability
to be elected as the CH. In the data transmission phase, the single hop mode is adopted for the
intra-cluster communication. As for the inter-cluster communication between the CHs, multi hop

6
manner is relatively favorable choice to reduce energy consumption for long distance
transmission.
The effect of FCM clustering depends on the initial means, which has a great influence on the
final clustering result. Most traditional FCM clustering methods employ random initialization to
determine the centroid of the cluster's assigned points, and it results in the fluctuation of clusters
and producing counterintuitive results. In this paper, the vicinal location with dense nodes is
chosen as the initial cluster center by setting the node’s density parameters. If satisfy the condition
that the number of objects is at least DensCluster within the scope of the field  , the object in
the high density area can be obtained. Here, the parameter of DensCluster and  is defined as:

 DensCluster  N / kopt

 S R (13)
   k  k
 opt opt

And the density function is defined as:

N
1
dist (i)   exp[ xi  x j *(
2
)2 ] (14)
j 1 DensCluster
Thus, the high density point set Sa can be established. The highest density data point can be
taken as the first cluster centroid p1 , and then the point farthest from the first cluster center in set
Sa is selected as second cluster centroid p2 . And so forth, the k -th cluster centroid pk should
be satisfied with:

max{ i 1 ( pk  pi )}
k 1
(15)

To avoid the distance between the node position and the cluster center be zero during the
course of clustering, the initial cluster center should be modified to ensure the clustering iteration.
According to the consideration mentioned above, the center of the nearest node far from
p1 , p2 , , pk is taken as the final initial centroid of cluster.

4.3 Formation of clusters

FCM algorithm is an iterative process, the purpose of which is to minimize the value of the
nkopt
objective function. Let P  [ p1 , p2 ,, pkopt ] denote the Vector of cluster centers, U  [uik ]

be clustering classification matrix, and ik be the Membership degree of sample xi to the

cluster centroid pk , uik [0,1] . Therefore, the objective function of the FCM algorithm is described

as follows:
kopt n
J m (U , P)   ik xi  pk
m
(16)
k 1 i 1

Where m  [1, ) is a constant to control the fuzzy degree of clustering results, and we take the
value of 2.
The clustering criterion is to minimize the value of J m (U , P) . Since the columns in the
classification matrix are independent and the constraint of the extreme values satisfies
n
with  ik  1 , it can be solved by Lagrange multiplier. To make the minimum value of J m (U , P) ,
i 1

ik and pk should be equal to:

7
  1
 ik  kopt d 2


 ( ik ) m 1
j 1 d jk

 n (17)
  ( ik ) m xi
 pk  i 1
 n



 i 1
( ik ) m

CH selection and cluster formation are as follows:

Step 1: Initialization: given the number of CHs kopt , and set the iteration stop threshold  .

Next, Initializing clustering centers P0 , and the iteration counter c can be set to zero;
Step 2: Using the formula (15) to calculate the new classification matrix U ;
Step 3: Using the formula (16) to update the clustering center pi ;

Step 4: if P(c 1)  P   , the algorithm can be terminated and output clustering center P ;

Each node is allocated an unique identity and joins into different clusters. Otherwise, c   and
turn to step 2.

4.4 Communication strategy

After initialization of the network, all sensor nodes are organized into kopt clusters based on

the optimal estimation and proposed Fuzzy-C means mechanism. Each node belongs to a certain
cluster, the weighted sum of the distance between the member nodes and the cluster node
attains the minimum meanwhile. During a complete round, the structure of the clusters will be
fixed, i.e. once each node’s identification is determined, it will not be altered until the next round.
The CH’s selection lies on candidate node's residual value Eres , and the nodes with more energy
will be of greater probability than other nodes. In the first round, the CH node is specified by the
base station. At the end of each round, the current CH will decide the possible alternatives for CH
in the next round. If there are a number of cluster nodes with the same probability, the nodes with
smaller ID will be selected. After the completion of the CH’s selection, the new CH will allocate
the time slots to all member nodes in terms of the size of the cluster, and then broadcast the
message to all nodes. Next, intra cluster communication and inter cluster communication are
discussed in detail.
 Intra cluster communication. Since usually the member nodes
are relatively close to their CH, a single hop communication mode can be used and the
ordinary nodes do not have to keep into the listening state. After transmitting the packets
to CH in assigned TDMA slots, they can turn into sleep state subsequently, which can
save their energy. The CH node needs to monitor the channel and receive the data sent
by its member nodes in the whole round. After receiving the messages from its all
members, the CH conducts data fusion with its own collection and shift into the inter
cluster communication.
 Inter cluster communication. Because the base station usually is far away from the
monitoring area and the distance between the CH and the base station will be relatively
large, the communication process between the CHs and the base station will consume
plenty of energy. The multi-hop communication mode can reduce the energy

8
 consumption effectively in contrast with the single-hop manner in long-distance
transmission. In the stage of cluster formation, the CHs should try their best to broadcast
the message, of which includes their ID and the distance from the base station, to the
surrounding nodes. Then, those messages will be received by adjacent CHs, which will
be stored in their memory and provides the references for the choice of the next hop.
Hence, after the accomplishment of data fusion, the CH find out the adjacent CH closer
to the base station in the neighboring CHs table as the next-hop node. If failed to
determine the appropriate node, it sends the data directly to the base station directly.

5. Simulation & performance evaluation

In this section, we evaluate the performance of our mechanism implemented with MATLAB.
For simplicity, we assume the probability of signal collision and interference in the wireless
channel is ignorable and no packet loss error is occurred. In view of the energy consumption of
data receiving, transmitting and fusion, the total residual energy and the number of nodes in the
network will be statistically analyzed to evaluate the energy efficiency. In each scenario, the
position coordinates of the nodes are randomly generated. To ensure the fairness and correctness,
the experiment conducted 100 times repeatedly and the average value of the simulation results can
be obtained for comparison between the different methods [28, 29]. The simulation parameters are
shown in Table 1.

Firstly, the total energy consumption index is defined as the sum of energy consumed by each
node as the network runs normally. The experiment compares the total energy consumption of
different protocols, and the results are shown in Figure 1. As can be seen from the results, the total
energy consumption of all protocols increases as the time goes. By employing Fuzzy-C means
method to determine the CH and the cluster scale, it can prevent the message flooding effectively.
In addition, the number of clusters in OCM-FCM is related to the optimal value, thus the running
time of the network is relatively long. In ECS-PSO, the reason of rapid expenditure of energy is
that the CH node collects the data in the cluster and then sends it to base station through the direct
transmission mode. In TLCR, the residual energy factor is considered in the selection of CH, and
the multi hop transmission is adopted in the inter cluster transmission. Therefore, the energy
consumption is significantly lower than that of the ECS-PSO. However, the message negotiation
mechanism and the cluster size being determined by heuristic method cause more energy
expenditure than our OCM-FCM mechanism.
Figure 2 shows the comparison of the network lifetime. In general, the round of the First
Node Die (FND) can be used to measure the stability of the network operation. However, while
some nodes are densely deployed in the uneven distribution, it is not comprehensive to use FDT as
test parameters. Therefore, the round of Half of Nodes Die (HND) and Last Node Die (LND) are
certainly worth recording to qualify or to support the reliability of the network in more aspects.
From the experimental results, it can be observed that the FDT in our protocol proposed in this
paper is 7.81% longer than that of ECS-PSO, 19.4% and 27.5% longer than that of TLCR in
aspects of HND and LND respectively.
The node’s average residual energy and variance of residual energy is shown in Figure 3 and
Figure 4. It is worth noting that the greater the mean of the residual energy is and the smaller the
variance is at a certain time, the better the energy balance achieves. From the figure we can

9
observe, both the average residual energy and the variance is large, and the fluctuation of the two
metrics is relatively obvious. It means that the insufficient mechanism of data collection lead to
the node’s load imbalance. In OCM-FCM, the variance is relatively stable and minimum, which
reflects the balance of the node’s energy consumption in wireless sensor networks.
Figure 5 shows the number of CHs in different protocols over time. As can be seen from the
figure, the number of CHs keeps stable before FDT occurs in OCM-FCM. Subsequently, the
number of CHs decreases gradually. However, the declined trend in OCM-FCM does not
demonstrate significantly as in ECS-PSO and TLCR. That is because the number of CHs in
OCM-FCM mainly depends on the node’s density and the transmission range of the node, and the
two factors are relatively fixed. The residual energy of nodes is one of the most important factors
to determine the cluster radius in ECS-PSO and TLCR, and the residual energy of plenty of the
nodes is greatly reduced after the appearance of FDT, the network death rate is accelerated, and
the residual energy of nodes is greatly reduced. Therefore, the number of CHs in ECS-PSO and
TLCR is also greatly influenced.
Figure 6 shows the number of alive nodes over time. It can be seen from the figure,
OCM-FCM protocol demonstrates good stability and the survival time of the active node is
obviously longer than the other protocols. Even if FDT appears in OCM-FCM, the survival
number of nodes is decreased linearly till the node all died, rather than the function of data
collection almost being paralyzed in ECS-PSO and TLCR as a result of excessive node death in
most of rounds. This is mainly due to the use of CH specified mechanism decreases the energy
consumption in the stage of cluster formation more greatly than the campaign and negotiation
mechanism applied in ECS-PSO and TLCR. In addition, the number of clusters in OCM-FCM is
equal to the optimal value so as to minimize the network energy consumption.
Figure 7 shows the distribution of the nodes when 50 percentages nodes died. As can be seen
from the results, the distribution of dead nodes is different due to the respective clustering
methods. In ECS-PSO protocol, the death node is mainly distributed in the area far away from the
base station. That is mainly because the CH forward data to the base station by means of direct
transmission. Since the energy consumption is increasing exponentially with the distance from the
source to the destination, the energy expenditure of the nodes far away from the base station
consumes more badly. Conversely, the dead nodes mostly are situated near the base station as a
result of multi-hop manner, which lead to heavy burden upon those nodes. Although the unequal
clustering method is introduced in TLCR to make the clusters near the base station as possible as
small so as to reduce the load of the related CHs, it can only alleviate the appearance of energy
holes rather than to avoid. Comparatively speaking, the position of dead nodes in OCM-FCM is
randomly distributed in the whole network, so it can avoid the energy hole problem better than the
other protocols

6. Conclusion
Aiming to balance the node’s energy consumption and prolong the lifetime under
energy-constrained wireless sensor networks, an energy-efficient clustering algorithm based on
Fuzzy-C means for wireless sensor networks is proposed. The overall energy consumption of the
whole networks is analyzed, and the optimal number of CHs is estimated based on node’s density.
Then, the distance from the node to the CH and the weight of the membership values are
considered to define the objective function, and the improved Fuzzy-C means clustering algorithm

10
is proposed to divide the sensor nodes into a specified number of clusters. The simulation results
show that the proposed algorithm can obtain uniform spatial distribution of CHs and balance the
energy consumption of network effectively.
In this paper, the objective function focuses on the distance from each member node to the
CH and weighted sum of membership value. In the coming research, we'll conduct intensive study
on the construction of comprehensive objective function of multi criteria, and make the clustering
algorithm fit for the actual WSNs applications perfectly.

Acknowledgements
This work was supported by National Natural Science Foundation of China (No. 61271114
and No. 61203325) and Innovation Program of Shanghai Municipal Education Commission
(No.14ZZ068).We wish to thank the anonymous reviewers who helped to improve the quality of
the paper. The authors gratefully acknowledge the helpful comments and suggestions of the
reviewers, which have improved the presentation.

References
[1] B. Srbinovski, M. Magno, B. O’Fiynn, et al, Energy aware adaptive sampling algorithm for
energy harvesting wireless sensor networks, In: Proceedings of 2015 IEEE Sensors
Applications Symposium (SAS), IEEE, 2015.
[2] H. J. Visser, R.J.M Vullers, RF Energy Harvesting and Transport for Wireless Sensor Network
Applications: Principles and Requirements, In: Proceedings of the IEEE, 2013.
[3] W. B. Heinzelman, A. P. Handrakasan, H. Balakrishnan, An application- specific protocol
architecture for wireless micro sensor networks, IEEE Transactions on Wireless
Communications 1(4) (2002) 660-670.
[4] O. Younis, S. Fahmy, HEED: A Hybrid, Energy-Efficient, Distributed Clustering Approach for
Ad Hoc Sensor Networks, IEEE Transactions on Mobile Computing 3(4) (2004) 366-379.
[5] C. Tunca , S. Iisk, M.Y. Donmez, et al, Ring Routing: An Energy-Efficient Routing Protocol
for Wireless Sensor Networks with a Mobile Sink, IEEE Transactions on Mobile Computing
14(9) (2015) 1947-1960.
[6] X.P. Fan, X. Yang, S.Q. Liu, et al, Clustering Routing Algorithm for Wireless Sensor Networks
with Power Harvesting, Computer Engineering 34 (11) (2008) 120-122.
[7] Y. K. Yao, G. Wang, Z. Ren, et al, Energy Balanced Clustering Algorithm for Self-Energized
Wireless Sensor Networks, Chinese Journal of Sensors and Actuators 26(10) (2013)
1420-1425.
[8] J. Yick, B. Mukherjee, D. Ghosal, Wireless sensor network survey, Computer Networks 52(12)
(2008) 2292-2330.
[9] Z. S. Chen, H. Shen, Energy-efficient multi-hop routing protocol for wireless sensor networks,
Computer Science 42(8) (2015) 90-94.
[10] B. A. Attea, E. A. Khalil, A new evolutionary based routing protocol for clustered
heterogeneous wireless sensor networks, Applied Soft Computing Journal 12(7) (2012)
1950-1957.
[11] Z. Y. Xu, Y. Yue, et Al, A Game-theoretic approach for efficient clustering in wireless sensor
networks, International Journal of Hybrid Information Technology 7(1) (2014) 67-80.
[12] B. Chen, H. Yao, M. Yang, B.J. LI, L.C. He, A Inter-Cluster Multi-Hop Routing Protocol
Improved Based on LEACH Protocol, Chinese Journal of Sensors & Actuators 27(3) (2014)

11
 373-377.
[13] Y.J. Sun, C.L. Lin, H.F. Jiang, An Energy Efficient Distributed Uneven Clustering Routing
Algorithm for WSNs, Chinese Journal of Sensors and Actuators 28(8) ( 2015) 1194-1200.
[14] W. X. Jiang, C. Lai, A Compressed Sensing Clustering Routing Algorithm for Heterogeneous
Sensor Networks, Journal of Chinese Computer Systems 36(2) (2015) 252-256.
[15] M. Yang,W. Wang, M.A. Chenming, A Topology Construction Method Based on CDS Tree in
Heterogeneous Wireless Sensor Network, Chinese Journal of Sensors & Actuators 29(2)
(2016) 248-255.
[16] G. Smaragdakis, I. Matta, A. Bestavros, SEP: A stable election protocol for clustered
heterogeneous wireless sensor networks, In: Proceedings of 2nd International Workshop on
Sensor and Actor Network Protocol and Applications, 2004.
[17] Q. Li, Q. Zhu, M. Wang, Design of a distributed energy-efficient clustering algorithm for
heterogeneous wireless sensor networks, Journal of Software 29(12) (2006) 2230-2237.
[18] J. Yang, D.Y. Zhang, A data transmission mechanism for wireless sensor networks using
unequal clustering, Journal of Xi’an Jiaotong University 43(4) (2009) 14-17.
[19] M. Xiang, W.R. Shi, C.J. Jiang, Y. Zhang, Energy efficient clustering algorithm for
maximizing lifetime of wireless sensor networks, AEU-Int’l Journal of Electronic and
Communication 64(4) (2010 ) 289-298.
[20] D. Kandris, M. Tsagkaropoulos, and I. Pllitis, Energy efficient and perceived QoS aware
video routing over wireless multimedia sensor networks, Ad Hoc Networks 9(4) (2011)
591-607.
[21] S. K. Gupta, P. K. Jana, Energy efficient clustering and routing algorithms for wireless sensor
networks: GA-based approach, Wireless Personal Communications 83(3) (2015) 2403- 2423.
[22] P. Kuila, P. K. Jana, Energy efficient clustering and routing algorithms for wireless sensor
networks: Particle swarm optimization approach, Engineering Applications of Artificial
Intelligence 33(5) (2014)127-140.
[23] N. Dimokas, D. Katsaros, Y. Manolopoulos, Energy-efficient distributed clustering in
wirelesssensor networks, Journal of Parallel and Distributed Computing 70(4) (2010)
371-383.
[24] T. Amgoth, P. K. Jana, BDCP: A backoff-based distributed clustering protocol for wireless
sensor networks, In: Proceedings of 2013 International Conference on Advances in
Computing, Communications and Informatics, IEEE, 2013.
[25] N.R. Pal, J.C. Bezdek, On cluster validity for the fuzzy c-means model, IEEE Transactions on
Fuzzy Systems 3(3) (1995) 370-379.
[26] Y. Chen, L. Zhang, D. Zhang, D. Zhang, Wrist pulse signal diagnosis using modified
Gaussian models and Fuzzy C-Means classification, Medical Engineering & Physics 31(10)
(2009) 1283-1289.
[27] G. Ran, H. Zhang, S. Gong, Improving on LEACH protocol of wireless sensor networks
using fuzzy logic, Journal of Information & Computational Science 7(3) (2010) 767–775.
[28] Razieh Sheikhpour, Sam Jabbehdari, A Two-Level Cluster based Routing Protocol for
Wireless Sensor Networks, International Journal of Advanced Science and Technology 45(8)
(2012) 19-29.
[29] B.Singh, D. K. Lobiyal, Energy-aware cluster head selection using particle swarm
optimization and analysis of packet retransmissions in WSN, In: Proceedings of 2nd

12
International Conference on Computer, Communication, Control and Information
Technology, 2012.

100

90

80

70
Total energy consumed(J)

60

50

40

30
OCM-FCM
ECS-PSO
20 TLCR

10

0
0 100 200 300 400 500 600 700
Round

Figure 1. Total energy consumed

700
FDT
HDT
LDT
600

500
Lifetime(Rounds)

400

300

200

100

0
OCM-FCM ECS-PSO TLCR

Figure 2. Comparison of the network lifetime

13
 2
OCM-FCM
ECS-PSO
1.8 TLCR

1.6

The average residual energy(J)

1.4

1.2

0.8

0.6

0.4

0.2

0
0 100 200 300 400 500 600 700
Round

Figure 3. The average residual energy

5
OCM-FCM
ECS-PSO
4.5 TLCR
The standard deviation of residual energy

3.5

2.5

1.5

0.5

0
0 100 200 300 400 500 600 700
Round

Figure 4. The standard deviation of residual energy

22
OCM-FCM
ECS-PSO
20 TLCR

18

16
The number of cluster heads

14

12

10

0
0 100 200 300 400 500 600 700
Round

14
 Figure 5. The number of CHs

110

100

90

80

The number of alive nodes

OCM-FCM
70 ECS-PSO
TLCR

60

50

40

30

20

10

0
0 100 200 300 400 500 600 700
Round

Figure 6. The number of alive nodes

100

90

80

70

60

50

40

30

20

10

0
0 10 20 30 40 50 60 70 80 90 100

(a) OCM-FCM

100

90

80

70

60

50

40

30

20

10

0
0 10 20 30 40 50 60 70 80 90 100

15
 (b) ECS-PSO

100

90

80

70

60

50

40

30

20

10

0
0 10 20 30 40 50 60 70 80 90 100

(c) TLCR
Figure 7. The distribution of the nodes when 50 percentages nodes died

Table 1. Detailed parameters for simulations

Parameters Description Value
Area of netwok 200*200
Location of the base station (100, 100)
nodes 100
Eelec 50 nJ/bit
 mp 0.0013 pJ/bit/m4
 fs 10 pJ/bit/m2
Packet size 500 bytes
Initial energy of node 2J

16