Adjust Fuzzy Model Parameters For Head Election in Wireless Sensor Network Protocols

421
Chapter 13
Adjust Fuzzy Model Parameters
for Head Election in Wireless
Sensor Network Protocols
Walaa Abd el Aal Afifi
ISSR-Cairo University, Egypt
Hesham Ahmed Hefny

ISSR-Cairo University, Egypt
ABSTRACT
The clustering routing protocols attract many research papers that result from their well topology
control, less demand resources, and less energy dissipation. The cluster routing protocols consist of
single hop communication and multi hop communication. Single hop is applied between sensor node
and its related cluster head. Multi hop is applied between cluster heads to base station. The previous
two communication modes depend on the cluster head election. Appropriate cluster-head election can
drastically reduce the energy consumption and enhance the lifetime of the network. The fuzzy models
are used frequently for cluster head election. The fuzzy models can be built either expert’s knowledge
or numerical data. The authors propose fuzzy model using adaptive Takagi-Sugeno for wireless sensor
network protocol (FATSN). The FATSN protocol is implemented by modified merging algorithm of fuzzy
clustering with expected value (MCFEV). The FATSN protocol compares with the famous cluster routing
protocol LEACH, EEUC, CHEF, and FCM protocols. The results show that FATSN protocol is efficiency
INTRODUCTION
Wireless sensor networks are receiving a considerable degree of research interest due to their deployment
in an increasing number and variety of applications. Wireless sensor networks are large scale networks
that consist of hundreds or thousands of sensor nodes. The limited energy resource is the main constraint
of the sensor node (Azzedin, 2009). Routing protocols are divided into cluster and multi hop routing
protocols. Cluster routing protocols are more energy efficient than multi hop routing protocols. They are
less demand resources and more scalable. Figure 1 shows cluster network topology. Sensor nodes are
DOI: 10.4018/978-1-5225-0773-4.ch013
Copyright © 2017, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Adjust Fuzzy Model Parameters for Head Election in Wireless Sensor Network Protocols
Figure 1. Cluster network topology
organized into 3 groups. Each group has a cluster head. Cluster heads aggregate data from other nodes
in the same group, and deliver aggregating data to base station via single hop. The cluster head election
has an important role in the increasing network life time, which is defined as the number of rounds until
the first node dead (I.F Akyildizet al., 2002; Mohammad & Imad, 2005).
The common cluster routing protocol is a low energy adaptive clustering hierarchy protocol (LEACH)
(Hinzelmanetal., 2000). Cluster heads are elected randomly. Rotating the cluster heads’ roles applies at
each round. Cluster heads connect to base station via single hop. The drawbacks of LEACH protocol are:
1. Unbalancing energy dissipation.

2. Cluster heads consumed a lot of energy in a transmitting data to base station.
3. Cluster head election doesn’t take into account the energy level of sensor node.
In recent years, the multiple criteria are used to elect cluster heads such as energy, node degree,
distance to base station, and centrality. (Chuen, 1990) Due to the malfunction sensor devices, the uncer-
tainty degree of the collecting data may be presented. The fuzzy set theory can deal with the uncer-
tainty data due to the partial membership values. The object belongs to multiple sets with membership
degree ∈ [0, 1] . The fuzzy models are the famous application of fuzzy set theory. The fuzzy models are
used more in the control applications and the decision making applications. The fuzzy models are built
either by expert’s knowledge or from numerical data. The experts are not easy to find. Even if you find
one, knowledge changes with time and is incomplete. There are multiple cluster routing protocols that
use fuzzy models. They led to reduce energy dissipation, but these routing protocols are built from ex-
pert’s knowledge.
422

This chapter proposes fuzzy model using adaptive Takagi–Sugeno for wireless sensor network protocol
(FATSN). The FATSN protocol is multi input and multi output Takagi–Sugeno models (MIMO TS) for
cluster heads and relay nodes election. The protocol aims to extend network lifetime and identify the
parameters of MIMO TS models from numerical data by using modified merging of fuzzy clustering
algorithm with expected value (MCFEV).
RELATED WORKS
This section describes the related works that are related to the challenges of cluster routing protocols:
the number of clusters and the cluster head election.
• Number of Clusters: Some research papers were based on the reducing the total energy dissipa-
tion and others were based on Xie and Beni’s fuzzy cluster validity index. First: total energy dis-
sipation was a sum of the energy consuming of the internal clusters communications (i.e. between
sensor nodes and related cluster head) and external cluster communication (i.e. between cluster
heads and base station) (Heinzelman et al., 2000). LEACH protocol introduced the optimal num-
ber of clusters in a single hop communication. The optimal number of clusters (C) is estimated as
follows:
N fs M
Copt = ∗ ∗ 2
(1)
2π mp d toBS
Equation 1 shows C value that is based on network area (M), number of sensor nodes(N), and com-
munication mode to base station d 2toBS (i.e. single hop communication), fs is an amplifier for shorter
distance than threshold, and mp is an amplifier for longer distance than threshold. Recently, cluster
routing protocol consists of a single hop communication and a multi hop communication. (Wenfeng,
2008) The authors calculated the optimal number of clusters in a single communication and a multi hop
communication. In single hop communication, optimal number of clusters depends on density of the
network, free space communication, and network area.
3∗ε
Copt = (2)
πλ(2a 2ε − 3e)
where ʎ is a density of network, ε is a free space communication, a 2 is the area of the network, e is an
energy dissipation for running transmitter’s circuit or receiver’s circuit, Copt is the optimal probability
of cluster heads. In multi hop routing; cluster head delivers aggregated data to its relay node. Cluster
head’s transmission range should cover at least two or more cluster range to guarantee the connectivity.
This method is applicable in the dense network. In addition, single space communication may not be
found in the real applications. These previous works showed the internal and external cluster commu-
nication that affected energy level of a cluster head; therefore, they should take into account the cluster
head election. Second: (Raghuvanshi et al., 2010), (Raghuvanshi et al., 2009) used FCM and GK fuzzy
423

clustering algorithms to partition network into C cluster. Xie and Beni’s fuzzy cluster validity index is
used to evaluate the cluster results. The minimum value of Xie and Beni’s index was the optimal number
of clusters. Authors ignored the energy analysis. The optimal number of clusters is based on energy
analysis are more energy efficient than Xie and Beni’s index.
• Cluster Head Election: Some related works used crisp single variable and others used many
linguistic variables. First: (Guihai et al., 2007) The EEUC was energy efficient unequal cluster
protocol. The election criteria was a crisp variable called energy. EEUC generated unequal trans-
mission range that was based on the distance to base station. EEUC ignored the energy dissipation
in the internal cluster communication. There was no guarantee well distributed clusters in the net-
work. EEUC didn’t generate shortest base to base station. (Mao Ye et al., 2005) Energy efficient
clustering schema protocol (EECS) was similar to EEUC protocol. The cluster head was elected
by the energy level. EECS also generated unequal transmission range. The transmission range was
the opposite of the EEUC. Cluster head near to base station had a transmission range greater than
the far cluster head. Cluster heads delivered data to base station via single hop. (Manal Abdullah
et al., 2015) authors used grid-methods and density based clustering algorithm for dividing the
network first into number of grids. And then creating clusters for each grid. The number of grids
and the number of clusters were challenges. The choice of cluster heads was based on energy level.
They preferred the nearest nodes to base station to be cluster heads nodes. They ignored the size
of cluster and the hot spot problem.
• Second: Linguistic variables were energy level, distance to base station, centrality, and node de-
gree (Gupta et al., 2005; Kim & Han, 2008). CHEF protocol used MISO Mamdani model, the
fuzzy rule bases consist of 27, 9 rules respectively. Cluster heads used single hop to sink node.
(Adonias et al., 2011) introduced cluster-head election algorithm using a Takagi-Sugeno model.
The CHEATS was zero order TS fuzzy model. The fuzzy rule base consists of 9 rules (2 linguis-
tic variables). Cluster heads delivered data to sink node by single hop. (Feng et al., 2013) The
ICT2TSK protocol was an improved of the CHEATS protocol. It used zero order TSK 2 type
fuzzy model. The experiment showed ICT2TSK achieved more improvement than the CHEATS
protocol, because 2 type fuzzy models handle un-centrality better than one type fuzzy model.
The outputs for ICT2TSK and CHEATS protocols were determined by authors. (Alexandre et al.,
2013) Authors used K- means algorithm to cluster the network at the first, and then used MIMO
Mamdani models to select cluster heads and relay nodes. This paper doesn’t use the information
of K mean to build fuzzy models. The rule base consisted of 27, 9 rules for cluster heads and relay
nodes election respectively. (Jiejie, 2012) used FCM clustering algorithm to divide the network
into k clusters at the first, and then each node decided whether to be cluster head or non-cluster
head according to the value of fitness function. The fitness function was based on energy level
and distance to sink node. The FCM clustering algorithm generated similar cluster sizes. The fit-
ness function ignored the internal cluster communication. (Akshay Kumaret al., 2015) authors
proposed DCHFC protocol. This protocol did additional step to detect malicious nodes and re-
moved them at first. The potential cluster heads was selected by MISO Mamdani model. The input
variables were energy, reachability (i.e. centrality), and reception power. The final cluster head
was selected based on the minimum distance between it and the neighbor cluster head nodes. The
DCHFC protocol was suitable for sparse network. After some round, the distance between cluster
424

heads increased therefore the energy consumption also increased. The distribution of clusters in
the network was not well.
The drawbacks of cluster routing protocols based on fuzzy models are:
1. These fuzzy models are based on expert’s knowledge; the experts are not easy to find. Expert’s
knowledge changes with time and is incomplete.
2. High overlap degree between linguistic terms for each linguistic variable, therefore, fuzzy models
are less interpretable. There is direct relationship between number of fuzzy rules, linguistic terms
and linguistic variables.
3. The fuzzy models represent the relationship between input variables and output variables. These
cluster routing protocols don’t discuss this relationship
(Vipin Palaet al., 2015) authors proposed a centralized LEACH-GA protocol. They used genetic
algorithm to select cluster heads. The fitness function was based on energy, intra cluster communica-
tion, distance to base station, and number of clusters. The number of nodes equals the population size.
The LEACH-GA was based on single communication between cluster heads and bases station. Cluster
heads consumed a lot of energy with communication with base station.
BACKGROUND
Fuzzy model is the most common way to represent the human knowledge as the natural language expres-
sions of the type “IF–THEN” form (Zadeh, 1988). Fuzzy model is used in several control applications
such as industry, water quality control, automatic train operation system, elevator control, and medical
science …etc. It is sometimes known as fuzzy logic control, expert system, fuzzy inference system, and
fuzzy rule based system.
Fuzzy model consists of four parts: a fuzzification, a defuzzification, an inference engine, and a
rule base as shown in Figure 2. fuzzification converts the crisp data into fuzzy sets. Defuzzification is
a conversion of a fuzzy set to corresponding crisp value. The inference engine is used to form infer-
ences and to draw a conclusion from the fuzzy rules. The output of the inference engine sends to the
defuzzification. Rule base is a set of “if then” rules. Each rule consists of antecedent or premise and
consequent parts (Kwang, 2005).
Figure 2. Fuzzy model
425

The fuzzy model consists of less number of fuzzy rules and the distinct fuzzy sets but with high overlap
degree between them that it will be hardly interpreted. Fuzzy model is built by expert’s knowledge or
from numerical data. In the first case, the experts are not easy to find. Even if you find one, knowledge
changes with time and is incomplete. In the second case, the fuzzy cluster algorithms and least square
errors are used to identify the parameters of fuzzy model. The number of fuzzy rules or clusters is pre-
requested. The fuzzy clustering algorithms should reflect the natural shape of clusters with different
density and parallel to axis. In addition, fuzzy model should be simplified to merge similar fuzzy sets
and then reducing the number of rules (Kevin, 1998).
Takagi–Sugeno model (TS) is an example of fuzzy model. It differs from Mamdani model: the con-
sequent is a crisp linear function of input variables, therefore no need for defuzzification method and
TS model enables the fuzzy part to participate in the calculation of the inferred output value (i.e. firing
degree) (Frank& Rudolf, 1997). The general form of TS fuzzy rule is as follows:
Ri: IF u is Ai and v is Bi THEN wi = fi (ui , vi ) = θ0 + θ1ui + θ2vi (2)
where wi = fi ( ui , vi ) is a crisp function; θ0 , θ1 , θ2 are consequent parameters, Ai and Bi are linguistic

terms, and ui and vi are real input variables(Michio & Takahiro, 1993). Figure 3 shows a simple example
for multiple input and single output (MISO) TS model. It consists of 2 rules; each rule consists of prem-
ise and consequent parts. In the case of singleton inputs (u=u0, v=v0), the output of fuzzy model (w0) is
calculated as follows:
1. Calculate output for each rule wi where i=1, 2 by substituting u0, v0 in f1 and f2 functions respectively.
Calculate the firing degree αi where i=1, 2.
Figure 3. TS model
426

αi = (µAi(u 0)^ µBi(v 0)) (3)
where ^ is a minimum operator µAi(u 0) is a membership value of u0 in fuzzy sets Ai, and µBi(v 0) is a
membership value of v0 in fuzzy sets Bi.
3. Calculate output model w 0 as follows
α1w 1 + α2w 2
w0 = (4)
α1 + α2
Multi-input and multi-output (MIMO) TS fuzzy model is a combination of multi- input single output
(MISO) TS models. It is possible to use the same or different input variables for each output variables
according to application requirements. Figure 4 shows the flowchart of identification parameters of
fuzzy models (Jang, 1997).
Structure Selection
It determines the relevant input/output data with respect to the goal of a system, and then represents them
in the regression matrix. Regression matrix represents the product space of the input - output data (x, y).
Fuzzy Clustering Algorithm
The clustering algorithm should reflect the natural shape of cluster with different densities, parallel to
axis, and separated clusters. The fuzzy c means algorithm (FCM) (Menahem & Abraham, 1999) and
Possibility c means clustering algorithm (PCM) (Krishnapuram & KELLER, 1993) are used in the iden-
tification parameter problem. They are based on minimization of the sum of weighted distance function
between data points and cluster centers. The drawbacks of FCM clustering algorithm:
Figure 4. Identification parameters of fuzzy models
427

1. Cluster shape is determined by distance function. Euclidean distance generates spherical cluster
shape.
2. Prototypes are affected by outlier; weight exponent is fixed over all iterations. Small cluster size
involved high weight exponent.
3. Membership function doesn’t describe the membership value of data point well due to the sum of
membership of point in all clusters which must be equal 1.
4. Cluster results are sensitive to initialization parameters.
5. Objective function ignores the separation measure.
Due to the outliers or noisy points, the fuzzy membership value can’t describe the degree of belong-
ing well. The PCM algorithm describes the degree of membership of point in all clusters independently
(Krishnapuram & Keller, 1993). The condition of sum of membership of point in all clusters removes.
This condition changes to be membership of point in cluster greater than zero. The drawbacks of PCM
algorithm are:
1. If there is an attracted cluster for all points, PCM produces identical clusters or one cluster.
2. PCM is sensitive to initial parameters.
The PCA algorithm uses the exponential membership function for estimating membership values
(Shen & Kuo, 2008). The PCA requires less number of adjusting constant parameters. The PCA deals
with outlier and noisy data as separated clusters, but it may be found two or more identical cluster centers.
In FATSN protocol, the modified merging of fuzzy clustering algorithm with expected value (MCFEV)
is used for cluster head election, relay node election and cluster formation phases as shown in Figure 5.
3. Initial Fuzzy Model: It involves tuning the membership function’s parameters in the premise part,
and the consequent parameters. (Jang, 1997). The free derivative optimization algorithms such as
Downhill simplex search method use to adapt the premise’s membership function. The least square
methods are used to adjust consequent parameters (Kevin & Stephan, 1998).
4. Model Simplification: Initial fuzzy models may contain more redundant membership functions
than necessary to describe the system. (Magneetal, 1998). Fuzzy similarity measures can be applied
to simplify or reduce the initial fuzzy rule base. In FATSN protocol, the authors use the similarity
measure (Hefny, 2007) which overlap degree is estimated by two parameters.
5. Model Validation: (Abdolrezaetal, 2013) explains the root mean square error (RMSE). The RMSE
returns root mean squared error between output model and original output model. If the value of
RMSE is less than 1 and close to zero, the fuzzy model will be accepted.
Figure 5. FATSN protocol
428

N
RMSE = ∑ (y − yˆ)^ 2 N
i =1
(5)
where y is the original output model, and ŷ is output of fuzzy model, and N is number of data Points.
THE FUZZY MODEL USING ADAPTIVE TAKAGI-SUGENO FOR

WIRELESS SENSOR NETWORK PROTOCOL (FATSN)
This chapter introduces the FATSN protocol for cluster head election and relay nodes election. The
FATSN protocol is MIMO TS models. The FATSN protocol is built from numerical data using the
modified merging of fuzzy clustering algorithm with expected value (MCFEV). The authors follow the
steps as in Figure 4 for identifying the parameters of MIMO TS models.
Network Configuration
N’ sensor nodes are distributed uniformly in an L * L with following characteristics:
1. Sensor nodes are static nodes.

2. It is impossible to replace the batteries.
3. All nodes can adjust transmission range.
4. Base station is static and powerful node.
FATSN Protocol
The FATSN protocol is similar to LEACH protocol. FATSN protocol is a centralized protocol. Figure
5 shows the FATSN protocol which consists of set up and transmission phases. Set up phase divides
into two phases: cluster formation and cluster head election phases. The cluster formation is applied
only one time rather than repeated at each time cluster heads election as in previous works. Base station
sends beacon message to all sensor nodes in the network. Each node calculates its distance to base sta-
tion according to the strength of signal. Each sensor node sends hello message to discover the neighbor
nodes and calculates the sum of distance between itself and its neighbors based on the strength of signal
(i.e. centrality), sensor nodes send control packet to base station with the sensor ID, its energy level, its
distance to base station and centrality.
Cluster formation divides the coordination system of sensor nodes into k clusters by modified merg-
ing of fuzzy clustering algorithm with expected value. At this step; we got the well distributed clusters
in the network area. This led to minimize internal cluster communication.
Cluster head election phase begins after finishing cluster formation step. It means which sensor
node in every cluster (cluster formation) should be cluster head node in the current round. Base station
constructs data set from received control packets. The data set is N * d dimensions, where N is number
of sensor nodes. The dimensions (d) represent the following attributes:
429

1. The energy level (EN).

2. Distance to base station (D_BS).
3. Centrality: The minimum value of centrality for sensor node Si means it is more central.
4. Chance: probability of sensor node becomes cluster head. The maximum value denotes the maxi-
mum probability of sensor node Si to become cluster head.
EN (i ) EN (i )
Chance(i ) = + (6)
centrality(i ) D _ BS (i )
where EN (i) is the current energy level of sensor node Si, D_BS (i) is distance to base station of Si,
centrality (i) is the centrality value of Si 1 ≤ i≤ N.
5. Cost is the same as the centrality, but it estimates the sum of distance between sensor node and
non- neighbor nodes with the condition d (Sj, BS) < d (Si, BS).
p
Cost (si) = ∑ d (Si, Sj ) (7)
l =1
6. Relay cost evaluates the distance between the sensor node and its relay node plus the distance to
base station of relay node.
p
Relay Cos t(i ) = ∑ (d (Si, Sj ) + d (Sj, BS )) (8)
l =1
The energy dissipation via relay node Sj can be evaluated by distance measure, d (Si, Sj) is distance
between Si and relay node Sj. P is number of nodes in relay list of sensor node Si, d (Sj, BS)is a distance
between Sj and base station BS.
At this step, the procedure of identification parameters of MIMOTS model (as shown in Figure 4) is
applied. This procedure is repeated for MISO cluster head election and MISO for relay nodes election.
The regression matrix (i.e. structure selection) consists of input- output data. The rows are N sensor
nodes, and columns are (d) dimensions.
1. For cluster head election, regression matrix consists of input variables: energy, centrality, and
distance to base station, and output variable is chance.
2. For relay node election, regression matrix consists of the input variables: energy, distance to base
station, and cost, and output variable is relay cost.
In fuzzy clustering algorithm, authors use MCFEV clustering algorithm. In initial fuzzy model,
Downhill simplex search is used to tune the Gaussian membership function in premise part. Least square
method is used to estimate the consequent parameters. In simplification model; similarity measure
(Hefny, 2007) estimates the overlap degree by membership function’s parameter:
430

• Inter Cluster Communication: Each cluster head node selects the relay node. These relay nodes
mention in the path to base station only one time, the path starts with cluster head node that has
the maximum relay cost, which means the farthest node to base station, this cluster head node
chooses its relay node having the maximum relay cost, this relay node choose other node, that has
maximum relay cost, the last node on path has the minimum relay cost, it will forward data packet
directly to base station.
• Transmission Phase: The sensor nodes perform measurements and append its residual energy
level to each packet before transmitting it to their cluster head. The cluster heads receive this data,
append residual energy level, aggregate the data packets and forward the resulting packet to the
next relay node. The cluster head rotation applies at each round, after a few round, if the perfor-
mance index (VAF) is less than 99; the base station updates data sets, and repeats the steps.
Modified Merging of Fuzzy Clustering Algorithm with Expected Value (MCFEV)
The fuzzy clustering algorithms are based on minimizing the sum of weighted distance function between
data point and cluster centers ignore the separation distance between cluster centers (Carl, 2001). The
fuzzy clustering algorithm with expected value uses merging algorithm to get distinct clusters. The
merging algorithm is based on Euclidean distance. It may be found two separated clusters but they are
overlapped. The Euclidean distance function is not a good indicator for the separated clusters, and also
it doesn’t define the objective function (Afifi & Hefny, 2014; Afifi & Hefny, 2015). In previous work,
the authors use fuzzy clustering with expected value except the merging algorithm. In this chapter, the
authors propose MCFEV clustering algorithm. The MCFEV changes the merging algorithm of fuzzy
clustering with expected value. The merging algorithm will be based on membership function. The
overlap degree is measured between two clusters by similarity measure. The similarity measure uses
centers and variances to estimate overlap degree.
The MCFEV algorithm is based on the minimization of the following objective function:
c N c M
Obj = ∑ ∑ FWijm
i =1 j =1
xj − µi
A
i =1 j =1
( )
+ ∑ σi ∑ FWijm log FWijm − FWijm (9)
where N is number of points belonging to cluster i, C is a number of clusters, xj is data point, σi is vari-
ance, FWijm is fuzzy weight or membership value of point j in cluster i, and µi is a cluster center. The
necessary conditions to minimize the objective function are FWijm and µi . Objective function is derived
with respect to the membership value or fuzzy weigh FWijm and cluster center µi or weighted fuzzy
expected value (WFEV). The membership function is Gaussian membership function (Equation 11),
and cluster centers are obtained by equation 12. Figure 6 shows the flowchart of MCFEV algorithm.
In Figure 6, the MCFEV algorithm requires the number of clusters (C), the number of iterations (it),
the number of sensor nodes (N), the eliminating threshold (E), the fault tolerance (ε) and the overlap
threshold (O).
Step 1: Draw C initial cluster prototypes from dataset.
431

Figure 6. MCFEV algorithm
Step 2: Run K mean algorithm for a few rounds (i.e. 10 rounds) to initial the crisp cluster centers and
the crisp variances.
Step 3: The fuzzy weight (FW), weighted fuzzy expected value or cluster prototypes (WFEV), and
weight fuzzy variance (WFE) are calculated by equations 10, 11, 12 respectively.
 
 x i − µj t 
exp  
 t 
( ) 2
 2σ j 
 
FWij(t ) = (10)
 M 
∑ FWij 
 i =1 
M
WFEVjt +1 = µtj +1 = ∑ FWi j(t )x i (11)
i =1
( ) ( )
(t +1) 2
WFVjt = σ 2j = ∑ FWij(t ) x i − µtj (12)
i =1
where M is number of points belonging to cluster j, xi is data point, σi is a weighted fuzzy variance.
432

Step 4: Assign the sensor nodes to each cluster that are based on the minimum distance.
Step 5: Repeat steps 3-4 until current round number (r) greater than it.
Step 6: Calculate crisp variance (VAR) instead of weight fuzzy variance (WFE), the fuzzy weight (FW),
weighted fuzzy expected value or cluster prototypes (WFEV).
Step 7: If the difference between cluster centers is less than fault tolerance (ε), then algorithms stop and
goes to step 8, otherwise goes to step 6.
Step 8: The merging algorithm requires the overlap degree (O). It is applied as follows:
◦◦ 8-1: create the similarity matrix S (C x C)
◦◦ 8-2: For each row
◦◦ 8-2-1 Extract S (.) >= O
◦◦ 8-2-1-1mergestwo clusters
◦◦ 8-2-1-2 S (.) =0
◦◦ 8-3 Next.
The advantages of MCFEV clustering algorithm:
1. It permits the cluster to take the natural shape.

2. It describes the degree of membership of point in each cluster separately.
3. It produces separated clusters by utilizing merging algorithm.
4. Overlap degree is measured by membership function’s parameters.
5. It utilizes the WFEV as a cluster prototype rather than traditional average method.
6. It is less sensitive to initial parameters (i.e. cluster centers).
7. Eliminate function enable to generate cluster with different densities.
8. The contour of the cluster is ellipse that can be seen under bell shape.
The next section provides the simulation result for proposed protocol (FATSN) against the famous
cluster routing protocol. It provides also the cluster results.
SIMULATION
This section provides the simulation results for the FATSN protocol against the famous cluster routing
protocols LEACH, CHEF, cluster routing based on FCM algorithm (Jielie, 2012) and EEUC protocol.
The simulation is implemented by Matlab 2010. The length of data packet equals 4000 bits; the first
radio (Hinzelmanetal., 2000) is used to calculate the energy dissipation. The energy model parameters
are explained as shown in Table 1.
The network life time is estimated by two metrics: first node die (FND) and half of the nodes die
(HND). The threshold for LEACH protocol in all scenarios is 1%. The number of clusters in FCM, and
FATSN protocol equal 10(cluster formation) in all clusters. The EEUC’s multi hop routing protocol
is applied in CHEF and FCM cluster protocols except LEACH protocol. The fuzzy model is built on
75% of the whole data set for each MISO fuzzy TS model. As in related works, the number of rules for
3 input variables and single output variable is 27 rules. The similarity measure (Hefny, 2007) is used
to estimate overlap degree by using an iterative merge algorithm (R. Babuska, 1998), and also used to
calculate similarity degree between input fuzzy sets and fuzzy set in the premise.
433

Table 1. Energy model
Parameters Values
Initial energy 0.5 J/bit
Eelec 50nJ/bit
εfs 10 PJ/bit/m2
εmp 0.0013 PJ/bit/m2
EDA 5nJ/bit.
The input fuzzy set with two parameters is a mean (feature values) and avariance. Authors calculate
the variance for each input variable as follows:
1. Calculate membership value for each data point in fuzzy set’s premise part as singleton.
2. The results of previous step are the membership values of data points, and centers equal the data
points.
3. Variance or width is calculated as equation in one absence parameter.
4. Take the mean for the calculated variances; then three variance values for three input variables
respectively.
The thresholds for merging similar fuzzy sets and removing fuzzy sets that they are similar to universal
set are above 0.40, 0.70 respectively. The mean results are estimated over all 20 runs.
Cluster Shape
Figure 7 shows the network topology after applying the modified merging algorithm of fuzzy cluster-
ing algorithm with expected value (MCFEV). Number of sensor nodes is 100. Network area is 100m
* 100m. The initial number of clusters equal 15, 10 clusters. The eliminating factor is2, stop criteria is
0.00001, and threshold for merging similar clusters is above 0.65. The number of clusters after eliminat-
ing the empty clusters and less cluster size equals 14, 10 clusters. The number of clusters after merging
algorithm equals 12, 9 as shown in Figures 7a and 7b respectively.
Cluster prototypes aren’t affected by outlier. Cluster prototypes are located in the dense area. The
Xie and Beni’s validity index is 1.33, 0.0561 as shown in Figures 7a and 7b respectively. The MCFEV
algorithm doesn’t affect outlier. The distribution of clusters in the space indicates the power of similarity
measure to detect overlapped clusters.
Number of Clusters
The FCM and MCFEV clustering algorithms run on 100 sensor nodes in 100 m * 100 m network area.
The Xie and Beni’s validity index is estimated for all experiments. The average results are plotted with
respect to number of clusters from 3 to 10 clusters. In MCFEV, we merge the clusters are overlapped
(i.e. above 0.8). Figure 8 shows the XB index values against the number of clusters.
434

Figure 7. The results of MCFEV clustering algorithm
Figure 8. XB index for FCM and MCFEV algorithms
The FCM and MCFEV agree the optimal number of clusters at the range from 3 to 6 clusters. In the
MCFEV algorithm, 10 clusters are also optimal number of clusters. These results agree with the optimal
number of clusters in the LEACH protocol. However, the calculation of number of clusters is based on
distance function.
RMSE Performance Index
The authors apply the FATSN protocol with FCM, PCA and MCFEV clustering algorithms. The pur-
pose of this experiment is comparing the fuzzy partition and possibility partition. The PCA algorithm
needs the good initial cluster centers. The cluster centers result from FCM algorithm that they are used
435

as initial the cluster centers. The averages results of RMSE performance index and average number of
fuzzy rules are estimated over 20 rounds. Number of sensor nodes equals 100 at 100 * 100 network area.
Figure 9a shows average number of fuzzy rules. The number of rules of MCFEV and PCA algorithms
is less than number of rules of FCM. The reduction in MCFEV algorithm is resulted from the effect of
similarity measure, eliminating function, and defuzzification method. The reduction in FCM algorithm is
limited to the similarity measure only. The reduction of PCA is resulted from two or more similar cluster
centers, and similarity measure. Although, the FCM is used to initial cluster centers of PCA algorithm.
Figure 9b shows the value of RMSE for three fuzzy clustering algorithms. The RMSE of cluster head
is similar in three fuzzy clustering algorithms. The RMSE of relay node of MCFEV is less than RMSE
of relay node of FCM and is similar to RMSE of PCA algorithm.
Figure 10 and Figure 11 provide examples of fuzzy rule bases in one experiment in scenario 1. The
rule base for MISO TS model for cluster head election is shown in Figure 10. The number of rules is 8
rules after simplification, and the consequent part of fuzzy rules is shown in Table 2. The rule base for
MISO TS model for relay nodes election is shown in Figure11. The number of rules is 12 rules after
simplification, and the consequent part of fuzzy rules is shown in Table 3.
Network Life Time
Scenario 1: Sensor nodes are distributed randomly in 100 m*100 m network area. The base station is
located at (50,175). The mean results are estimated over all 20 experiments. The FATSN proto-
col is compared with LEACH, CHEF, (Jielie, 2012) FCM and EEUC protocols. The maximum
transmission range for EEUC is 60. The threshold for CHEF and EEUC is 0.2. Network lifetime
is estimated by two metrics: first node dead (FND) and half node dead (HND). Table 4 explains
the network life time metrics. The LEACH protocol is the poorest one; cluster heads lost a lot of
energy to deliver data to base station. LEACH elects cluster heads randomly without taking into
account any criteria. The performance of CHEF protocol isin limited range because it consid-
ers two parameters to elect cluster heads (energy, and centrality), CHEF protocol achieves more
performance in dense network. Cluster routing based on FCM algorithm doesn’t consider energy
dissipation of internal cluster communication. EEUC protocol estimates radius competition based
Figure 9. Average numbers of fuzzy rules and RMSE for cluster heads and relay nodes
436

Figure 10. Premise part of fuzzy rules for cluster heads election
Table 2. Consequent part of fuzzy rules for cluster heads election
-0.00175 X1+ -2.519 X2+ -2.98388 X3+ 3.611

0.001097 X1+ -3.155 X2+ -3.22801 X3+ 4.261
-5.32E-04 X1+ -1.134 X2+ -1.84893 X3+ 2.008
0.018361 X1+ 2.683 X2+ -1.22005 X3+ -1.686
0.001173 X1+ -2.804 X2+ -3.05759 X3+ 3.894
-0.00101 X1+ -0.366 X2+ -0.94948 X3+ 0.957
-0.00199 X1+ -0.183 X2+ -0.62316 X3+ 0.623
0.000145 X1+ -0.485 X2+ -1.12208 X3+ 1.142
on distance between sensor node and base station. The up and down of number of clusters effect
on network life time especially FND. The FATSN protocol is the more performance protocols,
because we take into account the internal cluster communication in cluster head election, well
cluster distributed in cluster formation, and also inter cluster communication (cluster heads elect
relay node to deliver data to base station).
Scenario 2: The FATSN runs in different network configuration. Scenario two is a trial to create separated
data set. 100 sensor nodes are distributed randomly in 200 m* 200m. Base station is located at (100,
100) center of area. The maximum competition radius assigns for EEUC 70 m. The probability of
becoming cluster heads is 0.3 for EEUC and CHEF. Figure 12 illustrates the FATSN protocol is the
more efficient protocols. The HND of EEUC is the highest one because the base station is located
at the center of the network area. The difference between the EEUC and the FATSN protocols in
HND values is not significant.
437

Figure 11. premise part of fuzzy rules for relay nodes election
Table 3. Consequent part of fuzzy rules for relay nodes election
-0.84538 X1+ -0.29259 X2+ 0.591 X3+ 0.922

-3.54E+00 X1+ -2.54E+00 X2+ 0.491 X3+ 4.12
-3.42722 X1+ -1.91696 X2+ 0.462717 X3+ 3.750309
11.38294 X1+ 6.507744 X2+ 1.079055 X3+ -12.2574
-4.96268 X1+ -3.12256 X2+ 1.118068 X3+ 5.374011
2.405953 X1+ 3.854612 X2+ 0.647135 X3+ -3.79935
0.069157 X1+ 0.023547 X2+ 0.856898 X3+ -0.06922
-0.26463 X1+ -0.07326 X2+ 0.900628 X3+ 0.26719
-1.9779 X1+ -0.92953 X2+ 0.798301 X3+ 2.075328
-1.69607 X1+ -1.31187 X2+ 0.37526 X3+ 2.167877
-0.12466 X1+ -0.06755 X2+ 0.948959 X3+ 0.134202
0.243885 X1+ 0.451376 X2+ 0.9672778 X3+ -0.44097
Sum of Remaining Energy
Scenario 1: The sum of remaining energy indicates the level of energy dissipation at each round. As
mentioned above, the number of sensor nodes equals 100 nodes, location of base station equal
(50,175), and network area equals100m *100m. Figure 13 shows that the LEACH protocol is the
highest one in energy dissipation. The FATSN protocol is the least one in energy dissipation.
Scenario 2: The sum of remaining energy of FCM and EEUC are close to each other. Figure 14 shows
the FCM protocol is better than LEACH, and CHEF. The FATSN protocol is the highest one.
438

Table 4. The FND and HND
Protocols 100 Nodes 150 Nodes 200 Nodes

FND HND FND HND FND HND
LEACH 550 896 570 915 564 947
CHEF 695 1161 766 1198 763 1208
FCM 532 1250 329 1158 280 1156
EEUC 577 1147 437 1240 421 1235
FATSN 1067 1272 992 1247 1103 1239
Figure 12. the FND and HND metrics
Figure 13. Scenario1: sum of remaining energy
439

Figure 14. Scenario2: sum of remaining energy
FUTURE WORK
The FATSN protocol will experiment with mobile sensor nodes. Researchers will investigate other vari-
ables that they affect the network lifetime. We will study different multi hop routing protocols, different
fuzzy clustering algorithms, different similarity measures, and different fuzzy sets.
CONCLUSION
This chapter provides the FATSN protocol for cluster heads and relay nodes election. The FATSN proto-
col adapts TS model for the election process. The FATSN is based on numerical data. Then, there is no
need to expert’s knowledge. The fuzzy clustering algorithm MCFEV and least square method are used
to adapt the parameters of fuzzy TS model. The reduction of TS model is done by similarity measure
and fuzzy clustering algorithm. The fuzzy clustering algorithm MCFEV returns natural shape clusters
with different densities. The MCFEV describe the membership value for each node in the cluster sepa-
rately. Researchers introduce two functions to estimate the chance of sensor node to be cluster head and
to estimate the relay cost. The FATSN protocol achieves high performance against other cluster routing
protocols.
REFERENCES
Abdullah, M., Eldin, Al-Moshadak, Alshaik, & Al-Anesi. (2015). Density Grid-Based Clustering for
WirelessSensors Networks. International Conference on Communication, Management and Information
Technology, Procedia Computer Scienc.
440

Afifi, W., & Hefny, H. (2014). Adaptive Takagi–Sugeno Fuzzy Model using Weighted Fuzzy Expected
Value in Wireless Sensor Network. In Proceeding of The IEEE- 14th International Conference on Hybrid
Intelligent Systems (pp. 225-231).
Afifi, W., & Hefny, H. (2015). An Enhanced fuzzy model for cluster head election in wireless sensor
network. (Unpublished Master thesis). Institute of Statistical and Operation Researches, University of
Cairo, Egypt.
Akyildiz, I. F., Su, W., Sankarasubramaniam, Y., & Cayirci, E. (2002). A Survey on sensor networks.
IEEE Communications Magazine, 40(8), 393–422. doi:10.1109/MCOM.2002.1024422
Babuska, R. (1998). Fuzzy modeling for control. Boston: Kluwer Academic Publishers. doi:10.1007/978-
94-011-4868-9
Bagci, & Yazici. (2013).An energy aware fuzzy approach to unequal clustering in wireless sensor net-
works. International Journal of Applied Soft Computing, 13(4), 1741–1749.
Boukerche. (2009). Algorithms and protocols for wireless sensor networks. John Wiley & Sons.
Chen, G., Li, C., Ye, M., & Wu, J. (2007). An unequal cluster-based routing protocol in wireless sensor
networks. International Journal of Wireless Networks, 15(2), 193–207. doi:10.1007/s11276-007-0035-8
Chen, J. (2012). Improving life time of wireless sensor networks by using fuzzy c-means induced clus-
tering. In Proceedings ofIEEEWorld Automation Congress (pp. 1-4).
Friedman, & Kandel. (1999). Introduction to pattern recognition: statistical, structural, neural and fuzzy
logic approaches. World Scientific.
Hefny, H. (2007). Comments on Distinguishability quantification of fuzzy sets. International Journal
of Information Sciences, 177(21), 4832–4839.
Heinzelman, W. B., Chandrakasan, A., & Balakrishnan, H. (2000). Energy efficient communication
protocol for wireless microsensor networks. In Proceeding of System Sciences (pp. 1-10). IEEE.
Ilyas & Mahgoub. (2005). Handbook of Sensor Networks: Compact Wireless and Wired Sensing Systems.
CRC Press.
Jang, Sun, & Mizutani. (1997). Neuro-Fuzzy and Soft Computing. Prentice Hall. doi:10.1109/
ICCCT.2010.5640391
Kim, J., Park, S., & Han, Y. (2008). CHEF: cluster head election mechanism using fuzzy logic in wire-
less sensor networks. In Proceedings of IEEE the Advanced Communication Technology ICACT (pp.
654–659). doi:10.1109/ICACT.2008.4493846
Klawonn, F., & Kruse, R. (1997). Constructing a fuzzy controller from data. International Journal of
Fuzzy Sets and Systems, 85(2), 177–193. doi:10.1016/0165-0114(95)00350-9
Krishnapuram, R., & Keller, J. M. (1993). A possibilistic approach to clustering. International Journal
of Fuzzy Systems IEEE Trans, 1(2), 98–110. doi:10.1109/91.227387
Lee, K. H. (2005). First course on fuzzy theory and applications. Springer-Verlag Berlin Heidelberg.
441

Lee. (1990). Fuzzy logic in control system: fuzzy logic controller – part 1. International Journal of
Systems, Man and Cybernetics, IEEE Transactions, 20(2), 404 – 418.
Li, & Shen. (2008). Optimal Cluster Number Determination for clustered Wireless Sensor Networks. In
Proceeding of IEEE Transactions on Global Telecommunications.
Looney, C. G. (2001). Interactive clustering and merging with a new fuzzy expected value. International
Journal of Pattern Recognition, 35(11), 2413–2423. doi:10.1016/S0031-3203(01)00213-8
Michiosugeno, & Yasukawa. (1993). Fuzzy logic based approach to qualitative modeling. International
Journal of Fuzzy Systems on IEEE Transactions, 1(1), 7-31.
Silva, Maciel, & Corrêa. (2013). Multi-hop Energy-efficient Control for Heterogeneous Wireless Sen-
sor Networks Using Fuzzy Logic. International Journal of Soft Computing and Software Engineering,
3(3), 46–53.
Pahwa, Virmani, Sahil, Rathi, & Swami. (2015). Dynamic Cluster Head Selection Using Fuzzy Logic on
Cloud inWireless Sensor Networks. International Conference on Intelligent Computing, Communication
&Convergence,Procedia Computer Science.
Pal, Pal, Keller, & Bezdek. (2005) A Possibilistic Fuzzy c-Means Clustering Algorithm. International
Journal of IEEE Transactions on Fuzzy Systems, 13(4), 517-530.
Passino, K. M., & Yurkovich, S. (1998). Fuzzy control. Addison Wesley Longman.
Pires, Silva, Cerqueira, Monteiro, & Viegas, Jr. (2011). CHEATS: A Cluster-Head Election Algorithm
for WSN Using a Takagi-Sugeno Fuzzy System. In Proceedings of IEEE Communications LATINCOM.
Raghuvanshi, S., & Tiwari, S. (2009). G K Clustering Approach to Determine Optimal Number of
Clusters for Wireless Sensor Networks. In Proceeding of IEEE Wireless Communication and Sensor
Networks (pp. 1-6).
Riordan & Sampalli. (2005). Cluster–head election using fuzzy logic for wireless sensor networks. In
Proceedings of Communication Networks and Services Research (pp. 255-260).
Setnes, Babuska, Kaymak, & van Nauta Lemke. (1998). Similarity measures in fuzzy rule base simplifi-
cation. International journal of Systems, Man, and Cybernetics, Part B: Cybernetics, IEEE Transactions
on, 28(3), 376-386.
Tiwari, T., & Kishor. (2010). Optimal Number of Clusters in Wireless Sensor Networks: An FCM Ap-
proach. In Proceeding of IEEE Computer and Communication Technology (pp. 817 – 823).
Yanga, & Wu. (2008). Unsupervised possibilistic clustering. International Journal of Pattern Recogni-
tion, 39(1), 5–21.
Yazdani-Chamzini, Razani, Yakhchali, Zavadskas, & Turskis. (2013). Developing a fuzzy model based
on subtractive clustering for road header performance prediction. International journal of Automation
in Construction, 35, 111–120.
Ye, Li, Chen, & Wu. (2005). EECS: An Energy Efficient Clustering Scheme in Wireless Sensor Net-
works. In Proceeding of Performance, Computing, and Communications Conference (pp. 535– 540).
442

Yogitab, Singhc, & Yadavc. (2015). Cluster Head Selection Optimization Based on Genetic Algorithm
toProlong Lifetime of Wireless Sensor Networks. Third International Conference on Recent Trends in
Computing,Procedia Computer Science.
Zhang, F., Zhang, Q.-Y., & Sun, Z.-M. (2013). ICT2TSK: An Improved Clustering Algorithm for WSN
Using a Type-2 Takagi-Sugeno-Kang Fuzzy Logic System. In ProceedingsofIEEE Symposium on Wire-
less Technology and Applications (pp. 153 - 158). doi:10.1109/ISWTA.2013.6688759
KEY TERMS AND DEFINITIONS
Cluster Head: A senor node. It is responsible for collecting data from member nodes inside the
cluster. It is also responsible for aggregating and delivering data to base station.
Clustering Algorithms: The process of dividing objects into classes or clusters. The objects inside
the cluster are more similar and dissimilar to other objects outside cluster.
Energy: The main constraint of the sensor node. Sensors’ batteries are not chargeable in harsh en-
vironment. The energy is more consuming in the transmission phase than the sensing phase. Energy is
measured by joule unit.
Fuzzy Model: The most common way to represent the human knowledge as natural language expres-
sions of the type “IF–THEN” form. The IF part is called the antecedent and the THEN part is called the
consequent. The antecedent part partitions input space into fuzzy regions. The consequent part partitions
output space into fuzzy regions or represents linear function of input variables.
Number of Clusters: A pre-request for the almost clustering algorithms. They are based on mini-
mizing the objective function. The number of clusters is an indicator of the quality of the cluster results.
The cluster results should be more separated and more compacted. From the energy dissipation point
view, the number of clusters affect on the network life time. The large number of clusters increases
energy consuming in the multi hop communication. The small number may reduce energy consuming
but the high network density can have opposite effect on the network life time. So the determination of
the number of clusters is demand for the cluster routing protocols.
Routing Protocols: Determine the route between sensor nodes and base station. Routing protocols
divided into single hop or direct communication and multi hop communication. Sensor node sends data
directly to cluster head via single hop. Cluster heads send aggregated data packets to base station via
multi hop communication.
TAKAGI–SUGENO Model: An example of fuzzy model. The consequent part is a crisp linear func-
tion of the input variables. The antecedent part partitions input space into fuzzy regions. The advantages
of TS model are no need for defuzzification method and fuzzy antecedent participate in the calculation
of the inferred output value.
Wireless Sensor Networks: Large scale networks that consist of hundreds or thousands of sensor
nodes. Sensor nodes have the ability to monitor physical or environmental conditions such as temperature,
sound, pressure, etc. Sensor nodes work together to deliver data to base station.
443

Adjust Fuzzy Model Parameters For Head Election in Wireless Sensor Network Protocols

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Adjust Fuzzy Model Parameters For Head Election in Wireless Sensor Network Protocols

Uploaded by

Copyright:

Available Formats

421

Hesham Ahmed Hefny

Figure 1. Cluster network topology

1. Unbalancing energy dissipation.

The drawbacks of cluster routing protocols based on fuzzy models are:

Figure 2. Fuzzy model

Ri: IF u is Ai and v is Bi THEN wi = fi (ui , vi ) = θ0 + θ1ui + θ2vi (2)

where wi = fi ( ui , vi ) is a crisp function; θ0 , θ1 , θ2 are consequent parameters, Ai and Bi are linguistic

αi = (µAi(u 0)^ µBi(v 0)) (3)

3. Calculate output model w 0 as follows

Fuzzy Clustering Algorithm

Figure 4. Identification parameters of fuzzy models

Figure 5. FATSN protocol

THE FUZZY MODEL USING ADAPTIVE TAKAGI-SUGENO FOR

N’ sensor nodes are distributed uniformly in an L * L with following characteristics:

1. Sensor nodes are static nodes.

1. The energy level (EN).

Modified Merging of Fuzzy Clustering Algorithm with Expected Value (MCFEV)

Step 1: Draw C initial cluster prototypes from dataset.

Figure 6. MCFEV algorithm

The advantages of MCFEV clustering algorithm:

1. It permits the cluster to take the natural shape.

Table 1. Energy model

Figure 7. The results of MCFEV clustering algorithm

Figure 8. XB index for FCM and MCFEV algorithms

RMSE Performance Index

Network Life Time

Table 2. Consequent part of fuzzy rules for cluster heads election

-0.00175 X1+ -2.519 X2+ -2.98388 X3+ 3.611

Table 3. Consequent part of fuzzy rules for relay nodes election

-0.84538 X1+ -0.29259 X2+ 0.591 X3+ 0.922

Sum of Remaining Energy

Table 4. The FND and HND

Protocols 100 Nodes 150 Nodes 200 Nodes

Figure 12. the FND and HND metrics

Figure 13. Scenario1: sum of remaining energy

Figure 14. Scenario2: sum of remaining energy

KEY TERMS AND DEFINITIONS

You might also like