Energy Efficient Clustering and Data AggregationProtocol ...xajzkjdx.cn/gallery/61-june2020.pdf ·...
Transcript of Energy Efficient Clustering and Data AggregationProtocol ...xajzkjdx.cn/gallery/61-june2020.pdf ·...
Energy Efficient Clustering and Data AggregationProtocol using Machine
Learning in Wireless Sensor Networks
Sangeeta Kumari1*, Ph.D. Student, Maharishi University of Information Technology, Lucknow, UP
Dr. Rajat Updhyay2, Ph.D. Guide, Maharishi University of Information Technology, Lucknow, UP
Dr Vivek Deshapande3, Ph.D., Maharishi University of Information Technology, Lucknow, UP
Abstract: As the Wireless Sensor Networks (WSNs) are resource-constrained,
energy-efficient data transmission required by considering various applications. The
novel routing protocol designed to achieve the scalability, energy efficiency, QoS
optimization with minimum overhead in this research called Strong Clustering
Algorithm & Data Aggregation using Machine Learning (SCADA-ML). The
SCADA-ML design is mainly based on the use of machine learning techniques for
CH selection and robust data aggregation to minimize the energy consumption while
maintaining the other performances for different size of WSNs. In the first
contribution, we focused on optimal CH selection and cluster formation using the
supervised ML technique called Artificial Neural Network (ANN). The problem of
optimal CH selection for each cluster is formulated according to the architecture of
ANN (input layer hidden layer, and output layer) in which the every sensor node
properties such as residual energy, distance from the BS, and bandwidth allocated
are processed as input to ANN. At CH node, there may be the possibility of
redundant information, therefore in the second contribution the efficient data
aggregation performed by CH node of each cluster to minimize the energy
consumption using the Independent Component Analysis (ICA) ML technique. The
clusters with similar data need to perform the data aggregation. As compared to other
data aggregation methods, ICA is computation efficient and reduces the redundant
data based on differential entropy. The experimental results show that SCADA-ML
outperforms the existing ML-based clustering and data aggregation algorithms.
Keywords: Artificial neural network, clustering, cluster head selection, data
aggregation, data transmission, machine learning, independent component analysis.
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 628
I. Introduction
The Wireless Sensor Network (WSN) is consisting of sensor nodes. As a sensor
node has been small & have less power, energy consumption major concern &
deciding factor about network overall lifetime. Data collection, processing, sending,
receiving, forwarding processes consume more sensor node energy. Under WSN,
there has been several factors basis on which the energy efficiency has been
determined like the architecture of WSN, topology design, routing protocol, MAC
(medium access control) protocol, data aggregation schemes, etc [1]. The key terms
for WSNs have been energy efficiency & network lifetime defined as:
- Energy Efficiency: The processing of WSN should be extended as much as
possible. Under normal topology control protocol, every sensor consumes
similar energy for each network round either second. However, the topology
control protocol has been energy efficient if it has been able to extend the
overall network lifetime of WSNs. For every sensor node, energy
consumption should be minimized if we consider the fact that all sensor
nodes have been having similar importance.
- Network lifetime: The term network lifetime has been nothing but a number
of data collection rounds either overall life under minutes till to the first
sensor node under WSN dies. For example, under some WSN applications it
has been needed that operation of all sensor nodes should be done together,
then under such case lifetime has been nothing but the total number of rounds
of network till the first sensor node dies.
At the present time, energy consumption has been the most important research
problem for WSNs. Why energy efficiency has been necessary has been explained
as: 1) the single small radio sensor device has been having low power battery which
has been expected to operate several months after its deployment [2]. 2) If designing
& deployment of WSN has been done over the inaccessible region, then it’s needed
that all wireless sensor devices under such networks utilize their batteries efficiently
so that network lifetime should be extended.
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 629
The network layer protocols like routing methods have been the main component of
WSNs where the major operations performed & hence the energy consumption
concern among those operations. Several types of routing protocols designed to
achieve energy efficiency under WSNs [3] [4]. The clustering basis routing protocols
showed the better outcomes for energy efficiency compared to other techniques. The
cluster formation & CH selection has been also the NP-hard research problem & to
address the challenges various methodologies for optimal CH selection & cluster
formation presented so far. Utilizing clustering methods, data aggregation has been
another research problem under which the redundant data collected at CH node,
therefore there has been a need for appropriate method to fuse the common data
through discarding the duplicate information to reduce the transmissions as well as
energy consumption. The data aggregation has been performed through the CH node.
Creating productive calculations that has been reasonable for various application
situations has been a difficult errand. Specifically, WSN architects need to deliver
regular issues identified along information accumulation, information unwavering
quality, limitation, hub grouping, vitality mindful directing, occasions planning,
deficiency discovery, & security. ML was presented under the late 1950s as a
method for artificial intelligence (computer basis intelligence) [5]. After some time,
its center developed & moved more to calculations that has been computationally
reasonable & powerful. Under the most recent decade, AI systems have been utilized
broadly for a wide scope of errands including order, relapse & thickness estimation
under an assortment of utilize zones, for example, bioinformatics, discourse
acknowledgment, spam discovery, computer vision, misrepresentation identification,
& publicizing systems. ML strategies have been focal under creating WSN
applications since the beginning of WSNs, the same number of the issues under
WSNs could be put as advancement either displaying issues. ML under WSN assists
along finding important new connections, examples, & patterns, regularly already
obscure, through filtering through a lot of information, utilizing design
acknowledgment, factual & scientific procedures. It very well may be valuable not
just under information revelation, that has been, the recognizable proof of new
wonders, yet under addition it can help under improving our understanding of known
marvels. At the end of the day, AI systems can help assemble choice guide devices
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 630
& encourage investigating of sensor information acquired from WSNs. The two
concerns of WSNs like clustering (i.e. optimal CH selection) & data aggregation can
be solved for employing the ML techniques effectively as the key motivation of this
paper. From the literature review, it has been noticed that the various researchers
having various applications, preferences, & assumptions under utilizing ML
methods. Such differences & assumptions lead the main challenge for the other
researchers to build upon the present works. Hence, the generalized framework for
WSN machine learning has been necessary. In this paper, designing the generalized
ML-basis WSN algorithms (clustering & data aggregation) that can achieve the
tradeoffs between QoS performances & energy efficiency of WSNs regardless of
end-user applications is main objective of this research. In section II, brief review of
ML-based clustering and data aggregation methods presented. In section III, the
design of SCADA-ML protocol presented. In section VI, experimental results are
presented. In section V, conclusion and future work presented.
II. Related Works
This section presents the review of recent works of ML-based clustering and data
aggregation for WSNs.
A. ML-based Clustering
In [6]-[19], various algorithms designed for optimal clustering and CH selection
using the ML techniques. In [6], creator proposed productive cross breed vitality
mindful grouping correspondence convention for green IoT system registering; Hy-
IoT, yet likewise gives a genuine IoT organize design for looking proposed
convention contrasted with usually existed conventions. Effective bunch head
determination supports the utilization of the hubs vitality substance and thusly builds
the system life time and furthermore the bundles transmission rate to the base station.
Hy-IoT utilized different weighted race probabilities for choosing a Cluster-head in
context of heterogeneity dimension of the locale. In [7], creator revived Internet of
Things (IoT) gadgets is particularly used in different fields, for example, regular
checking, associations, fast home and so on. Under such occurrence, a group head is
chosen among the diverse IoT gadgets of WSN based IoT system to keep up the time
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 631
tested system with proficient information transmission. To achieve the productive
group head choice, they utilized Fuzzy C-Means (FCM) bunching count. In [8],
creator proposed IoT related issues mostly on vitality proficiency. The development
of these contraptions in a conveying inciting framework that makes the Internet of
Things (IoT) captivating, wherein sensors and actuators were blend reliably with
nature around us, and the information is shared across over stages to develop a
regular working picture.
In [9]-[15], the protocols designed for IoT applications. In [9], proposed the different
directing answers for buy in techniques that incorporate substance and setting based
steering for IoT empowered WSNs. They planned the Energy-Efficient Content-
Based Routing (EECBR) convention for the IoT that limits the vitality utilization in
WSNs. In [10] [11], creators talked about the requirement for green IoT and the
different software and equipment based advancements required to empower its
acknowledgment. Vitality proficient between hub correspondence and improved
steering systems has been recognized as the issues that should be routed to
encourage enormous scale appropriation of green IoT. In [12], the distinctive
methodology proposed wherein an evaluated open detecting structure is proposed for
open information conveyance accumulated from cloud and heterogeneous assets.
The work is information driven, centered on free market activity chain of open
information from cell phones. In [13], creator proposed information gathering in cell
gadgets utilizing gadget to gadget interchanges in an IoT and Smart City setting.
This outcomes in progressively productive asset use and limits vitality utilization.
They utilize one gadget that totals information from a few encompassing gadgets and
after that sends the information to cell station, rather than every gadget is sending
information independently. In [14], a diagram of utilizing customary WSN
conventions for accomplishing gadget to gadget correspondence in IoT has been
exhibited. In [15], the latest work proposed for the information total in WSNs
utilizing the fluffy c-implies bunching approach. They structured similitude mindful
information collection utilizing a fluffy c-implies approach. The fluffy c means used
to play out the bunching to sort out sensors into groups dependent on information
closeness. They utilized help degree work for the anomaly discovery.
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 632
In [16], author proposed novel energy efficient routing protocol using machine
learning algorithm. The reinforcement learning technique is used for energy
efficiency. In the first step of the protocol, a new clustering method is applied to the
network and the network is established using a connected graph. Then data is
transmitted using the 𝑄-value parameter of reinforcement learning technique. In
[17], author proposed novel machine learning based approach for the sensor node
validation. Authors designed validation sensors in the domain using spectral
clustering technique have been proposed that is detecting a bad sensor and deleting it
from the domain. Sensors have been indexed by their location using simple model of
spectral clustering. In [18], author recently introduced the new routing protocol for
OppNets called MLProph based on machine learning (ML) algorithms. The ML
techniques used in this work are neural networks and decision tree to discover the
probability of successful deliveries. TheMLmodel is trained by using various factors
such as the predictability value inherited from the PROPHET routing scheme,
popularity of nodes, energy consumption of node, location of node, and mobility
speed. In [19], author proposed robust distributed clustering method without a fusion
center. The algorithm combines distributed eigenvector computation and distributed
𝐾-means clustering. A distributed power iteration method is used to compute the
eigenvector of the graph Laplacian.
B. ML-based Data Aggregation
In [20], an alternate strategy has been proposed under the Center at Nearest (CNS)
calculation. Under this calculation, every center that recognizes an event sends its
data to a specific center point, called aggregator, through using a most short way. For
this situation, the aggregator has been the nearest center to the sink (under bounces)
that recognizes an event.
In [21], the Directed Diffusion calculation has been one of the soonest answers for
moreover proposed characteristic premise controlling. Under these cases,
information can be keenly totaled when they meet at any transitional center point.
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 633
In [22], under perspective on Directed Diffusion, the Greedy Incremental Tree (GIT)
approach was proposed. The GIT calculation develops essentialness beneficial way
and eagerly interfaces different sources onto the developed way.
In [23] and [24], information accumulation has been realized under a certified world
testbed and the Tiny Aggregation Service (TAG) structure has been introduced.
Name uses a most restricted way tree, and proposes upgrades, for instance, sneaking
around premise and hypothesis testing-premise progressions, dynamic parent
trading, and the utilization of an adolescent hold to evaluate information adversity.
In [25], author proposed an information conglomeration strategy for remote sensor
frameworks using fake neural frameworks. The information combination tree has
been developed to diminish the packs stream and can revive the leaf centers
intensely.
In [26] author introduced a joint arrangement of information assortment along the
coordinating innovation, and displayed a network premise guiding and aggregator
determination intend to achieve low essentialness dissemination and low idleness
without giving up quality. Through inquiring about information combination along
correspondence imperative between the combination center and each sensor.
In [27] author showed an information combination instrument for target following
under remote sensor frameworks reliant on quantized improvements and Kalman
filtering.Through including some defer time, every one of the information gathered
through transfer hub can be combined at one time under order to decrease the vitality
utilization.Planning to guarantee the information quality
In [28] author proposed various measurements for QoS (nature of administration)
during the time spent information total, including lifetime, information deferral, &
retransmission rate. Likewise, the methodology has been examined to guarantee
above QoS measurements under subtleties. Likewise to the tree-basis methodologies,
bunch basis plans additionally comprise of a various levelled association of the
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 634
system. Be that as it may, under these methodologies, hubs have been isolated into
groups.
III. SCADA-ML Protocol
This section presents the design of proposed SCADA-ML protocol for WSNs by
considering the design of ML-based clustering and ML-based data aggregation. The
SCADA-ML proposed for generalized solutions to enhance the network lifetime,
improve QoS performance with acceptable or minimum overhead for WSNs by
using the ML techniques. The SCADA-ML routing protocol proposed consists of
two contributions such as:
ML based Optimal CH Selection: In the main commitment, we concentrated
on optimal CH selection and bunch formation utilizing the directed ML
technique called Artificial Neural Network (ANN). The issue of optimal CH
selection for each group is formulated by the architecture of ANN (input
layer concealed layer, and yield layer) in which the each sensor hub
properties such as remaining energy, good ways from the BS, and data
transmission allocated are prepared as contribution to ANN. The usefulness
of the concealed layer performed to choose CH in the yield layer utilizing the
versatile learning of ANN. After the clustering, the between group and intra-
bunch data transmissions performed.
ML based Efficient Data Aggregation: At CH hub, there might be the
chance of repetitive information, along these lines in the second commitment
the efficient data aggregation performed by CH hub of each bunch to limit
the energy utilization utilizing the Independent Component Analysis (ICA)
ML technique. The scales with comparable data need to play out the data
aggregation. When contrasted with other data aggregation strategies, ICA is
computation efficient and diminishes the repetitive data based on differential
entropy.
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 635
A. ML-based Clustering and Data Transmission
In traditional methods of clustering in sensor network, cluster heads selections
are mainly rely on two parameters firstly residual energy and secondly distance
from base station node whereas in Machine learning based clustering, the cluster
head are elected on basis of cost metrics function. In traditional approach of
clustering of WSN, there are overcrowded cluster heads whereas in sparse area
there would be very few clusters heads or there is possibility that no cluster head
is present. Such methodology of clustering leads to reduce the network lifetime
which can be avoidedby using Artificial Neural network (ANN based approach)
for selection of Cluster head. For electing cluster heads ANN uses layered
architecture with primarily three types of layers input layer, hidden layer, and
output layers. Figure 1 shows the proposed ANN based CH election architecture.
Figure 1. Proposed ANN based CH selection approach
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 636
Artificial Neural Networks (ANN) could find applications in various domains of our
life including fraud detection, speech recognition, health care applications,
handwriting recognition, face recognition applications or pattern recognition
applications and many more. As ANN has good potential to provide the solutions of
numerous problems because of good learning abilities through hidden layers.This
paper provides an optimal solution for election of periodic cluster head through
Artificial Neural Network (ANN). ANN based cluster head selection is not only fast,
simple to implement and flexible rather it is adaptive too, which makes this
architecture most suitable for adhoc networks
As showing in figure 1, the two layer feed-forward NN designed to select the
optimal CH node from set of sensor nodes. The input layer consist of n number of
sensor nodes 𝑆 = {𝑆1,𝑆2, … , 𝑆𝑛} those are competing each other in hidden layer to
become the CH. In hidden layer, each node evaluated through its cost function
𝑓(𝑆𝑖, 𝐶𝑖), 1 ≤ 𝑖 ≤ 𝑛 which is computed using two parameters residual energy and
distance from BS node as:
𝑓(𝑆𝑖, 𝐶𝑖) = 𝐶𝐸𝑆𝑖 + (200 − 𝑑𝑖𝑠𝑡 (𝑆𝑖, 𝐵𝑆)) (1)
Where, 𝐶𝐸𝑆𝑖 denotes the current available energy of node 𝑆𝑖 and 𝑑𝑖𝑠𝑡 (𝑆𝑖, 𝐵𝑆)
computes the geographical distance from node 𝑆𝑖 to BS.
Algorithm 1 elaborates the process of Cluster head (CH) election. Artificial neural
network based competitive learning technique has been used in this paper for
evaluations of cost functions of all the sensor nodes. Each nodes of input layers as
shown in the figure represents sensor nodes of the cluster and its cost function value
computed prior applying the hidden layer.
At second layer which is hidden layer of neurons as shown in figure 1,
contendamong themselvesand neuron with highest cost function
value𝑓(𝑆𝑖, 𝐶𝑖)𝑖𝑠𝑐𝑎𝑙𝑐𝑢𝑙𝑎𝑡𝑒𝑑. As each neurons of hidden layer is based on the
adaptive learning technique where the learning rate ε decides the vector adaption
towards the input pattern and it is directly concerning with convergence. If learning
rate is zero (ε =0) means, it there is no learning. If learning rate is one (ε =1), it
means learning is very fast. Generally, the learning rate arekept fixed value over the
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 637
period of time. For periodic selection of optimal cluster head(CH) ,the process is
repeated at regular interval of time in a network.
Algorithm 1: ANN-based CH election
Input
𝑠 = {𝑠1,𝑠, … , 𝑠𝑛} ,
𝐿𝑒𝑎𝑟𝑛𝑖𝑛𝑔 𝑟𝑎𝑡𝑒 𝜀 𝑤ℎ𝑒𝑟𝑒 0 ≤ 𝜀 ≤ 1
Output
𝐶𝐻: 𝑒𝑙𝑒𝑐𝑡𝑒𝑑 𝐶𝐻 𝑓𝑟𝑜𝑚 𝑠𝑒𝑡 𝑤
Input Layer
1. Initialize 𝑤 = {𝑤1,𝑤, … , 𝑤𝑛} at input layer for the 𝐶𝐻 competition
2. Initialize the cost function for each sensor of input layer 𝑓(𝑤𝑖, 𝐶𝑖) = 0.
Hidden Layer
3. For each layer
4. Compute the cost function 𝑓(𝑤, 𝐶𝑖) for each sensor using Eq. (1)
5. 𝑓(𝑤𝑖, 𝐶𝑖)= neuron 𝑛𝑖 as weight function.
6. Neurons are competition each other with learning rate 𝜀
7. Update 𝑛𝑖 weight value after each learning 𝜀
𝑓(𝑤𝑖, 𝐶𝑖) = 𝑓(𝑤𝑖, 𝐶𝑖) + 𝜀(𝑤𝑖 − 𝑓(𝑤𝑖, 𝐶𝑖))
8. Estimate the 𝑛𝑖with highest cost function value
9. Repeat steps 4-8.
10. End For
Output Layer
11. The estimated neuron 𝑛𝑖labelled to actual sensor node 𝑤𝑖
12. 𝐶𝐻 = 𝑤𝑖
13. Return 𝐶𝐻
After the cluster formation, according to the objective function defined equation (2),
this section presents the route formation and data transmission functionality in
clustered WSN. The link cost is among two nodes 𝑤𝑖 and 𝑤𝑗 is defined as the
amount of energy consumed to send and receive the packet successfully. The link
cost model is represented as:
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 638
LC = (𝐸𝑖
𝐷
𝐸𝑡(𝑆𝑖,𝑆𝑗)+𝐸𝑟(𝑆𝑖,𝑆𝑗)) (2)
Where 𝐸𝑖𝐷is energy associated with the delivery ratioof the packet originating from
source node 𝑆𝑖and correctly received at destination node, while𝐸𝑡 (𝑆𝑖, 𝑆𝑗) is the
energy used in transmitting from 𝑆𝑖 to 𝑆𝑗and 𝐸𝑟 (𝑆𝑖, 𝑆𝑗) is the energy used in
receiving thepacket. Data routing from every CH to the sink is done over multi-hop
paths, which is given byminimizing equation (2).
Consider that BS node labeled as 0, and all the current CH nodes labeled as 𝐶𝐻𝑖,
where 𝑖 = 1, 2, … 𝑘. The problem is formulated as with objective:
Minimize ∑ 𝐿𝐶1≤𝑖≤𝑘
Subject to following constraints
∑ 𝑃𝑖𝑗1≤𝑗≤𝑘 - ∑ 𝑃𝑗𝑖1≤𝑗≤𝑘 = 𝑝𝑖 (3)
𝑃𝑖𝑗 ≥ 0, 1 ≤ 𝑗 ≤ 𝑘, (4)
𝑝 <= 𝐸𝑡ℎ𝑟 (5)
Where, the constraint (3) represents the amount of data transmitted 𝑝𝑖, constraint (4)
represents the 1 or more packets to be transmitted among two nodes. The constrain
(5) limits the maximum energy consumption 𝑝 of any sensor node should be below
the predefined energy threshold value 𝐸𝑡ℎ𝑟.
B. ML-based Data Aggregation
This is another contribution for SCADA-ML routing protocol. Data aggregation
leads the reduced number of transmissions which further optimizes the energy
efficiency and QoS performances.
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 639
Figure 2. ICA-based data aggregation in SCADA-ML protocol
For data aggregation phase, we proposed the ML-based approach in this work. In
this work, after the node deployment, the clustering performed using the ANN based
approach. Further in data aggregation phase, the data similarity among the
neighbouring sensor nodes computed in the form of data correlation. Using the
predefined value, the decisions regarding to whether the two data items belongs to
cluster with similar data or not. If the correlation value more than set threshold value,
then two data items is belong to cluster with same data. The data aggregation applied
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 640
on only the similar data clusters to prevent the loss of information. The data clusters
with different data perform the data transmission operations without applying the
data aggregation technique.
The Independent Component Analysis (ICA) algorithm proposed for cluster-based
data aggregation in this work which is applied on similar data cluster. The ICA
applied by the CH node. The ICA method performs efficiently as compared to other
techniques like PCA and compression methods. Itminimizes mutual information
using the concept of differentialentropy. Aggregated data from similar clustersare
forwarded to sink node. Hence the computation andenergy consumption can be
reduced due to less numberof aggregation processes. Figure 2 shows the working of
ICA-based data aggregation in SCADA-ML protocol.
III. Experimental Results
This section presents the simulation results of SCADA-ML protocol in two sections.
In section A, we present the comparative analysis with existing ML-based clustering
protocols. In section B, we present the comparative analysis with existing ML-based
data aggregation protocols. The simulation parameters used for both studies are
described in table 1. The performances are measured in terms of average throughput,
average delay, Packet Delivery Ratio (PDR), average energy consumption, and
network lifetime.
Table 1. Density variation parameters
Sensor nodes 40-400
BS position (150, 160)
Simulation Time 150 seconds
Network size 150 x 150
MAC 802.11
Packet size 512 bytes
Sensor nodes transmission range 100 m
Channel bit rate 10 kbps
Initial energy 5000 nJ
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 641
Mobility Model Random Way-point model
Sensor speed 0 m/s
Packet sending rate 20 packets/sec.
A. ML-based Clustering Protocols Evaluation
The proposed clusteringalgorithm SCADA-ML is compared with existing ML-based
clustering protocols such as Decision Tree (CHDT) [29], clustering using Q-learning
(QLIQUE) [30], and recent clustering using K-means (CHKM) [31] protocols.
Figure 2. Average throughput analysis vs. density of ML-based clustering methods
This section presents the performance investigation of proposed SCADA-ML (with
ML-based data aggregation technique) with the various ML-based clustering
methods. The figure 2 and 3 demonstrate the average throughput and PDR
performances with varying number of sensor nodes. From these results it is noticed
that increased density leads to increased throughput and PDR performances using all
the ML-based clustering techniques. The SCADA-ML protocol delivered the
improved performances for throughput and PDR compared to other techniques due
to innovative clustering and data transmission algorithms proposed. In SCADA-ML,
along with optimal CH selection the focus is on the optimal data transmission as well
by considering the energy and distance constraints. Among the other ML-based
0
20
40
60
80
100
120
40 80 120 160 200 240 280 320 360 400
Ave
rage
th
rou
ghp
ut
(kb
ps)
Number of Sensor nodes
CHDT QLIQUE CHKM SCADA-ML
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 642
clustering methods, the recent K-means algorithm initiated clustering having better
performances than Q-learning and decision tree machine learning algorithms.
Figure 3. PDR analysis vs. density of ML-based clustering methods
Figure 4. Average end to end delay analysis vs. density of ML-based clustering
methods
Figure 4 showing the outcomes of average end to delay using different ML-based
clustering protocols. The delay increasing significantly with increased number of
sensor nodes in network due to long links establishment and increased number of
data transmissions. The SCADA-ML protocol shows the reduction in delay
0
20
40
60
80
100
120
40 80 120 160 200 240 280 320 360 400
PD
R (
%)
Number of Sensor nodes
CHDT QLIQUE CHKM SCADA-ML
0
0.5
1
1.5
2
2.5
3
40 80 120 160 200 240 280 320 360 400
De
lay
(Se
con
ds)
Number of Sensor nodes
CHDT QLIQUE CHKM SCADA-ML
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 643
compared to existing methods as the data transmission performed through the robust
link optimisation technique along with the optimal CH selection compared to
existing methods. The another point observed in these results that QLIQUE shows
the less delay performance compared to CHKM protocol as the Q-learning takes less
time for operations compared to K-means.
Figure 5. Average energy consumption analysis vs. density of ML-based clustering
methods
Figure 6. Network lifetime analysis vs. density of ML-based clustering methods
Figure 5 and figure 6 demonstrates the energy efficiency performances of ML-based
clustering protocols in terms of average energy consumption and network lifetime
0
200
400
600
800
1000
1200
40 80 120 160 200 240 280 320 360 400Ave
rage
co
nsu
me
d e
ne
rgy
(nJ)
Number of Sensor nodes
CHDT QLIQUE CHKM SCADA-ML
0
2000
4000
6000
8000
10000
12000
40 80 120 160 200 240 280 320 360 400
Ne
two
rk L
ife
tim
e (
Ro
un
ds)
Number of Sensor nodes
CHDT QLIQUE CHKM SCADA-ML
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 644
respectively. The energy consumption and network lifetime performances are
correlated to each other, it means if the energy consumption increased, then the
network lifetime decreased. From both outcomes of energy efficiency, it shows the
SCADA-ML achieved the significant improvement in energy efficiency performance
over the existing protocols as the energy aware cost function designed to establish
the data forwarding in network as well as optimal CH selection performed using
energy as core parameter. The other ML-based clustering methods only work on
optimal CH selection using ML technique by exploiting energy parameter of sensor
nodes and do not bother about the energy efficient data transmission.
B. ML-based Data Aggregation Protocols Evaluation
In this section, SCADA-ML with data aggregation is compared with other existing
ML-based data aggregation techniques such as data aggregation with self
organization map (CODA) [32] and data aggregation using PCA (DAPCA) [33].
Figure 7. Average throughput analysis vs. density
0
50
100
150
200
250
40 80 120 160 200 240 280 320 360 400
Ave
rage
th
rou
ghp
ut
(kb
ps)
Number of Sensor nodes
CODA DAPCA SCADA-ML
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 645
Figure 8. PDR analysis vs. density of ML-based aggregation methods
The last model of SCADA-ML investigated in this segment with consideration of
ML-based aggregation technique alongside ML-based clustering calculation. The
presentation of SCADA-ML is contrasted and late ML-based aggregation techniques
such as CODA and DAPCA. The figure 7 and 8 demonstrate the normal throughput
and PDR exhibitions with differing number of sensor hubs for each aggregation
strategy. These outcomes show that with expanded thickness, the throughput and
PDR exhibitions increments for all investigated techniques. The SCADA-ML
convention conveyed the improved throughput and PDR contrasted with different
techniques due ICA-based aggregation calculation acquainted with diminish the
quantity of transmissions and ANN-based optimal CH selection technique prompts
stable bunches in network. The optimal data transmission utilizing the connection
cost optimization promotes to improve the exhibitions of SCADA-ML.
Figure 9 demonstrates average end to delay results using different ML-based
aggregation protocols. Here the delay performance becomes worst with increased
number of sensor nodes in network due to long links establishment and increased
number of data transmissions. The delay performance of SCADA-ML is efficient
compared to CODA and DAPCA algorithms as the ICA technique is more robust,
efficient, and fast for SCADA-ML protocol.
0
20
40
60
80
100
120
40 80 120 160 200 240 280 320 360 400
PD
R (
%)
Number of Sensor nodes
CODA DAPCA SCADA-ML
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 646
Figure 9. Average end to end delay analysis vs. density of ML-based aggregation
methods
Figure 10. Average energy consumption analysis vs. density of ML-based
aggregation methods
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
40 80 120 160 200 240 280 320 360 400
De
lay
(Se
con
ds)
Number of Sensor nodes
CODA DAPCA SCADA-ML
0
200
400
600
800
1000
1200
40 80 120 160 200 240 280 320 360 400
Ave
rage
co
nsu
me
d e
ne
rgy
(nJ)
Number of Sensor nodes
CODA DAPCA SCADA-ML
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 647
Finally the figure 10 shows the average energy consumption performance of all the
ML-based aggregation techniques for each WSN. The results shows that the
SCADA-ML delivered the significant improvement in energy efficiency
performance over the existing protocols as the energy aware link cost function
designed to establish the data forwarding in network, optimal CH selection
performed using energy as core parameter, and effective ICA-based data aggregation
functions.
V. Conclusion and Future Work
The design of the SCADA-ML protocol neatly presented and described in this paper.
The ANN technique effectively used in the process of optimal CH selection and
cluster formation in SCADA-ML. The neurons weights computed using the sensor
nodes properties such as residual energy, geographical distance, and bandwidth
availability. After the clustering, the inter-cluster and intra-cluster data transmissions
performed via the link cost-based optimal path selection process to ensure the
reliability and QoS performance in WSNs. Finally, to reduce the energy
consumption, the ICA based data aggregation algorithm designed in SCADA-ML
which shows superior behaviors compared to other ML-based methods for data
aggregation. The simulation results presented with varying density and data rate
parameters to claim the scalability and reliability performances of SCADA-ML
compared to exiting ML-based clustering and data aggregation methods. The
SCADA-ML shows the improved performances for throughput, delay, PDR, and
energy efficiency over all the investigated algorithms.
References
1. Cagalj, M., Hubaux, J.-P., and Enz, C. C., “Energy-efficient broadcasting in
all-wireless networks," Wireless Networks, 11(1/2), 177–188, 2005.
2. Polastre, J., Szewczyk, R., and Culler, D., “Telos: Enabling ultra-low power
wireless research,” In Proceedings of international symposium on
information processing in sensor networks (pp. 364–369), 2005.
3. Chen, Y. P., Wang, D. and Zhang, J., “Variable-base tacit-communication: a
new energy efficient communication scheme for sensor networks,”
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 648
Proceedings of the First International Conference in Integrated Internet Ad
Hoc and Sensor Networks, InterSense 2006, Nice, France, May 30-31, 2006.
4. Chen, Y. P., Liestman, A. L., & Liu, J., “Energy-efficient data aggregation
hierarchy for wireless sensor networks,” In Proceedings of 2nd international
conference on quality of service in heterogeneous wired/wireless networks
(QShine ’05), Orlando, 2005.
5. Bharat Sundararaman, Ugo Buy and Ajay D. Kshemkalyani, “Clock
Synchronization for Wireless Sensor Networks: A Survey”, Journal of Ad
Hoc Networks, Vol.3,2005, pp. 281-323.
6. Rowayda A.Sadek, “Hybrid energy aware clustered protocol for IoT
heterogeneous network” https://doi.org/10.1016/j.fcij.2018.02.003, 2018.
7. Praveen Kumar Reddy,RajasekharaBabu“An Evolutionary Secure Energy
Efficient Routing Protocol in Internet of Things” Vellore Institute of
Technology University, Vellore, Tamil Nadu, India, 2017.
8. Vellanki M, Kandukuri SPR and Razaque A “Node Level Energy Efficiency
Protocol for Internet of Things” Vellanki et al., J Theor. Comput.Sci. 2017.
9. Samia Allaou, Chelloug, “Energy-Efficient Content-Based Routing in
Internet of Things,” Journal of Computer and Communications, 2015, 3, 9-
20Published Online December 2015 in SciRes.
10. Faisal Karim Shaikh, Sherali Zeadally, and Ernesto Exposito, “Enabling
technologies for green internet of things,” IEEE Systems Journal,11(2):983–
994, 2017.
11. Chunsheng Zhu, Victor CM Leung, Lei Shu, and Edith C-H Ngai, “Green
internet of things for smart world,” IEEE Access, 3:2151–2162, 2015.
12. Al-Fagih, A.E.; Al-Turjman, F.M.; Alsalih, W.M.; Hassanein, H.S. A priced
public sensing framework for heterogeneous IoT architectures. IEEE Trans.
Emerg. Top. Comput. 2013, 1, 133–147.
13. Orsino, A.; Araniti, G.; Militano, L.; Alonso-Zarate, J.; Molinaro, A.; Iera, A.
Energy efficient iot data collection in smart cities exploiting D2D
communications. Sensors 2016, 16, 836.
14. Bello, O.; Zeadally, S. Intelligent device-to-device communication in the
Internet of things. IEEE Syst. J. 2016, 10, 1172–1182.
15. Runze Wan1, Naixue Xiong, Qinghui Hu, Haijun Wang, and Jun Shang,
“Similarity-aware data aggregation using fuzzy c-means approach for
wireless sensor networks,” EURASIP Journal on Wireless Communications
and Networking, (2019) 2019:59.
16. Farzad Kiani, Ehsan Amiri, Mazdak Zamani, Touraj Khodadadi, and Azizah
AbdulManaf, "Efficient Intelligent Energy Routing Protocol in Wireless
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 649
Sensor Networks", Hindawi Publishing Corporation, International Journal of
Distributed Sensor Networks, 2015
17. Abdo M. T. Nasser, V. P. Pawar, "Machine Learning Approach for Sensors
Validation and Clustering", International Conference on Emerging Research
in Electronics, Computer Science and Technology – 2015
18. Deepak K. Sharma, Sanjay K. Dhurandher, Isaac Woungang, Rohit K.
Srivastava, Anhad Mohananey, and Joel J. P. C. Rodrigues, "A Machine
Learning-Based Protocol for Efficient Routing in Opportunistic Networks",
IEEE Systems Journal ( Volume: 12, Issue: 3, Sept. 2018 )
19. Gowtham Muniraju, Sai Zhang, Cihan Tepedelenlio?glu, Mahesh K.
Banavar, "Location Based Distributed Spectral Clustering for Wireless
Sensor Networks", Sensor Signal Processing for Defence Conference
(SSPD), 2017.
20. B. Krishnamachari, D. Estrin, and S.B. Wicker, “The Impact of Data
Aggregation in Wireless Sensor Networks,” Proc. 22nd Int’l Conf.
Distributed Computing Systems (ICDCSW ’02), pp. 575-578, 2002.
21. C. Intanagonwiwat, R. Govindan, D. Estrin, J. Heidemann, and F. Silva,
“Directed Diffusion for Wireless Sensor Networking,” IEEE/ACM Trans.
Networking, vol. 11, no. 1, pp. 2-16, Feb. 2003.
22. C. Intanagonwiwat, D. Estrin, R. Govindan, and J. Heidemann, “Impact of
Network Density on Data Aggregation in Wireless Sensor Networks,” Proc.
22nd Int’l Conf. Distributed Computing Systems, pp. 457-458, 2002.
23. E.F. Nakamura, H.A.B.F. de Oliveira, L.F. Pontello, and A.A.F. Loureiro,
“On Demand Role Assignment for Event-Detection in Sensor Networks,”
Proc. IEEE 11th Symp. Computers and Comm. (ISCC ’06), pp. 941-947,
2006.
24. S. Madden, M.J. Franklin, J.M. Hellerstein, and W. Hong, “Tag: A Tiny
Aggregation Service for Ad-Hoc Sensor Networks,” ACM SIGOPS
Operating Systems Rev., vol. 36, no. SI, pp. 131-146, 2002.
25. L.Y. Sun, X.X. Huang, W. Cai, Data aggregation of wireless sensor
networks using artificial neural networks. Chinese Journal of Sensors and
Actuators. 24(1), 122–127 (2011).
26. J.N. Aikaraki, R. Uimustafa, A.E. Kamal, Data aggregation and routing in
wireless sensor networks: optimal and heuristic algorithms. Comput. Netw.
53(7), 945–960 (2009).
27. J. Xu, J.X. Li, S. Xu, Data fusion for target tracking in wireless sensor
networks using quantized innovations and Kalman filtering. Science China:
Information science edition. 55(3), 530–544 (2012).
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 650
28. H. Li, H.Y. Yu, Research on data aggregation supporting QoS in wireless
sensor networks. Application Research of Computers. 25(1), 64–67 (2008).
29. G. Ahmed, N. M. Khan , Z. Khalid and R. Ramer, Cluster head selection
using decision trees for Wireless Sensor Networks, IEEE International
Conference on Intelligent Sensors, Sensor Networks and Information
Processing, 2008.
30. A. Forster and A. L. Murphy, CLIQUE: Role-Free Clustering with Q-
Learning for Wireless Sensor Networks,29th IEEE International Conference
on Distributed Computing Systems,2009.
31. G. Muniraju, S. Zhang, C. Tepedelenlioglu and M. K. Banavar Location
Based Distributed Spectral Clustering for Wireless Sensor Networks, IEEE
Sensor Signal Processing for Defence Conference (SSPD), 2017.
32. SangHak Lee and TaeChoong Chung, Data Aggregation for Wireless Sensor
Networks Using Self-organizing Map,Springer-Verlag Berlin Heidelberg
2005.
33. A. Morell, A. Correa and M. Barceló and J. L. Vicario, Data Aggregation and
Principal Component Analysis in WSNs,IEEE Transactions on Wireless
Communications, vol.15, Issue.6,PP.3908-3919, 2016.
Journal of Xi'an University of Architecture & Technology
Volume XII, Issue VI, 2020
ISSN No : 1006-7930
Page No: 651