Multi-objective Cluster-head-based Energy-aware Routing using a Separable Convolution
Neural Network in a Wireless Sensor Network
AtherDanish1,*
MaryJ. Prisca2
SinghPooja3
GargKanika4
PriyadharshiniT.R.5
PrincyB. Anni6
TiwariMohit7
-
( Department of IT and Engineering, Amity University in Tashkent, Uzbekistan danishather@gmail.com)
-
( Department of Computer Science and Engineering, NPR College of Engineering & Technology,
Dindigul, Tamil Nadu, India ramana3483@gmail.com)
-
( Department of Physics, Faculty of Science, SGT University, Gurugram, Haryana, India
pujasingh0409@gmail.com)
-
( Department of Computer Science and Engineering, SRM Institute of Science and Technology,
NCR Campus, Modinagar, Ghaziabad, India kanikagarg.kg@gmail.com)
-
( Department of Information Technology, Hindusthan College of Engineering and Technology,
Pollachi main road, Coimbatore, India Mail Id: cse.priyadharshinitr@gmail.com)
-
( Department of Computer and Communication, Panimalar Engineering College, Chennai,
Tamil Nadu, India ccehod@panimalar.ac.in)
-
( Department of Computer Science and Engineering, Bharati Vidyapeeth’s College of Engineering,
Delhi, India mohit.tiwari@bharatividyapeeth.edu)
Copyright © The Institute of Electronics and Information Engineers(IEIE)
Keywords
WSN, Pelican optimization algorithm (POA), Cluster head (CH), Separable convolution neural network (SCNN)
1. Introduction
Wireless sensor networks (WSNs) have received more attention throughout the world
with the proliferation of Micro-Electro-Mechanical Systems (MEMS) that are used for
the development of sensors [1]. The results of testing done in homes and offices come out well, and MEMS are going
to make their separate ways onto the market. In future, there will be more developments
in homes and offices [2]. WSNs are used in many business activities, like surveillance, decision making, health
care, monitoring, artificial intelligence, machine learning, virtual reality, infrastructure,
and smart homes. To make business policies, predictions, forecasts, data collections,
storage, and longtime observations, a stakeholder’s network has been used [3]. WSN is a set of multiple sensor nodes dispersed across a geographic area with stable
battery power supplying attached to each sensor node (SN). The node can work until
the battery gets depleted, and nodes die early in a WSN because energy depletion causes
network holes [4,5].
1.1 Problem Statement and Motivation
The energy used for data transmission in a WSN must be kept to a minimum. Various
approaches have been used to deal with this issue, but most of them focus primarily
on the energy factor, and pay less attention to other crucial factors like delay,
distance, cluster density, and traffic rate. Some protocols face challenges in choosing
ways that give the highest throughput and the least amount of latency while finding
the best possible paths [6-9]. The objective and motivation behind this work is to increase the lifespan of a wireless
sensor network by using energy-efficient routing, appropriate cluster head selection,
and optimal clustering.
1.2 Contributions
This work’s contributions can be understood in two ways: (i) the proposed method utilizes
a Separable Convolution Neural Network (SCNN) to deliver a solution to the issue of
WSN lifespan expansion, and (ii) multiple objective clustering is applied with efficient
fitness function to choose the most effective CH. Every cluster has CH dependent on
a number of factors, such as proximity, communication cost, residual energy, coverage.
POA is applied to choose the best routes. Finally, the WSN successfully transfers
the data to the base station (BS).
Continual paper is structured as: literature review is depicted in part 2, the proposed
methodology is explained in part 3, the results are proved in part 4, and the conclusion
is depicted in part 5.
2. Literature Review
Numerous studies in the literature related to multiple-objective cluster heads in
wireless sensor networks; a few works are reviewed here,
Mehta and Saxena [6] suggested an energy-aware optimized routing method depending on a multiple-objective
cluster head in a WSN. For the best CH based energy-aware routing protocol, the multi-objective
Hybrid Woodpecker with Flamingo Search Optimization Approach (Hyb-WFSOA) executes
an Internet of Things routing process through CH selection (CHS). CH selection requires
the same amount of energy as the sensors, and Hyb-WFSOA-CHS-IoT transmissions were
scrutinized for distance, delay, energy consumption, and throughput, providing a maximum
network lifetime along with a minimum packet delivery ratio (PDR). Nabavi and Najafi
[7] presented ideal CH selection and the consolidation of the Multi-objective Grasshopper
Optimization approach and Harmony Search in WSNs. An interesting strategy to reduce
energy consume was clustering the nodes and choosing the cluster head based on the
available transfer factors. As a result, the network lifespan was extended while the
nodes' average energy usage dropped. The presented approach provides a distinctive
optimization technique for WSN clustering. After the CH was chosen under grasshopper
optimization, data was sent between the nodes of the cluster head and the sink node
using nearly optimal routing based on the harmony search. It provides high energy
consumption and low throughput. Alghamdi [8] suggested dependable data storage in diverse WSNs by combining optimized routing
and storage node exploitation. The presented method makes an effort to create a novel
clustering model that has the best CH selection by deeming 4 key factors: energy,
latency, distance, security. To select the best CH, a hybrid dragonfly and firefly
algorithm was presented. It provides maximum packet delivery ratio and minimum network
lifetime.
Arunachalam et al. [9] presented clustering depending on energy-efficient spider monkey optimization with
data aggregation for WSN. Their energy-efficient routing protocol was established
for the chosen cluster head, which helps decrease the network’s energy consumption.
The classic Bellman-Ford algorithm was employed to determine best paths, and although
it has high energy consumption, it provides high throughput.
Sha et al. [10] suggested multi-objective-derived energy-efficient routing in a WSN by utilizing
a hybrid African vultures-cuckoo search optimization process to enhance the network
lifespan. Although it has high energy consumption, it provides a high packet delivery
ratio.
Chaurasia et al. [11] suggested a routing algorithm depending on meta-heuristic optimized CH selection.
It activates two sub processes: the optimum CH Selection Algorithm (CHSA) and the
Route Search Algorithm. Under node density, residual energy, distance betwixt the
CH and the BS, inter-cluster formation, CHSA employs an energy level matrix. It obtains
a longer network lifetime but with high energy consumption.
Biradar and Mathapathi [12] presented Safety and Energy-aware Clustering-based Routing at WSN. A hybrid nature-incited
approach to choosing the optimum CH introduces a cluster-based routing model into
the WSN, wherein it chooses a CH with the help of an optimization approach. A hybrid
seagull rock swarm with opposition-based learning was also suggested. The ideal CH
selection depends upon distance, security, and energy, and it has high energy consumption,
but a high PDR.
Debasis et al. [13] presented the Energy-Efficient Clustering Approach (EECA) for upgrading the lifespan
of a WSN under machine learning. EECA lengthens the WSN lifespan by lessening the
energy consumed in sensor nodes. The target area under EECA is viewed as a collection
of minor regions. It has high energy consumption and low throughput.
3. Proposed Methodology
Energy-aware routing utilizing the MOCH-based Separable Convolution Neural Network
Pelican Optimization Algorithm in a WSN (MOCH-SCNN-POA-WSN) is proposed in this manuscript
[14]. The proposed methodology has clustering, multiple-objective cluster head selection,
and optimization-based route recognition as the three main aspects of data transmission.
Four stages implement them: network structure, shared energy modeling, node clustering,
route finding. POA determines better routes for transmitting data during the route
identification step [15]. The proposed MOCH-SCNN-POA-WSN approach is shown in Fig. 1.
Fig. 1. Illustration of the proposed MOCH-SCNN-POA-WSN approach.
3.1 Network Setup
Here, random sensor positioning uses 2 dimensional Cartesian scheme. Initially, non-rechargeable
power sources drive the sensor nodes. Secondly, altering locations is not possible
once nodes are installed. Here, every sensor node has the same initial energy, and
hence, sensor nodes spread information to various locations between the transmission
links.
3.2 Energy Modeling
Energy for radio electronics is supplied through the receiver. The transmitter supplies
energy for both power amplifiers and radio electronics. When the distance (denoted
‘dist’) amongst the sender and receiver is higher than the set threshold, $'dist^{2'}$
is a square of the distance for energy dissipation in free space mode, and $'dist^{4'}$
denotes the energy dissipated in the multi-path-fading paradigm. Consequently, the
energy usage for packet transmission is shown in Eq. (1):
The electronic circuit of the SN is denoted $a_{pq}$, which is used for energy simulation.
Multiple path fading is $a_{rst}*dist^{2}$ or $a_{lmn}*dist^{4}$. Aggregating the
data energy spent at the CH is expressed with Eq. (2):
where b denotes bits in the packet, c is the message count, and $a_{Epqq}$ is the
quantity of energy aggregated.
3.3 Cluster Formation with CH Selection
In the network, unequal nodes are clustered and connected. If there are fewer unequal
clusters, there are various advantages, such as longer lifetimes, improved load balancing,
scalability, and less overload on the nodes. The remaining equal clusters are reformed,
and broadcast their messages through the network to the BS. By using messages, the
BS estimates the range of the network, and if it is a short range the message is analyzed
and compared with energy utilization. Cluster formation and cluster head selection
is explained below.
3.3.1 Network Area
The network area consists of parameters for width, $W_{dth}$, and height, $H_{ght}$,
with the total number of layers, $T_{l}$, calculated with Eq. (3):
where $Q$ is number of nodes. At the first level, some quantities are formed by sub
layers. The nearest layer to the BS is determined with Eq. (4):
where $T_{D{L_{1}}}$ denotes the number of clusters altered, $N$ is the size of the
sensor network, and $e^{2}toBS$ denotes the distance of average square from every
nodes to base station. The BS obtains the addresses of all nodes and sends control
messages to the nodes.
3.3.2 Cluster Formation
To identify adjacent sensor nodes, neighboring sensor nodes must be able to find them.
In finding nodes, clusters are formed. Messages from the base station consist of each
node’s location, its cluster ID, node ID, activation flag, and layer ID. The process
of forming more nodes is depicted in Eq. (5):
where $Location_{K}$ and $Location_{M}$ denote the locations of two nodes classified
by an adjacent node based on distance and given in Eq. (6);
If inactivate any sensor node, then the formation of cluster node is implemented.
3.3.3 Cluster Head Selection
Cluster head selection is made in a probabilistic manner using the Separable Convolution
Neural Network. The best node of the SCNN selects the CH depending on residual energy,
coverage, communication cost, and proximity. The SCNN consists of frameworks such
as Keras and Tensor Flow with a fulfilled convolution. Functions of the separable
convolutions are executed for channel-wise spatial convolution performance of the
nodes. In this section, the SCNN adopts the convolution technique denoted in Eq. (7):
Let $s$ is the width of a sensor node, $Z(i)$ is the input signal to the SCNN, and
$l$ is the length of a sensor node. By applying values to length $l$, the values are
derived. The depth-wise separable convolution calculation is shown in Eq. (8):
where $L^{\wedge }$ denotes a separable convolution,$j,k,m,n$ denote convolutions
in the kernel, and $G$ is the output of the $n^{th}$ channel on $H\wedge $.
CH tasks like transmission, data collection, and receiving from sensing nodes use
more energy than the other nodes. The SCNN is used for the multi-objective CH selection
procedure consisting of the following steps. The fitness factors for effectual clustering
involve distance and energy. The closeness of neighboring nodes depends on distance
as expressed in Eq. (9):
where $e$ represents all the nodes, $R$ represents all the cluster heads, $T^{O}$
denotes a normal node, a cluster head is $T^{d}$, a sink node is $T^{l}$, ${T}_{i}^{d}$
denotes the $i^{th}$ cluster head, and ${T}_{k}^{o}$ denotes the $k^{th}$ node. Energy
is needed to forward data among the nodes and cluster heads so it must be at maximum.
Energy is calculated as cluster energy, expressed in Eq. (10):
where ${R}_{D}^{B}(g)$ specifies the energy of a cluster head, ${\sum }_{g=1}^{p}{R}_{D}^{B}(g)'$
specifies the cumulative energy in all the cluster heads, and $P\times Ma{x}_{g=1}^{P}[\alpha
({R}_{g}^{m})]\times Ma{x}_{g=1}^{P}[\alpha ({R}_{g}^{B})]$ is a product of the overall
cluster heads.
3.4 Path Selection using the POA
The POA determines the best routes for data transmission over the network. This is
a new, stochastic, nature-incited optimization approach. Throughput is estimated with
Eq. (11):
when data are transmitted to the BS from a CH, with the maximum amount labeled $CH\left[EV_{CHj}\left(LOC_{C{H_{j}}},LOC_{BS}\right)\right]$.
CH data transmission speed is $EV_{CHj}$, with the positioning of the CH and the BS
denoted $\left(LOC_{C{H_{j}}},LOC_{BS}\right)$. The energy of a CH is predicted by
using Eq. (12):
The major concept in designing the proposed POA is to simulate the natural behavior
of pelicans while they are hunting. The vital phases of the PAO-based routing process
are path selection, data transmission, path alteration.
By choosing the best adjacent CH, path selection is made. Moreover, the POA updates
potential solutions by simulating the tactics and behavior of pelicans during attacks
and while hunting. The path selection history is determined with Eq. (13):
where the recently chosen CH's position is depicted $C{H}_{j}^{head}$, the location
of the CH already designated as best and the initial cluster head, is $W^{i}CH_{old}$,
and $z$ is a random integer between 0 and 1. The coefficient of the $jth$ iteration
is $\lambda _{j}$, which is determined with Eq. (14):
Here, PD is path density.
3.4.1 Path Alteration
Once the best route has been chosen from a variety of options, data communication
begins. This might result in a CH's demise or a decrease in its vitality. Path alterations
are determined with Eq. (15):
This also determines if any CH’s energy falls below the minimum threshold.
4. Results and Discussion
This segment describes the proposed MOCH-SCNN-POA-WSN approach. The simulations were
executed in MATLAB. The performance metrics is analyzed. Parameters and values utilized
during simulation are shown in Table 1.
Table 1. Parameters and values for the simulation.
|
Parameter
|
Utilized value
|
|
Network zone
|
200 m × 200 m
|
|
Nodes
|
500
|
|
Number of clusters
|
5
|
|
Initial node energy
|
20 J
|
|
Simulation time
|
|
|
Packet size
|
10,000bits
|
|
Max network throughput
|
1Mbps
|
|
Sink position
|
(50,100)
|
|
Node range
|
30-40 m
|
|
Packet size
|
1024 bytes
|
|
Number of rounds
|
3000
|
|
Basic routing protocol
|
LEACH
|
4.1 Performance Metrics
4.1.1 Network Lifetime
A number of cycles before the field nodes cease to function. It is determined by Eq.
(16):
where$PM_{ij}$ is the matrix coverage, $Q_{i}$ represents the lifetime of the SN,
and $R_{j}$ denotes nodes in coverage area K.
4.1.2 Energy Consumption
The energy count used through the sensor nodes is computed by Eq. (17):
where the energy used by the CN is denoted $CN_{Q}(s)$, and the cluster member uses
energy denoted $CM_{Q}(t)$.
4.1.3 Throughput
Throughput is the amount of data transfer to the BS from the nodes, and is scaled
by Eq. (18):
With 100 to 500 nodes, the data transmit at a 0.97 Mbps maximum speed and a 0.80 Mbps
minimum speed.
4.1.4 Packet Delivery Ratio
For an increasing number of nodes, the ratio of messages sent to the BS is expressed
in Eq. (19):
4.2 Performance Comparisons
Figs. 2-5 depict the performance metrics. The proposed technique performance is compared
with existing MOCH-EORA-WSN and MOCH-MOGOH-WSN models.
Fig. 2 depicts the network lifespan analysis. The proposed MOCH-SCNN-POA-WSN attained 72.65%
and 39.63% longer network lifespans for round 0, 66.96% and 27.63% longer lifespans
for round 1200, 39.63% and 59.63% longer network lifespans for round 2400, and 46.63%
and 62.29% longer network lifespans for round 3000, compared with MOCH-EORA-WSN and
MOCH-MOGOH-WSN, respectively. The increased network lifespans are attributed to the
integration of energy consumption rate components in the fitness function for MOCH-SCNN-POA-WSN.
The average distance between the CHs and nodes is lessened because of the high number
of neighboring nodes.
Fig. 2. Network lifespan analysis.
Fig. 3 is the throughput analysis. The proposed methodology transmitted data at a 0.978
Mbps maximum speed and a 0.815 Mbps minimum speed with 100 to 500 nodes. MOCH-EORA-WSN
and MOCH-MOGOH-WSN only reached 0.75 Mbps and 0.82 Mbps with 100 nodes. From the
simulated outcomes, the proposed technique increased throughput more than the existing
models in almost in all rounds. The increase in throughput means the proposed technique
has less overhead, shorter latency, and a more uniform distribution of energy among
the nodes. The proposed MOCH-SCNN-POA-WSN methodology attained 42.27% and 31.65% higher
throughput for round 0, 32.67% and 29.65% higher throughput for round 1200, 32.65%
and 69.85% higher throughput for round 2400, and 31.65% and 72.63% higher throughput
for round 3000, compared with MOCH-EORA-WSN and MOCH-MOGOH-WSN, respectively. Thus,
the attained outcomes prove the superiority of MOCH-SCNN-POA-WSN over other schemes.
Fig. 3. Throughput analysis.
Fig. 4 depicts energy consumption. The proposed methodology consumed less energy compared
to the other models. The proposed technique needed 0.6 mJ energy for 500 nodes, but
MOCH-EORA-WSN and MOCH-MOGOH-WSN consumed 0.93 and 0.95 mJ of energy with 500 nodes.
MOCH-SCNN-POA-WSN attained 69.76% and 48.63% lower energy consumption for round
0, 48.63% and 53.69% lower energy consumption for round 1200, 48.63% and 34.65% lower
energy consumption for round 2400, with 25.64% and 74.47% less energy for round 3000
compared to MOCH-EORA-WSN and MOCH-MOGOH-WSN, respectively.
Fig. 4. Energy consumption analysis.
PDR is depicted in Fig. 5. Here, the proposed routing method delivered packets effectively at 99.19% rate with
500 nodes, and 98.26% with 100 nodes. MOCH-EORA-WSN and MOCH-MOGOH-WSN attained delivery
ratios of only 93.66%, and 69.89% for 500 nodes. The proposed MOCH-SCNN-POA-WSN attained
53.26% and 46.56% better PDR for round 0; 65.63% and 31.63% better PDR for round 1200;
52.45% and 45.65% better PDR for round 2400; and 59.45% and 69.85% better PDR for
round 3000 compared to MOCH-EORA-WSN and MOCH-MOGOH-WSN, respectively.
Fig. 5. Packet delivery ratio analysis.
5. Conclusion
A multi-objective CH-basis energy-aware routing under an SCNN in WSN was implemented
successfully in this paper. The proposed methodology addresses the issue of short
lifespans in sensor nodes close to the base station. The low-lifetime issue develops
due to heavy traffic flow to the sink from several cluster heads. The proposed SCNN
method utilizes energy-efficient CH selection that depends upon multi-objectives,
such as residual energy, coverage area, communication cost, and proximity to the BS.
The POA is employed to optimum route selection via CH to sink node. The simulation
is activated in MATLAB; its efficacy was assessed with several performance metrics,
and the proposed MOCH-SCNN-POA-WSN methodology attained 45.56% and 55.90% maximum
throughput, with 66.12% and 78.88% maximum PDR compared with MOCH-EORA-WSN and MOCH-MOGOH-WSN,
respectively. Future research can develop new techniques to address a number of security
and privacy concerns that strain the integrity, accessibility, and data secrecy in
sensor nodes of WSNs. Future work will include blockchain-based security modes for
improving the protection and privacy of sensor nodes.
REFERENCES
J. John, P. Rodrigues, ``MOTCO: Multi-objective Taylor crow optimization algorithm
for cluster head selection in energy aware wireless sensor network,'' Mobile Networks
and Applications, vol. 24, no. 5, pp. 1509-1525, 2019.
M. Malakar, ``TLBO based cluster-head selection for multi-objective optimization in
wireless sensor networks,'' Nature inspired computing for wireless sensor networks,
pp. 303-319, 2020.
S.A. Sert, H. Bagci, A. Yazici, ``MOFCA: Multi-objective fuzzy clustering algorithm
for wireless sensor networks,'' Applied Soft Computing, vol. 30, pp. 151-165, 2015.
G. Hacioglu, V.F.A Kand, E. Sesli, ``Multi objective clustering for wireless sensor
networks,'' Expert Systems with Applications, vol. 59, pp. 86-100, 2016.
X. Hong, M. Tao, ``Multi-objective cluster head election in cluster-based wireless
sensor networks,'' International Journal of Computational Science and Engineering,
vol. 15, no. 1-2, pp. 87-95, 2017.
D. Mehta, S. Saxena, ``MCH-EOR: Multi-objective cluster head based energy-aware optimized
routing algorithm in wireless sensor networks,'' Sustainable Computing: Informatics
and Systems, vol. 28, p. 100406, 2020.
S.R. Nabavi, M. Najafi, ``Optimum Cluster Head Selection with a Combination of Multi-Objective
Grasshopper Optimization Algorithm and Harmony Search in Wireless Sensor Networks,''
International Journal of Smart Electrical Engineering, vol. 11, no. 02, pp. 71-82,
2022.
T.A. Alghamdi, ``Energy efficient protocol in wireless sensor network: optimized cluster
head selection model,'' Telecommunication Systems, vol. 74, pp. 331-345, 2020.
G.S. Arunachalam, G. Ramalingam, R. Nanjappan, ``A classy energy efficient spider
monkey optimization based clustering and data aggregation models for wireless sensor
network,'' Concurrency and Computation: Practice and Experience, p. e7492, 2023.
A. Sha, N. Verma, I. Poonguzhali, ``Multi‐objective‐derived energy efficient routing
in wireless sensor networks using hybrid African vultures‐cuckoo search optimization,''
International Journal of Communication Systems, p. e5438, 2023.
S. Chaurasia, K. Kumar, N. Kumar, ``MOCRAW: A Meta-heuristic Optimized Cluster head
selection based Routing Algorithm for WSNs,'' Ad Hoc Networks, p. 103079, 2023.
M. Biradar, B. Mathapathi, ``Security and Energy Aware Clustering-Based Routing in
Wireless Sensor Network: Hybrid Nature-Inspired Algorithm for Optimal Cluster Head
Selection,'' Journal of Interconnection Networks, vol. 23, no. 01, p. 2150039, 2023.
K. Debasis, L.D. Sharma, V. Bohat, R.S. Bhadoria, ``An energy-efficient clustering
algorithm for maximizing lifetime of wireless sensor networks using machine learning,''
Mobile Networks and Applications, pp. 1-15, 2023.
T. Zhang, X. Zhang, J. Shi, S. Wei, ``Depthwise separable convolution neural network
for high-speed SAR ship detection,'' Remote Sensing, vol. 11, no. 21, p. 2483, 2019.
N. Alamir, S. Kamel, T.F. Megahed, M. Hori, S.M. Abdelkader, ``Developing Hybrid Demand
Response Technique for Energy Management in Microgrid Based on Pelican Optimization
Algorithm,'' Electric Power Systems Research, vol. 214, p. 108905, 2023.
Author
Danish Ather is Currently working as an Associate Professor, at the Department
of Information Technology and Engineering, Amity University Tashkent, Uzbekistan.
He has 16.5+ years of experience in teaching, research & administration. He also has
experience in heading many academic and administrative positions such as IoT Research
Lab Coordinator at Sharda University, India, Chief Proctor, Programme Coordinator,
Examination In-Charge, and being one of the founding members to start the IEEE Conference
at Teerthanker Mahaveer Univerity, Moradabad and Technical Coordinator of 4th Technovation
Hackathon cum Innovation Budge at Sharda University India, in which 1500+ participants
participated from India & Abroad. Designed several training modules on Internet of
Things Applications using Arduino & Python.
J. Prisca Mary working as an assistant professor in the department of CSE at NPR
College Of Engineering & Technology, India
Pooja Singh working in the department of Physics, Faculty of Science at SGT University,
Gurugram, Haryana, India
Kanika Garg working in the department of Computer Science and Engineering at SRM
Institute of Science and Technology, NCR Campus, Modinagar, Ghaziabad, India
T. R. Priyadharshini has received B.E - Computer Science and Engi-neering in MPNMJ
Engineering College, Erode District in the year 2009. M.E degree in Computer Science
and Engineering in Gnanamani College of Technology, Rasipuram in the year 2012. I
am currently working as an Assistant Professor in the Department of Information Technology
in Hindusthan College Of Engineering and Technology, Coimbatore.
B. Anni Princy working as a professor in the department of Computer and Communication
at Panimalar engi-neering college, chennai, tamilnadu, India
Mohit Tiwari working as an assistant professor in the department of Computer Science
and Engineering at Bharati Vidyapeeth’s College of Engineering, A-4, Rohtak Road,
Paschim Vihar, Delhi, India