Content uploaded by Mohammed Zaki Hasan
Author content
All content in this area was uploaded by Mohammed Zaki Hasan on Jul 15, 2020
Content may be subject to copyright.
10344 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 6, DECEMBER 2019
Optimization of Sensor Deployment for Industrial
Internet of Things Using a Multiswarm Algorithm
Mohammed Zaki Hasan ,Member, IEEE, and Hussain Al-Rizzo
Abstract—A bio-inspired metaheuristics canonical particle
multiswarm optimization (CPMSO) algorithm is presented to
investigate the optimal deployment pattern for sensors to improve
the connectivity in Industrial Internet of Things (IIoT). The
proposed algorithm ensures effective deployment by building κ-
connected network to tolerate failure while satisfying quality
of service (QoS) in terms of energy consumption, delay, and
throughput. We prove the effectiveness of the CPMSO algo-
rithm by deploying multitopologies satisfying the QoS and results
are compared against the conventional canonical particle swarm
optimization (CPSO) and fully particle multiswarm optimization
(FPMSO) algorithms. It is shown that CPMSO and FPMSO
improve the throughput by approximately 95.23%, while min-
imizing the energy consumption by 87.5%, and the delay by
95.00% as compared with CPSO.
Index Terms—Canonical swarm optimization, connectivity,
fault tolerant, fully particle multiswarm optimization (FPMSO),
Industrial Internet of Things (IIoT), Internet of Things (IoT),
multiswarm.
I. INTRODUCTION
INTERNET of Things (IoT) is a network of connected
objects, each with low storage, limited energy, and process-
ing capabilities [1]. These objects interact in a complex way to
enhance reliability, performance, and security of their infras-
tructure [2]. The existing definition of Industrial IoT (IIoT) is a
framework for IoT components as a basis for analyzing the use
and deployment of IoT technologies and systems in industrial
settings [3]. IIoT is required to realize the desired performance
metrics of the IoT, such as energy conservation and cost reduc-
tion [4]. Hence, the question that arises is how to satisfy the
application requirements in a heterogeneous IoT network in an
industrial environment? We believe that recently emerged IoT
applications, such as smart cities, home automation, health-
care, agriculture, and renewable energy sources [3], [5], [6],
will further increase the scope of the existing definition of 5-D
experience [2]. This definition emphasizes the importance of
power distribution, topology control, bandwidth, availability,
and delay sensitivity since IIoT applications aim at creating
Manuscript received May 9, 2019; revised June 03, 2019, June 21,
2019, and July 27, 2019; accepted August 17, 2019. Date of publication
August 29, 2019; date of current version December 11, 2019. (Corresponding
author: Mohammed Zaki Hasan.)
M. Z. Hasan is with the College of Computer Sciences and Mathematics,
University of Mosul, Mosul 41002, Iraq, and also with the Systems
Engineering Department, Donaghey College of Engineering and Information
Technology, University of Arkansas at Little Rock, Little Rock, AR 72204
USA (e-mail: mzallayla@ualr.edu).
H. Al-Rizzo is with the Systems Engineering Department, Donaghey
College of Engineering and Information Technology, University of Arkansas
at Little Rock, Little Rock, AR 72204 USA (e-mail: hmalrizzo@ualr.edu).
Digital Object Identifier 10.1109/JIOT.2019.2938486
demand-side energy management, power system monitoring,
coordination of distributed power storage, and integration of
renewable energy generators. Therefore, most industries tend
to propose a new IIoT framework to help operators acceler-
ate network construction to ensure robust development and
deployment of IoT services and applications in large-area
locations [7].
The emergence of IoT’s services and applications requires
integration framework of IIoT within the context of IoT to
allow shifting IIoT planning and deployment to industrial
applications to maximize the IIoT value and benefits. The
algorithm used to achieve these objectives should take into
consideration the practical realization of the IoT’s services
paradigm that can be classified into three categories based
on applications: 1) individual; 2) public; and 3) industrial.
This allows us to establish a framework to analyze these
categories to understand the nature of IIoT. Each category
comprises a heterogeneous mix of devices with applications
in several different domains. Diverse services in each category
within different applications have divergent and challenging
requirements. Precisely, survivability of connection is one of
the diverse services that is considered as the most demand-
ing in urgent IoT applications such as IoT-based real-time
scenarios for remote monitoring systems for the industrial sec-
tor where successful operations presume the qualities of the
5-D experience definition based on energy efficiency, con-
nectivity, throughput, availability, and delay sensitivity [8].
This may be classified under the title of dependability, which
implies the ability of critical systems for sensitive applica-
tions must provide uninterrupted operations under survivability
services when limiting conditions are satisfied and to main-
tain the ability to resume normal services once the fault
has been tolerated. Therefore, survivability of connection
is considered important in these applications due to IIoT
needs to guarantee certain quality properties as the minimum
requirements to meet service providers’ and the end users’
requirements [9].
Connectivity can be treated as a deployment problem aimed
at the optimization of the QoS needed for guaranteeing the
existence of network links between any pair of devices [10].
For example, a smart meter requires a higher energy efficiency
but has a low requirement for delay. Meanwhile, a vehicu-
lar network is rather delay sensitive. Therefore, it is essential
to understand the mechanism of controlling the neighboring
set of sensors in different network technologies by adjusting
the transmission range and/or selecting appropriate sensors to
carry out a specific task.
2327-4662 c
2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
HASAN AND AL-RIZZO: OPTIMIZATION OF SENSOR DEPLOYMENT FOR IIoT USING MULTISWARM ALGORITHM 10345
TAB LE I
IIOTREQUIREMENTS IN INDUSTRY CONTROL
It should be noted that an optimal deployment strategy
ensures network connectivity along with efficient energy uti-
lization and network stability. In this article, we present a
robust bio-inspired metaheuristic optimization algorithm tak-
ing into consideration the 5-D experience definition based
on energy efficiency, connectivity management strategies,
throughput, availability, and delay sensitivity. This evaluation
can then be used to define a typical IIoT experience, such
as smart meters, connected vehicles, and industrial control. A
robust algorithm is proposed for information sharing mecha-
nism within particle swarm in order to optimize QoS param-
eters for various network topologies. Sharing information
depends on the mapping solution by updating the velocity
of optimal solution for each particle corresponding to differ-
ent types of devices in terms of sensors and Fog nodes. The
optimal solution is interpreted on what robust optimization
is: finding optimal solution given uncertain objective function,
finding optimal solution that remains stable under fluctuations
of the input variables, and finding optimal solution in dynamic
environment. We focus on defining a neighborhood relation-
ship among the particles toward finding robust optimal on
continuous optimization problem with uncertainty or fluctu-
ations in the input variable. Updating the sharing of a sensor’s
information means that the best sensor position of limited
neighborhood to which sensor is connected instead of the
whole swarm and therefore we can control the exploration,
exploitation behavior, as well as the convergence rate of the
algorithm. It is well known that particle swarm optimization
(PSO) is unable to balance exploration versus exploitation.
Exploration inhibits the swarm convergence [11], meanwhile
exploitation tends to cause the particles to hastily congregate
outside the feasible search space, leading to premature conver-
gence [12]. We employed a multiswarm strategy to slow down
the speed of convergence of swarm populations in terms of the
velocity parameter of each particle to balance population diver-
sity. Moreover, we implemented several base PSO algorithms
for their ability to provide particles with a collaborative learn-
ing strategy for optimizing QoS parameters and for testing
their applicability in the context of IIoT deployment [13]. The
major contributions of this article are summarized as follows.
1) A mathematical model is developed to satisfy QoS
requirements in terms of energy consumption, delay, and
throughput for the deployment of sensors in IIoT.
2) A robust multiswarm PSO algorithm is developed to
solve the optimization problem. The proposed algorithm
allows the deployed IIoT to achieve connectivity among
devices.
Extensive simulation experiments are conducted on differ-
ent κ-connectivity degrees. Then, the results are compared
against the traditional canonical PSO (CPSO) algorithm to
demonstrate that our proposed scheme is more suitable for the
deployment of sensors and participating devices in IIoT. The
remainder of this article is organized as follows. Section II
provides a review of literature related to work reported in
this article. In Section III, the PSO algorithm is introduced.
Section IV introduces the system model. Section V intro-
duces the proposed swarm optimization algorithm. Section VI
provides the analysis of the performance of the proposed
algorithm. Finally, Section VII concludes this article.
II. RELATED WORK
Connectivity is a crucial issue since IoT relies upon various
communication technologies, such as WiFi, Thread, ZigBee,
Bluetooth, and RFID, deployed in various network infrastruc-
tures [14]. These technologies alter the way IoT technology is
implemented as well as the operational mode of service and
device connectivity when integrated with the IoT paradigm in
the automation industry. Although there are several IIoT appli-
cations in different domains of academia and industry, they
all share the objective of improving the operational mode of
industrial processes by optimizing the interactive connectiv-
ity and collaboration of devices, which is a challenging task.
For example, a smart meter requires a score of five experi-
ence in terms of connectivity energy efficiency but imposes a
low requirement for delay. Meanwhile, vehicular network is
delay sensitive in real-time acquisition but has a low energy
consumption as depicted in Table I.
Therefore, controlling the network topology is vital since it
ensures efficient data exchange as well as the way communi-
cation protocols are used [15]. It is essential to understand the
mechanism of controlling the neighbor set of sensors in dif-
ferent communication technologies by adjusting transmission
range and/or selecting specific sensors to forwarding messages.
The topological structure of IoT infrastructure is self-managed
to cope with the increasing complexity and operational cost
of IIoT infrastructure management. Accordingly, different
organizations, such as Cisco Systems, Huawei, and Texas
Instruments, present the common requirements of IIoT by
providing self-configuring, self-optimizing, and self-protecting
functions for various deployment schemes of devices.
Meng et al. [16] highlighted the major technical gaps
of homogeneous communications among devices in IIoT
and presented a reference collaboration mechanism to
enhance self-configuration by focusing on connectivity.
Meng et al. [17] addressed the challenges of homogeneous
schemes in industrial systems and proposed a new data-
orientation of messages based on zero message quality (ZMQ)
protocol for rich sensing applications. Indeed, in a homo-
geneous scheme, the complex structure and heterogeneity
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
10346 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 6, DECEMBER 2019
problems of IIoT applications are distributed regularly where
each device or sensor is able to collaborate and transmits data.
Therefore, the characteristics of these applications are identi-
fied for device communications and then a data-orientation
messaging mechanism is presented. Although this mecha-
nism increases the reliability, it may not be suitable for IIoT,
especially for large-scale deployment since sensors nearby
the gateway more often consume energy and deplete ear-
lier, or may face temporal death and disconnect from the
network, which makes the lifetime of the whole network short.
Fortunately, energy harvesting (EH) makes it possible for sen-
sor networks to exploit energy from several sources, such as
temperature, magnetic fields, solar power, wind, etc., without
requiring a conventional recharging, as well as they can be
deployed in different places with several infrastructures. Tang
and Tan [18], [19] considered a temporal death model for EH
devices using a 3-D stochastic process model to describe the
dynamic policy of EH-WSN transmission of the data pack-
ets taking into consideration the data buffer queue length and
package blocking probability.
Khalil et al. [20] proposed a specific mechanism for the
lower layer (i.e., sensing layer) of operation mode for service
level in order to satisfy the requirement of IoT applications.
Designing virtualized wireless sensor network (VWSN) for
IoT environment was proposed in [21] by providing a group
of users within a reliable VWSN network by assigning redun-
dant resources according to each user’s demand and providing
a recovery method to incorporate environmental changes.
Handte et al. [22] presented an urban bus navigator (UBN)
of IoT to improve the ability of seamless availability (cover-
age area) and connectivity between bus passengers and public
bus infrastructure. The IoT architecture is based on three lay-
ers to support direct connectivity between the passengers and
bus vehicles. A new paradigm was presented in [23] for an
IoT application in the new generation of environmental infor-
matics. The authors focused on key integration of technologies
from both integrated information system (IIS) as well as IoT
for regional environmental monitoring based on four layers:
1) perception; 2) network; 3) middleware; and 4) application.
The authors supported their investigation with a case study
to show where IIS and IoT solutions can provide competi-
tive advantages and play an important role in environmental
monitoring and management. The purpose of constructing het-
erogeneous sensor deployment in multitiered IIoT scheme is to
ensure that the connectivity of sensors in the monitored area
is maintained over multiple hops and starts from the lower
layer, while the Fog and the gateway are placed in the upper
layer. Although the sensors have the functionality to commu-
nicate directly with their neighbors and forward the data, the
sensors must be equipped with more complex chips which
are very costly. Therefore, it is important to investigate the
problem of availability and connectivity by optimization of
sensors topology in IIoT systems.
Topology control is considered a common way to improve
and preserve the connectivity [13], therefore, there exist sev-
eral topology control methods in [24] and [25] that can be
classified based on two factors. The first factor is the met-
rics to be improved, such as availability, number of devices,
reliability, transmit power control, managing duty cycles, and
Fig. 1. Topology control (centralized): adjust the transmission range of
specific nodes to improve the connectivity.
Fig. 2. Topology control (distributed): the new nodes are added to recover
the connection.
clustering for data aggregation [23], [26]–[28]. Additionally,
security is a challenge due to several issues: how to balance
security on one side with ability to utilize the potentials of
IIoT on the other side [29]–[32] which is outside the scope
of this article. The second factor is the integration of sensor
network into the IoT infrastructure [33]–[35]. Generally, there
are three main deployment approaches for WSN correspond-
ing to these metrics which must be considered to improve the
connectivity.
1) Predeployment, where the sensors are randomly posi-
tioned and the network is not operational. However,
constructing network topology is needed at this stage
to satisfy connectivity as shown in Fig. 1.
2) Post-deployment, where the positions of the sensors are
already known, however, adjusting the transmitted power
is required to achieve the desired connectivity as shown
in Fig. 2.
3) Redeployment, where new sensors are populated inside
the original network to maintain the connectivity.
Connectivity improvement should be applied only at one
stage in each strategy, therefore, in predeployment, there are
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
HASAN AND AL-RIZZO: OPTIMIZATION OF SENSOR DEPLOYMENT FOR IIoT USING MULTISWARM ALGORITHM 10347
two cases: 1) randomly positioned sensors’ locations can
be controlled using centralized or distributed algorithms and
2) construct κ-connected networks [10]. For both cases, the
number of sensors must be known so that the whole network is
κ-connected. Meanwhile, the topology control methods have
been recently presented for energy efficiency by relying on
the ability of sensors to control their transmitted power [15].
However, the control of transmitted power may be insufficient
for the network lifetime. Therefore, connectivity should be
taken into consideration. Note that adjusting transmitted power
and sensor placement are NP-hard even if the network is con-
nected with κ=1. Moreover, applying these schemes to IoT
network may cause a great amount of overhead to construct
and select paths as well as topology maintenance.
Several metaheuristic algorithms have been proposed to solve
the coverage and connectivity problem using on-demand tech-
niques employing specific heuristic algorithms for searching
the optimal solution [36]. In [37] and [38], a diversity concept
is proposed to avoid premature convergence of swarm by set-
ting a lower and upper bound of feasible region to ensure that
the swarm has a good search ability in finding optimal solu-
tions for various real-world applications. Other studies use the
strategy of duty management in order to schedule the smallest
group of nodes to be in active mode, while putting others in a
sleeping mode [39], whereas others use geometry [40], local-
ization [41], direct information techniques, and combination of
these techniques [42] to address the κ-connectivity in WSN.
In this article, however, we present a new and efficient
PSO algorithm which is nature inspired with population
based on metaheuristic. We used several techniques of robust
optimization in terms of canonical, fully informed, and mul-
tiswarm to take advantage of the combination of proactive
and reactive routing mechanisms to exchange demanding
calculation to maintain multinetwork topology, and record
the objective function value for the selected connecting
paths. Moreover, we define the whole swarm’s neighbor-
hood relationship among sensors according to the mecha-
nism of information exchange. Therefore, different topologies
have been defined and used for interparticle communication.
Updating a sensor’s information means that the best sensor
position of limited neighborhood to which the sensor is con-
nected instead of the whole swarm and thus can control explo-
ration and exploitation behavior and convergence rate of the
algorithm. By using these techniques of robust optimization,
we can show that multiswarm topologies can remarkably influ-
ence the performance of the algorithm. The motivation behind
using this mechanism is to combine the information of network
topology for exploitation and to converge toward optimal con-
figuration, or route whereas the sensors collaborate to keep
diversity up, which differs from the previous works [36]–[38].
Finally, the research reported in this article considers the mul-
titiered heterogeneous scheme for IIoT deployment, whereby
the deployment scheme involves Fog nodes with relatively
strong computing and storage capabilities.
III. PARTICLE SWARM OPTIMIZATION ALGORITHM
The earliest PSO attempted to use the concept of social
behavior of swarm resulting in a set of particles that spread
over the search space. PSO has undergone many changes rang-
ing from minor parameters adjustments to complete restructur-
ing of the algorithm. The original PSO starts from the initial
population of a swarm Scandidate solutions called particles
which explore with random positions Xid and velocities υid in
adth n-dimensional hyperspace ı≤dth ≤nof problem or
objective functions (where nrepresents the number of optimal
parameters to be determined). A particle ıoccupies its posi-
tion Xid and velocity υid. Each particle is evaluated through an
objective function f(x1,...,xn), where f:n→denotes
the number of particles (i.e., sensors) that are uncovered by
particle (sensor), therefore PSO tries to achieve close to ideal
coverage determined by the network connectivity. Meanwhile,
the velocity υid constitutes the momentum component, best
position constitutes the cognitive component, and the best
previous position of particle’s neighbor constitutes the social
component. The evaluation of fitness of the particle implies
that particle close to the global solution is lower than a par-
ticle that is not connected. PSO guides each particle to move
through the search space in a direction by combining some
aspects of global solution and best-fitness locations with one
of the whole members of the swarm. The global-best version
of PSO defines the position of a particle which has its lowest
cost stored as best previous position. Besides, the global-best
solution defines the position of the best particle. We hereby
note that both the particle and sensor node are interchange-
ably used to refer to a “sensor.” Thus, at each iteration, the
algorithm determines for each particle which other particle
was its nearest neighbor, then assigns the particle velocity
to another particle which communicates among others. The
update process is repeated until either an acceptable global-
best solution is achieved or a fixed number of iterations is
reached. Therefore, the updating velocity rules for calculat-
ing the next position of a particle in each iteration follows a
cooperative and social pattern as [43]
υκ+1
ı,j =υκ
ı,j +c1r1Xbestκ
ı,j −xκ
ı,j +c2r2Xgbestκ
j−xκ
ı,j .
(1a)
And
xκ+1
ı,j =xκ
ı,j +υκ+1
ı,j (2a)
ı,j =1,2,3,...,N+M(2b)
where ı,j denote the sensor’s index; xı,j,dκ
ı,j , and υκ
ı,j are the
jth component of the ıth sensor’s position, distance, and veloc-
ity vector, respectively, in the κth iteration; r1and r2are the
two random numbers generated from a uniformly distributed
function in the range [0,1]; Xbestκ
ı,j and Xgbestκ
jindicate the
best and global positions experienced so far by the ıth sen-
sor and whole swarm topology, respectively; and c1and c2
represent the confidence particle as in cognition and social
behavior, respectively.
The communication topology determined by the best and
global positions overall the swarm particles. Whereas the
Xbestκ
ı,j determine the simplest form of local communication
topology, what is known as ring communication topology,
where each particle connects to only two other particles in
swarm as depicted in Fig. 5. Meanwhile, Xgbestκ
jdetermines
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
10348 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 6, DECEMBER 2019
Fig. 3. Schematic movement of a particle based on (3).
Fig. 4. Schematic movement of a particle based on (4).
the simplest form of global communication topology, what is
known as mesh communication topology, where every parti-
cle is able to obtain information from the very best particle
in the entire swarm population as depicted in Fig. 6. These
two parameters are set equal to 2 in the earlier versions of
PSO to control the overfly of a particle through target about
half time and then to control the balance between exploration
and exploration tendencies. Canonical-based PSO algorithms
aim at better control of exploration/exploitation searches by
defining a neighborhood relationship among the particles by
linearly increasing the connectivity of swarm’s topology and
performing an adjustable mechanism. These topologies as seen
in Figs. 5 and 6 show how data are hold on whether a parti-
cle neighbor is connected with another particles. Hence, ring
and mesh swarm topology has the advantage of having higher
diversity and therefore being able to find the global optimal
solution. The motivation of using these swarm topologies
is to combine the benefits of local and global swarm com-
munication topology to converge toward an interesting area.
Most heuristic deterministic methods would always likely get
optimal results but they are easy to fall into local optimum.
There are several studies that have improved the PSO by
maximizing velocity based on various neighborhood defini-
tion (such as velocity selection, acceleration constant, inertia
constant weight, or constriction factor section) in order to
prevent a particle from leaving the search space; others dis-
cussed the role of the three terms of (1) by introducing a
new factor, named inertia weight ωused to maintain bal-
ance between the exploration and exploitation searches of
sensors [43]. Therefore, (1) is rewritten as
υκ+1
ı,j =ωυκ
ı,j +c1r1Xbestκ
ı,j −xκ
ı,j+c2r2Xgbestκ
j−xκ
ı,j.
(3)
Fig. 5. Local swarm communication topology.
Fig. 6. Global swarm communication topology.
We consider how a swarm behavior in information commu-
nication could be implemented into IIoT through ambient
intelligence, smart devices, or virtual realities and how other
new ways of combining computational advances for swarm
behavior with smart possibilities. Consequently, we can benefit
from adding in new smart devices to form a complex organ-
ism on the IIoT. However, questions arise, could this benefit
be shared in multimodal information mechanisms or just in
control of selected devices? Which pattern of swarm behavior
becomes possible on large-scale network if distributed over a
huge area? Will it trigger a new step of evolution in an indus-
trial area? Typically, there are several PSO algorithms which
undergo some modifications on updating velocity rules for
finding a robust optimal solution. We investigated these modi-
fications to find the difference between the types of canonical
PSO algorithms through the substantial coupling between the
swarm behavior and the mechanism of deployment of devices.
Therefore, a network process can contribute in providing a way
of controlling and managing the connectivity of devices at its
several iteration.
A. Canonical PSO
Existing literatures [27]–[32] provide interpretations on
which metaheuristic is capable of finding an optimal solution.
In this article, we focus on finding a robust optimal solution
for continuous optimization problems with uncertainty in QoS
parameters for large networks as shown in Figs. 1 and 2 in
dynamic IIoT environments. However, the difficulty comes
when the size of the network becomes substantial. Thus, it
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
HASAN AND AL-RIZZO: OPTIMIZATION OF SENSOR DEPLOYMENT FOR IIoT USING MULTISWARM ALGORITHM 10349
is better to call upon a version of PSO taking into account
the size and combinational aspects of the connection problem
through the definition of the management of true distance in
the search space. These rely on two significant aspects of the
algorithm: 1) the definition of a proximity distribution and 2)
the search for the best information of a particle. To perform the
robust optimal solution, exchange of best information among
the particles is satisfying. Moreover, we define several network
topologies to insure connectivity using the interparticle com-
munication. Therefore, the robust solution depends on swarm
topologies, which influences the performance of the algorithm
through either using all information from other sensors in its
neighborhood or just the best one. Some studies have focused
on the parameters in terms of the number of particles, total
number of iterations, inertia weight, and social and/or cogni-
tion behavior of the particle to guarantee convergence of the
particle swarm algorithm in finding the optimal solution. By
referring to (1), mathematical operations (subtraction −, exter-
nal multiplication ⊗, addition ⊕, and displacement )were
added to address the two significant aspects of the algorithm
according to the size of the problem. Therefore, an alternative
formulation for velocity as referred in (3) is defined as [43]
υκ+1
ı,j =χυκ
ı,j +c1r1Xbestκ
ı,j −xκ
ı,j+c2r2Xgbestκ
j−xκ
ı,j.
(4)
Fig. 3 illustrates the schematic movement of a particle (i.e.,
satisfying exchanging information between particles), where a
weight factor ωhas been defined to achieve and to balance
the exploration/exploitation searches of particles
χ=2
2−ϕ−ϕ2−4ϕ
(5)
where ϕ=c1+c2,ϕ>4. Fig. 4 illustrates the schematic
movement to satisfy information exchange between sensors,
where different topologies have been defined to ensure the con-
nectivity by using the interparticle communication. Updating
velocity implies the best sensor position of limited neigh-
borhood in order to determine the connectivity with other
sensor’s neighborhood instead of the whole swarm topology.
Therefore, swarm topologies in PSO can remarkably influence
the performance of the algorithm. Furthermore, CPSO uses the
fully connected topology in which all sensors are neighbors.
This leads to connect a sensor directly to a global best sensor
and affects it simultaneously. Therefore, the swarm topology
in CPSO does not explore other areas of the search space and
mostly trapped in local optimal solution. Meanwhile, a sensor
in the fully informed PSO (FIPSO) uses information from all
other sensors in its neighborhood instead of just the best one.
The velocity update function in FIPS is defined as [43]
−→
υ:=χ.−→
υı+1
κı
κı
m=1
−−−→
Xbest(0,φ)+−→
pnbrm−−→
xı(6)
where κıis the number of neighbors for sensor ıand −→
pnbrm
is the mth neighbor of sensor ı. This modification enhances
the performance of FIPSO in two ways. First, it helps the sen-
sor to receive information about good regions of search space.
Fig. 7. Schematic movement of a particle of (3, 3) m, n-connectivity based
on (6).
Second, forbidden error implies sensors from taking part in
the movement of the swarm to improve the exploration capa-
bilities of the algorithm. Figs. 4 and 7 illustrate the schematic
movement of particle based on (6).
B. Multiswarm PSO
Another PSO modification is multiswarm which aims at
increasing the diversity of the search space such that the algo-
rithm does not converge to local optimal solution. Multiswarm
has the ability to improve the convergence rate of the
search space by generating and grouping new subswarms.
Additionally, these subswarms are regenerated again creat-
ing a new group of subswarms to assist in escaping from
local optimal and finding the global solution. Generally, mul-
tiswarm uses a main swarm Mwith nparticles (xı,υ
ı,yı,rcı),
ı∈{1,...,n}, where xıdefines the position of a particle, υı
is the velocity of the particle, yıdefines the objective func-
tion of the problem, and rcdefines the transmission of a
particle. We briefly highlight the mechanism of multiswarm
as when a new subswarm is generated from the main swarm
by observing the objective function according to the value
of transmission of a particle. Thus, if a particle is connected
with the closest particle, then the new particle merges into a
new subswarm and is removed from the main swarm. Later,
each subswarm independently updates its velocity, hence, a
lot of different multiswarm variants are possible depending on
which full PSO mode is used by the subswarms. The proposed
algorithm utilizes this self-adaptive mechanism of multiswarm
to explore constrained search spaces. Multiswarm diversity
mechanism combined with a comparison mechanism of CPSO
which uses feasibility based on rules used to guide the search
toward the global optimal solution. In addition to combina-
tion, multiswarm improves the exploration and exploitation in
search space by avoiding a high selection pressure of candi-
dates as well as maintaining infeasible solutions. This is an
idea that allows us to explore the individual solution with
the lowest amount of constraint violation for the next popula-
tion. Thus, in this article, we present an update to the best of
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
10350 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 6, DECEMBER 2019
Fig. 8. Schematic movement of a particle of multiswarm.
neighborhood velocity function of the CPSO and FIPS using
multiswarm in fully connected network topology in order to
find the optimal solution. This is because simultaneous attrac-
tion to multiple sensors provokes random behavior of the
swarm. Therefore, this behavior will offer new opportunities
to explain the performance of multiswarm with various con-
nected network topologies. Fig. 8 illustrates the processes of
creating multiswarm and merging into a new swarm.
IV. SYSTEM MODEL
Typically, an IIoT consists of a network of numerous wire-
less devices deployed in a wide area, leading to a large-scale
complex system. We assume a distributed scenario where
sensors are deployed in IIoT using fault-tolerant topology
control in a heterogeneous multitiered layers (HML-IIoT)
framework [44]. This framework consists of Cloud back-end,
a Fog middle layer, and a sensor layer, as depicted in Fig. 9.
The Fog middle layer consists of a number of resource-rich
Fog nodes, denoted by S. The sensor devices are constrained
by limited battery capacity and unmitigated QoS constraints,
denoted as R. Each sensor can vary its transmission range
by varying power level within network topology depending
on deterministic or nondeterministic workload to transmit or
receive a message. A range assignment is defined as power set-
ting for each sensor, and cost of a range assignment is either
the average power setting or the maximum power setting in
that assignment. We seek a solution that minimizes power con-
sumption while maintaining key network properties such as
connectivity. In fact, the cost of transmitting a message among
sensors is independent of the number of receiving sensors,
however, it depends on a function of their maximum distance
for sending a message from sensors. Therefore, if the network
is multihop, then it is possible to maintain connectivity without
every sensor transmitting at maximum power. This is one of
the common fault-tolerance solutions to establish and maintain
connectivity among disjoint sensors with proactive and reac-
tive routing mechanisms. This allows us to seek optimal power
range assignment for connectivity and other related network
issues. In other words, it provides an idea to find solutions to
Fig. 9. Structure of IIoT.
such complex problems applying several types of CPSO in a
dynamically IIoT environment.
Existing work on connectivity for large-scale IIoT has
been reviewed in Section II of this article. Most works have
addressed the issue of optimal power range assignments that
maintain connectivity as NP-hard in the Euclidean plane
or in κ-fault-tolerant network. Some approaches concentrate
on local heuristics without considering other sensor network
issues such as lifetime. Other approaches focused on provid-
ing approximation algorithms depending on several heuristic
techniques, such as Steiner trees or κ-fault tolerant to main-
tain the connectivity. Both approaches pursue a provisioned
methodology by deploying redundant sensors, or reposition-
ing to restore lost connectivity. Although these approaches
are heuristic, however, they do not have provable bounds on
solution cost (i.e., objective function).
Most of the proposed approaches perform arbitrarily worse
than the optimal solution. We concluded that two challenges
must be addressed. One is how to ensure the reliability of the
network while maintaining network coverage (or availability).
The second is how to adapt the network topology control uti-
lizing a swarm algorithm in several IIoT scenarios. We notice
that network coverage is highly determined by the sensors’
distribution and it is crucial in several IIoT systems such as
monitoring. Moreover, several existing works assume that the
sensing range of a given sensor is the same as its transmis-
sion range. This implies that once the network is connected,
the network coverage is guaranteed. However, this assumption
is not true in IIoT scenarios since some sensors can commu-
nicate through a short distance while possessing long routing
abilities. Typically, it is feasible to deploy dynamic routing
protocols for small-scale IIoT, however, a large amount of
energy is spent. Therefore, for large-scale IIoT, it is reasonable
to adopt static routing for higher energy efficiency, optimal
network reliability, and scalability. This article investigates
the κ-fault tolerant range assignments with the objective of
minimizing power consumption and delay.
We present three algorithms for solving this problem
depending on different considerations to demonstrate the
framework in [44] by centralized and distributed routing pro-
tocol using robust swarm optimization for different network
topologies based on range metric.
First, we present a simple algorithm to obtain approxi-
mation for κ-connectivity for 2 ≤κ≤4, then we extend
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
HASAN AND AL-RIZZO: OPTIMIZATION OF SENSOR DEPLOYMENT FOR IIoT USING MULTISWARM ALGORITHM 10351
the investigation using multiswarm techniques to obtain better
approximation for optimal-power assignment while maintain-
ing κ-degree connectivity. Within the changing IoT environ-
ment, demanding traffic patterns, variation to traffic patterns,
and addition or removal of sensors can occur. To realize the
dynamic IoT environment, a mixed deployment of HML-IIoT
objects can provide a balance between the performance of
dynamic and static routing mechanisms on the one hand and
the cost of IIoT on the other hand [10]. However, IoT elements
are based on the 5-D experience requirements and the service
model.
Our investigation of variable power assignment as well as
the relationship between the topology cost and path connection
follows the approach reported in [44]. Therefore, by placing
IoT elements in a swarm topology framework, the proposed
deployment scheme provides more flexibility, more scalabil-
ity, and increased manageability of devices. Hence, a major
benefit of an HML-IIoT is that optimal deployment using a
connectivity factor does not require sophisticated computation
to run routing mechanisms and therefore significantly reduces
the IIoT topology cost. It should be emphasized that to con-
figure more dynamic routing, we consider the topology as
follows. Let υıand υjbe two sensors joined in Euclidean
distance denoted as dκ
ı,j . The construction of κ-disjoint paths
to achieve more than three connectivity degrees and full cov-
erage for sensor deployment is a major challenge due to low
battery capacity in performing path computations. Moreover,
we consider a default wireless channel where Fog nodes and
sensors are able to ensure the required connectivity degree by
a many-to-one traffic pattern. Furthermore, a need exists to
exchange route information regularly which causes overhead
traffic as well as more energy consumption for extra connec-
tivity. Before introducing the model, we briefly introduce some
necessary definitions as follows.
Definition 1: Let n=p1,p2,...,pN+Mbe a set of paths
on graph Gthat are said to be energy-node-disjoint, meaning
they have no common nodes.
Every node-disjointness is a set of all nodes which are
closer to p1(υı,υ
j)used in building the network topol-
ogy with the κ-disjoint paths routing. In order to increase
the number of alternative paths, we assume every sensor-
disjointness is a set of all sensors used in building the
network topology with κ-disjoint paths routing, thus mak-
ing the network more reliable. Our model is based on the
observation in [13] that to achieve an optimal deployment
pattern with different degrees of connectivity between neigh-
bors not on a κ-disjoint path from a sensor to a Fog node,
we must determine which neighbors are on a path and which
are not.
Definition 2: n paths are said to be node-disjoint, if and
only if they have no common nodes.
Definition 3: A WSN is said to be four-connected if for
every two interior nodes of |N|there are at least four node-
disjoint paths joining them.
According to Definitions 1–3, the node-disjointness rela-
tions are modeled as a directed graph G(V,E)ina2-D
space, where the cardinality of sensors, Fog nodes, and
links in graph Gare denoted by |V|and |E|, respectively.
|V|={υ1,υ
2,...,υ
N,υ
N+1,...,υ
N+M}are finite vertices
representing the number of sensors. N⊂Rdenotes the sensors
and M⊂Sdenotes the Fog nodes. |E|represents the links,
with each link assigned a non-negative value representing the
cost of the link in terms of energy consumption, hop-distance,
delay, and throughput. All sensors are assumed to be identical,
and their reliable power transmission range denoted as rcis
greater than or equal to the sensing range which is denoted
as rs. Each sensor discovers the neighbors in its transmission
range by periodically sending the message hello and collecting
the energy consumption, hop distance, delay, and through-
put of its neighbors. Therefore, the relation between a pair
of nodes is the number of links |E|in Gwhich is a sub-
set of paths. |E|={(υı,υ
j)|Hop(υı,υ
j)≤rc},Hop(υı,υ
j)
depicts the distance between vıand υj. A path P(υı,υ
j)
from vertex υıto vertex υjin a graph Gis a sequence of
links that are traversed when going from υıto υj, where
P(υı,υ
j)={pı(υı,υ
j)∈P:∀pı(υı,υ
j)∈P,Hop(υı,υ
j)≤
rc,ı= j=1,2,...,N+M}. Therefore, the paths are defined
as a set of alternative paths pı(υı,υ
j).E(υı,υ
j∈pı(υı,υ
j))
represents a node-disjoint between pı(υı,υ
j), (υN,υ
N+M),
and e(e∈pı(υı,υ
j), (υN,υ
N+M)) represents the direct link
between two nodes. Based on the above, the communica-
tion policy of direct links among nodes can be outlined
as follows.
1) For any υıand υj∈N⊂R,υıand υjcannot
communicate with each other even if dκ
ı,j <rc.
2) For any υı∈M⊂Sand υj∈N⊂R,υıand υjcan
communicate with each other even if dκ
ı,j ≥rc.
Thus, the κ-disjoint paths in |G|can be obtained by consid-
ering the QoS parameters affecting the selection mechanism
of the optimal paths, including energy consumption, delay,
and throughput. These parameters are used to evaluate the
objective function of the selected paths, while the derivations
in [45] are used to solve the objective functions for mini-
mizing energy consumption and average delay. The κ-disjoint
paths are constructed to transmit data collected by the sensors
to the Fog nodes for HML-IIoT. By combining these com-
ponents of HML-IIoT into IIoT framework, sensors become
multifunctional depending on their communication and elec-
tronic limitations as well as their application specific require-
ments. However, how can we use such resource-constrained
sensors to meet certain application requirements of devices
in IIoT?
Networked sensors face challenges imposed by integra-
tion within other co-existing heterogeneous wireless systems.
This coexistence and integration may substantially affect the
performance of sensors which rely on diverse performance
metrics to optimize the QoS. These metrics often conflict with
each other, hence carefully balancing the tradeoffs among sen-
sors is vital in terms of optimizing the overall performance of
the IIoT. Network connectivity is closely related to energy
efficiency. Thus, we need to define a relationship among the
number of sensors that remain active and connected with
acceptable QoS. Hence, we focus on the κ-vertex Fog node
connectivity to obtain fault-tolerant topology control as a
transmission range assignment problem where each sensor is
connected to at least one Fog node by κ-disjoint paths in the
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
10352 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 6, DECEMBER 2019
TAB L E I I
NOTATI ON
network. The connectivity can be tolerated by neighbors that
are not on one of the κ-disjoint paths from a selected sensor
to one of the Fog nodes. To achieve this, we need to deter-
mine which neighbors are on one of these paths and which
are not. In this case, the objective function aims at saving the
energy achieved by minimizing transmitted power as well as
end-to-end delay. The optimal paths have been obtained by
applying the objective functions to all of the deployed sensors
while ensuring the κ-disjoint paths between these sensors and
the Fog nodes. This topology requires each sensor in the IIoT
environment to be κ-disjoint connected to at least one Fog
node. Table II lists the notations used throughout this article.
A. Modeling QoS for IIoT
There are many sensors and Fog nodes to be distributed in
a certain geographic area, however, the goal of deploying κ-
disjoint sensors in IIoT is to determine the optimal number of
hops between the sensors and Fog nodes while satisfying the
QoS parameters. The model is presented as an optimization
problem for QoS in IIoT in terms of energy consumption,
delay, and throughput. We propose a multiswarm algorithm to
solve the problem. To obtain the optimal deployment pattern
with a minimum number of sensors and Fog nodes certainly
requires defining the deployment pattern as a constraint. To
define this constraint, a neighborhood relationship among the
sensors and Fog nodes is used. First, we define ℵı,j as the set
of disjoint sensors and the next neighborhood of the κ-disjoint
path as
ℵı,j =ı,j = ı|
υı−υj
≤rcυ(ı,j).(7)
Let πbe the conditional adjacency matrix of G(V,E)to ensure
that a link exists between two nodes, then
π=⎛
⎜
⎜
⎜
⎜
⎜
⎝
c11 c12 ... c1|N|
c21 c22 ... c2|N|
c31 c32 ... c3|N|
.
.
..
.
.....
.
.
c|N|1c|N|2... c|N||N|
⎞
⎟
⎟
⎟
⎟
⎟
⎠
(8)
where
c(ı,j )=1,if υ(ı,j)∈ℵ
ı,j
0,otherwise.(9)
Second, to address the topology requirements, we consider the
following constraints which are determined by three variables:
1) the connectivity factor; 2) the intermediate distance between
two sensors along the selected path; and 3) the number of
hops. Denoting rcυ(ı,j) as the power transmitted by one sensor
to the next hop sensor, then, (9) defines a binary connectivity
constraint to determine whether a sensor is within its power
transmission range and if dκ
ı,j ≤rcυ(ı,j) . We can rewrite (9)
to satisfy a new connection to be added in directed graph
G(V,E)as
c(ıj)=1,if ı,j = ı|υı−υj≤rcυ(ı,j)∈ℵı,j ⊆S∪R
0,otherwise.
(10)
1) Energy Consumption Constraint: The IIoT energy con-
sumption model depends on expenditure in transmit and
acquisition, while processing and sensing expenditures are
less than data transmission/receiption. Selecting the subse-
quent hop is achieved via exploiting the nearest neighbor, and
therefore each sensor has a transmission range to communi-
cate with neighbors. The model for energy consumption per
bit is
Energyυsd =
S∪R
ı,j ∈N∪M
Etrans
elecυN∪M+0rα
cυ(ı,j)+Erec
elecυN∪MLp(11)
where Etrans
elecυN∪Mis the distance-independent energy consumed
by the transmitter, Erec
elecυN∪Mis the energy utilized by the
receiver while αis energy loss due to path loss assuming that
the channel is free of obstacles, and rcυ(ı,j) is the transmission
range. The data rate from sensor υıto υjis equal to data rate
from υjto υıin a time unit and denoted as Fıj. Therefore,
the end-to-end energy consumption per unit time is
Energyυı=
j∈R
cıjFıjEtrans
elecυN∪M⊂S∪R+0rα
cυ(ı,j).(12)
Energy consumption in receiving per unit time is given by
Energyυj=
ı∈S∪R
cıjFıjErec
elecυN∪M⊂S∪R.(13)
Therefore, the overall energy consumed Energyυsd is
Energyυsd =
ı,j ∈S∪R
cıjFıj2EelecυN∪M+mprcıjα.(14)
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
HASAN AND AL-RIZZO: OPTIMIZATION OF SENSOR DEPLOYMENT FOR IIoT USING MULTISWARM ALGORITHM 10353
Note that we consider the transmission and reception of the
sensed data among the set of disjoint sensors of the κ-disjoint
paths which can vary during transmission and reception. It
might lead to disconnecting several neighbors and partitioned
paths, unless the constraints are satisfied. Hence, the second
constraint is the intermediate distance between two sensors
along the selected path, written as
Hop =α
dυ(ı,j )3εmp
2Eelecυ(ı,j)≤rcıj.(15)
This constraint guarantees that the optimal number of hops
among the selected κ-disjoint paths can be obtained, which
is an important design and performance parameter in IIoT.
There is a linear relationship among the intermediate distance
between two sensors and the hop count. This relationship does
not work unless the sensor’s connectivity degree is large, and
this variation is most prominent for a connectivity degree of
three. Thus, the theoretical hop count is obtained as an integer
number for the selected κ-disjoint paths given by
hops count =total distance
ηoptimal (16)
where
ηoptimal =α
dυ(ı,j )30
2Eelecυ(ı,j).(17)
Therefore, the optimization problem is solved by finding the
robust optimal solution of objective functions in terms of
energy consumption model constrained by (16) and (17).
Moreover, these constraints adaptively control network topol-
ogy based on the distance-based connectivity of the selected
κ-disjoint paths. Due to uncertainty of the connectivity degree
of the deployed topology that may change due to the dynamic
environments, these constraints are used for each objective
function to decide the optimal pattern of sensors deployment
when monitoring a given region of interest. Consequently, the
constraints in (16) and (17) define the dynamic objective func-
tion in terms of the lower and upper boundaries of the feasible
solution space. As optimization progresses, energy consump-
tion is minimized by changing the connectivity degree of the
deployed topology.
2) Delay Constraint: The delay constraint is accounted for
by finding the optimal number of hops in a path which might
lead to different delay guarantees in the network. Let the
delivery delay between two sensors be denoted as ℘(ξ
ı,ξ
j).
This delay might be classified into several types as queu-
ing, propagation, processing, transmission, retransmission, and
idle. Then, the mean delay can be computed as
ξ=Dqueu +Dprop +Dproc +Dtrans +Dretrans +Didle.(18)
The strategy for finding the optimal number of forwarding
hops as in (17) aims at minimizing the delay of successful
transmissions. This means that a sensor may receive data via
multiple hops, but it aggregates and sends data only once.
Then, we jointly optimize the hops and the delay. For each
sensor, the delay from its one-hop neighborhood is period-
ically computed. This is applied to all source as well as
the intermediate sensors. The end-to-end QoS dictated by the
devices are met when the accumulated QoS requirements are
satisfied at each hop [45]. At each hop, the proposed algorithm
evenly divides the bounded delay as defined below:
℘ξı,ξ
j=℘sdLp∪ξı,ξ
j
∀ξ=ℵ
ıj=ξı,ξ
j∈N∪M⊂S∪R.(19)
Therefore, the end-to-end delay, ξSource and ξDestination, over
the set of paths Pis given by
℘sdLp=⎧
⎨
⎩
ξı,ξj∈N∪M⊂S∪R
℘ξı,ξ
j⎫
⎬
⎭
.(20)
The delay denoted by ℘sd occurs due to data transmission
along the set of paths between ξsand ξd. Since the delay
℘(ξ
ı,ξ
j)is defined as the time required to successfully
transmit data after the first sensor receives it, thus
N∪M
(ξı=1,ξj=1)
℘ξı,ξ
j≤℘ (21)
where ℘ is the bounded delay, which depends on two factors:
1) the number of hops taken and 2) the delay of a sensor, which
are additive and denoted as ηıjand ℘e, respectively. Therefore
℘ =℘Source
0+℘ξ+1
η1+℘ξ+2
η2+···+℘Destination
ηN+M.(22)
L℘
eis the per-hop delay from source to sink with the number of
hops ηıwhich depends on the κ-connectivity degree at sensor
ξı. The per-hop delay is determined by
L℘
e=℘ −℘e
ηı.(23)
Then, the constraint is expressed as
N∪M
(ξı=1,ξj=1)
℘ξı,ξ
j≤L℘
e.(24)
Finally, the total amount of successfully transmitted data
packets through the optimal number of hops follows as:
Throughput ="Dtransmission
ξ#∗F∀ıj∈N$M.(25)
V. CANONICAL PARTICLE SWAR M OPTIMIZATION
ALGORITHM
To represent the population of swarm topology in the
deployment of sensors and Fog nodes, we employ com-
plex network connectivity to balance the tradeoff among the
κ−1 failure paths and communication overhead. Combination
of proactive and reactive routing mechanisms are initial-
ized to exchange demanding calculation on every sensor and
record the objective function which minimizes the energy
consumption and delay. Consequently, the degree of sen-
sors’ connectivity is adaptively increased or decreased through
matching sensors’ velocity to make a proper selection. Hence,
the optimization model for IIoT connectivity deployment is
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
10354 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 6, DECEMBER 2019
defined for minimizing energy spent and delay of a sensor
when transmitting a data packet of length Fıjbits by
min ⎡
⎣
υı,j ∈N∪M
−→
Z⎤
⎦(26a)
min )
sd∈N∪M
℘Lp*.(26b)
subject to
Energyυsd ,∀ı,j ∈S∪R(27a)
α
dυ(ı,j )3εmp
2Eelecυ(ı,j)≤rcυ(ı,j )(27b)
N∪M
(ξı=1,ξj=1)
℘ξı,ξ
j≤L℘
e(27c)
dα
υ(ı,j)≤rcυ(ı,j )≤cıjEnergyυıj
≤Energyυsdmax ∀ı,j ∈N∪M.(27d)
Based on our definition of the energy model for IIoT intro-
duced in Section IV-A1, it can be stated that if Energyυıjis
the energy consumed by sensor υıto connect with its neigh-
bors υj, then each sensor can adjust its transmitted power
to reach its neighbor. In other words, the hop-distance is
Hop =dα
υ(ı,j) ≤rcυ(ı,j ) , where αis the path-loss parameter.
Therefore, we add another two constraints to the optimization
model depending on the lower and upper bound values of
the power transmission range as well as the number of hops
according to the cut-off values of the search space. This
energy model of the selected path may be adaptively changed
within a closed interval of the lower and upper bound val-
ues such as [Energyυsdmin ,Energyυsdmax ], where Energyυsdmin
and Energyυsdmax determine the minimum and maximum con-
straints of the path. The optimization model is NP-hard, and
the key step to solve this model is to assume that the κ-disjoint
paths algorithm assigns each sensor, the transmission range
according to the hop-distance using (17) for each neighbor to
utilize the diversity of the swarm topology. However, each sen-
sor connects with its neighbors within a range of transmission
power which is limited to a small boundary. By employing
multiswarm network topology, each swarm in our formulation
works as a local optimizer with a different number of design
variables and constraints.
Using this feature provides an opportunity to control the
network topology in terms of post-deployment utilizing a
local search algorithm in each swarm. Additionally, employing
multiswarms raises the issue of information exchange among
swarms. In other words, each sensor has the ability to improve
collaborative learning behavior by exchanging messages with
their neighbors. Upon request of receiving these messages,
each sensor computes the disjoint paths as well as the local
path information updates. According to the objective functions
computed from (26a) and (26b), new potential paths are gen-
erated. Therefore, the sensors that are selected adaptively with
a velocity v(ı,j) are updated every iteration to satisfy the right
direction of selected paths. The pseudocode of the proposed
canonical particle multiswarm optimization (CPMSO) algo-
rithm is depicted in Algorithm 1, whereas the generic steps
of the algorithm start as follows. First, every swarm of sensor
nodes is initialized. Second, each swarm is identified by the
dominant connected sensor nodes. Then, each sensor connects
to other sensors generated in a new subswarm. The sensor con-
tinues to perform this operation until the network topology is
generated. The positions and velocities of each sensor are ini-
tialized. Then, each sensor evaluates the objective functions
in (26).
Specifically, assume that the κ-disjoint paths have m
variables, the position and velocity of sensor νare rep-
resented by m-dimensional vectors |ν|={ν1,ν
2,...,ν
N,
νN+1,...,ν
N+M}and ν={ν1,ν
2,...,ν
N,ν
N+1,..., ν
N+M},
respectively. Two positions, referred to as personal-best
position pbest and global-best position gbest are defined in
the proposed algorithm. During the solution of the objec-
tive functions in (26a) and (26b) for energy consumption
and delay, the sensors are connected in the whole search
space in each iteration. These sensors are influenced by
the individually exchanged messages, the personal extreme
value point as well as global extreme value point to allow
the sensors to select the next hop toward the extreme value
point within the scope of the search space in each iteration.
Therefore, the sensors deviate from the constraints domain
and hardly converge to the extreme value point of the
constraint domain. The whole personal-best positions of the
swarm imply the distribution of good objective functions as
defined in (26a) and (26b) that are related to the exchanged
messages satisfying constraints in (27a) and (27b). In each
path, the personal-best position of particle v(ı,j ), denoted
as p(best,ν(ı,j ))=(p(best,ν1),(best,ν2),...,(best,νN),(best,ν(N+M)))
and global-best position, denoted as g(best,ν(ı,j ))=
(g(best,ν1),(best,ν2),...,(best,νN),(best,νN+M)). The degree of influ-
ence of the personal-best position pbest is defined by the
coefficient of constraints φ1.
Likewise, the influence of the best global pbest is defined
by the coefficient of constraints φ2. The second term in (28)
is the “social” model, and the third term is the “cognition”
model. We use the symbol ⊗to represent the elementwise
multiobjective matrix-vector multiplication between objective
function −→
Zthat influences the personal-best position pbest
defined by coefficient of constraints φ1with objective function
−→
Zthat influences the best-global position gbest defined by
the coefficient of constraints φ2. Thus, the velocity update
function, which drives the CPSO, is defined as
−→
νv:=χνv+−→
Z(0,φ
1)⊗−→
pv−−→
xv+−→
Z(0,φ
2)⊗−→
gv−−→
xv
(28)
where −→
Zis a distribution of objective functions satisfying con-
straints defined in (27). χis a constriction coefficient, which
helps to balance the global exploration and local exploitation.
It is defined as [45]
χ=2
φ+φ2−4φ
,with φ=φ1+φ2>4.(29)
During the evolution process, the velocity update function is
referred to as a momentum representing the currently selected
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
HASAN AND AL-RIZZO: OPTIMIZATION OF SENSOR DEPLOYMENT FOR IIoT USING MULTISWARM ALGORITHM 10355
Algorithm 1: CPMSO
input:The number of sensor N;
input:The number of Fog nodes M;
input:The degree of connectivity κ;
for κ-disjoint paths from the Nto one of the M; do
Calculate Objective functions −→
Zfor each N
Set the velocity V:={
−→
υ1,...,−−−→
υN+M}:=
InitNode(−−−−−−−→
lowerbound,−−−−−−−→
upperbound)→∀
υ∈
{1,...,N+M}:−−−−→
N+Mυ:=
−→
U(−−−−−−−→
lowerbound,−−−−−−−→
upperbound)
−→
V:={
−→
υ1,...,−→
υυ}:=
InitParticleVelocities(−−−−−−−→
lowerbound,−−−−−−−→
upperbound)→
∀υ∈{1,...,N+M}:−−−−−−→
υ1,...,N+M:=
(−−−−−−−→
upperbound −−−−−−−−→
lowerbound)+−→
U(0,1)−
1
2(−−−−−−−→
upperbound −−−−−−−−→
lowerbound)
ϒ={
−→
ϒ1,....,−−−→
ϒN+M}:=
EvaluateObjectfunction(−→
Z)→∀
υ∈
{1,...,N+M}:ϒυ:=f(−→
Zυ)
Set the personal position vector of
P={
−→
p1,...,−−−→
pN+M}:=Initllocallyoptimal(Z)→Z
Set the personal positions vector of
P={pZ
1,...,pZ
υ}:=InitObjectivefunction(ϒ) →ϒ
Set the Global positions vector of G=
{−→
g1,...,−−−→
gN+M}:=Initgloballyoptimal(P,T)→P
Set the Global positions vector of G=
{gZ
1,...,gZ
1}:=Initgloballyoptimal(PZ,T)→PZ;
end
input:The network topology;
while termination condition not met do
for each node υof N +M; do
Update inertia weight
Update the velocity
−→
υυ=χ.(−→
υυ+−→
Z(0,ϕ
1)+(−→
pυ−−→
υN+M)+
−→
Z(0,ϕ
2)+(−→
gυ−−→
υυ))
−→
υp:=−→
Zp+−→
υp
end
Evaluate the object function of selected path
ϒ:=(Hop,Z)
Update the locally optimal
P,PZ:=(Hop,Z)→∀
p∈{1,...,N+M}:−→
pp,pZ
p:=
,−→
Zp,yiif ϒυbetter than pf
p
−→
pp,pf
potherwise
Update the global position
G,Gf:=Updategloballyoptimal(P,PZ,T)→∀
p∈
{1,...,N+M}:−→
gp,gZ
p:=
best(PTp,PZ
Tp), where Tpare the neighbors of p
end
best solution found
sensor’s direction, the social component which is the force
of attraction toward the best solution evaluated so far by the
neighbors, and the cognitive component which is the force
of attraction toward the previous solution the sensor was
aware of. The difference between the CPSO and fully particle
multiswarm optimization (FPMSO) algorithms is the velocity
update function, which describes not just the best position that
the sensor takes into account but also all of its neighbors.
Therefore, the velocity update function is defined as
−→
υv=χ−→
υv+1
N
N+M
υ=1
−→
Z(0,φ
1)⊗−→
pυ−−→
υυ.(30)
Taking full advantage of information exchange of all personal-
best messages avoids being trapped into local optimal solution.
This also strengthens the sensor’s ability to learn from other
sensors’ experience to guide its direction selection. Therefore,
the performance of each algorithm depends on the way the
sensors are influenced in the search space to achieve the objec-
tive functions. Fig. 10 illustrates the movement of particles
based on the steps of multiswarm and fully informed swarm
mechanism which are stated as follows.
Step 1: Calculate pbest’s objective function
in (26a) and (26b) in terms of energy con-
sumption and average delay. Then figure out the
minimal value of the objective function among
pbest’s objective functions for the κ-disjoint paths.
Step 2: Calculate a constriction coefficient χas in (29) in
order to prevent velocity explosion.
Step 3: Update the velocity value.
Step 4: Select the optimal solution to improve the fault-
tolerant paths routing.
VI. PERFORMANCE EVAL U AT IO N
Realistic WSNs experiments provide a more accurate, real-
istic, and replicable validation mechanism for our proposed
algorithm. However, the wide variety of available sensors
and Fog computing hardware can make it rather difficult to
replicate results produced by real experiments. Especially, in
some IIoT applications, where dangerous conditions should be
avoided, for example, banking industries, and the healthcare
industry which uses several sensors and Fog computing fitting
to the hospital wards to monitor patient’s health conditions
for various ailments. Therefore, in such cases, a real experi-
ment as the first testbed is not an appropriate choice. Besides,
there are a few testbed facilities that have emerged to overcome
the difficulties in testbed-based evaluation environments which
can satisfy some complex application needs. MATLAB pro-
vides several toolboxes to facilitate research in sensor network
programming environments, communication protocols, system
design, and applications. Therefore, MATLAB provides a reli-
able and cost-effective work environment for the researcher
enabling a broad range of innovations. Finally, MATLAB
offers the ability to rapidly prototype new ideas as well as
the ability to duplicate real-world scenarios.
Compared to Castalia, a simulator for WSNs and body area
networks (BANs), MATLAB has many advantages. First, the
developers and/or programmers can expect the reduction of
the costs of both deployment and maintenance of a large-
scale testbed. Second, a variety of platforms are available with
different capabilities, and hence, the deployment of hetero-
geneous and homogeneous sensors and other nodes, such as
Fog and cloud computing, becomes feasible. Third, it pre-
vents unexpected or erroneous behavior from interfering with
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
10356 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 6, DECEMBER 2019
(a)
(b)
Fig. 10. (a) CPMSO. (b) FIPSO.
production services or other testing. Fourth, the level of exper-
tise needed to set up experiments is less compared to that in
conventional testbeds. Finally, the developers and/or program-
mers can focus their development and evaluation efforts on
the area of contribution without worrying about other network
tasks.
A. Simulation Setup
We conducted simulations by implementing the proposed
algorithm in MATLAB to generate the network topology and
evaluate the objective functions. In order to present differ-
ent IoT connectivity, we generated the network topology by
deploying various numbers of homogenous and heterogenous
sensor and Fog nodes uniformly distributed over a 2-D area of
1000 m×1000 m as shown in Fig. 11. The sensors are placed
√3.rsapart without overlap, with or without holes. We sim-
ulate different types of topologies to investigate the effect of
κ-degree of connectivity on the performance of the algorithm
relative to the number of evaluations of objective functions
in terms of energy consumption, delay, and throughput. We
assume that each sensor consumes energy of 50 nJ/bit in
transmitting/receiving data packets, meanwhile the transmitter
consumes extra 100 pJ/bit/m2. The initial value of transmis-
sion range of the sensors is set to be 12.00 m and varies from
9to39mwhile(rc/rs)varies from 0.3to1.8 to guarantee
the connectivity among the Fog nodes and sensors and to sat-
isfy the constraints of the problem. The parameters used in
the simulations are summarized in Table III.
B. Simulation Results
1) QoS Parameters Versus the Number of Deployed Nodes:
Fig. 12 illustrates the effect of using the swarm mechanism on
Fig. 11. Possible topology connectivity of each particle during initialization.
TABLE III
SIMULATION PARAMETERS
HML-IIoT topology. The results reveal that the connectivity
of a particle increased using the multiswarm allowing a sensor
to select a new neighborhood. This happens during the search
process to keep track of the searching performance of each par-
ticle and then makes appropriate adjustments on the particle’s
connectivity. Basically, the number of hops required to report
an event increased when new sensors are connected. However,
the CPMSO, FPMSO, and CPSO algorithms search to find the
optimal number of hops that minimize delay and energy con-
sumption. Therefore, the energy consumption decreases with
the increasing number of iterations while satisfying the con-
straints in terms of the number of hops required to report the
event without exceeding the intermediate power transmission
radius. Moreover, the number of constraints is 2|N[ι]|+2,
where |N[ι]|defines the number of sensors in the networks.
Therefore, the size of the search space increases proportion-
ately to the sensor density. From Fig. 12(a), we notice that the
total κ-disjoint energy consumption generated by the proposed
algorithm is better than the CPSO. This is because the objec-
tive function of the CPSO encounters difficulties in discovering
the κ-disjoint paths after recovery from the fault-tolerance
error messages since the search space is large, resulting in the
inability to substitute with the κ-disjoint paths and thus high
energy consumption. Another interesting observation is that
the number of sensors and Fog nodes is regular. Therefore, the
maximum number of hops decreases as the transmission range
increases. This is because the maximized transmission range
allows to cover more of the neighboring set of sensors, thus
reducing the number of Fog nodes required for deployment.
Additionally, we address the impact of the relationship
between the number of hops and the exchange of messages for
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
HASAN AND AL-RIZZO: OPTIMIZATION OF SENSOR DEPLOYMENT FOR IIoT USING MULTISWARM ALGORITHM 10357
(a)
(b)
(c)
Fig. 12. QoS of deployed IIoT with 50 sensor nodes with different transmis-
sion ranges. (a) Energy consumption of deployed IIoT with 50 sensor nodes.
(b) Average delay of deployed IIoT with 50 sensor nodes. (c) Throughput of
deployed IIoT with 50 sensor nodes.
fault-tolerance between sensors and Fog nodes. We observe
that the CPSO performs worse than FPMSO and CPMSO.
This is because the CPSO requires more control messages to
be exchanged between the Fog nodes and sensors deployed in
IIoT. Therefore, the CPSO needs to find the κ-disjoint paths in
its reachable neighborhood, whereas the FPMSO and CPMSO
can directly search for paths using fewer control messages
between the reachable neighborhoods. Furthermore, FPMSO
and CPMSO make less use than CPSO of control messages
between the reachable neighborhoods produced by all of the
new reachable neighborhoods qualified for determining the
new position. Therefore, the new particle swarm regenerates
optimal κ-disjoint paths for FPMSO and CPMSO that can
achieve lower total energy consumption compared to the CPSO
algorithm. We notice that the average delay of the κ-disjoint
generated by the proposed algorithm is better than the CPSO.
Again, the number of sensors and Fog nodes is regular,
while the maximal number of hops has decreased as the trans-
mission range increases. Therefore, the number of sensors to
be deployed in IIoT decreases to make fewer paths available.
Meanwhile, the minimal number of hops leads to proportion-
ally increased number of sensors and Fog nodes deployed in
the IIoT framework as well as more paths becoming avail-
able for simultaneously routing data and control messages to
destinations. Fig. 12(b) shows the average delay of packet
transmission in the selected κ-disjoint paths.
The number of subswarms being irregular, therefore the
sensors used in FPMSO and CPMSO are reduced with the
increase of the transmitted power. This is because the num-
ber of sensors selected inside a subswarm is less than the
total number of sensors. The number of sensors deployed
using CPSO is larger than those of the FPMSO and the
CPMSO algorithms, leading to a lower delay per-hop. This
can be attributed to sensors that successfully updated their
information for the next selection and maintenance of the
κ-disjoint paths which can satisfy the hop availability require-
ment by selecting the next hop in the neighborhood of each
sensor. This information enables sensors to improve the con-
nectivity which subsequently enables FPMSO and CPMSO
to exchange fewer control messages for fault tolerance com-
pared to CPSO for selecting and maintaining single-hop
neighborhood. This indicates that both FPMSO and CPMSO
provide better searching accuracy and convergence speed of
the feasible search space for the κ-disjoint paths than CPSO.
Throughput may frequently be degraded due to high bit-
error rate or other conditions such as environmental ones. In
Fig. 12(c), we present the effect of number of hops to solve
the objective function in (26a) and (26b). We observe that
when the delay is minimized with increasing optimal num-
ber of hops, this leads to proportionally increased number of
sensors and Fog nodes to be deployed in IIoT framework as
well as more paths becoming available, leading to throughput
degradation. Actually, this is expected as minimizing delay
under the aforementioned constraints with optimal number of
hops leads to a lower number of exchanged control messages
for fault tolerance.
Finally, we conclude that the performance attained depends
on the construction of the network topology, such as the
distance-independent and optimal number of hops for each
sensor to achieve a connection to all sensors in the swarm
as neighbors. Therefore, the results of the comparison show
that the average objective function values in terms of energy
consumption for the FPMSO algorithm followed by CPMSO
are approximately equal to 92.39% and 84.52%, respectively.
Thus, it can be observed that the best performing algorithm
is the FPMSO algorithm, followed by the CPMSO algorithm
for 50 sensors. Meanwhile, CPSO has the worst value which
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
10358 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 6, DECEMBER 2019
(a)
(b)
(c)
Fig. 13. Total optimizing energy consumption with the number of connectiv-
ity of devices deployed IIoT. (a) Canonical particle swarm. (b) Fully particle
multiswarm. (c) Canonical particle multiswarm.
is approximately equal to 71.83% under the same setting.
Although the CPSO achieves fully connected topology, i.e.,
when each sensor is connected to all sensors in the swarm as
neighbors, it exhibits a particularly inferior performance com-
pared to others. This is because of the simultaneous attraction
of the κ-disjoint paths provoking random behavior for each
sensor to discover, construct, and select paths. Meanwhile,
this behavior could support optimal performance in FPMSO
and CPMSO with full connectivity. In other words, solving
network connectivity using multiswarm provides the capabil-
ity to achieve optimal solution using lower number of fitness
(a)
(b)
(c)
Fig. 14. Total optimizing average delay with the number of connectivity
of devices deployed IIoT. (a) Canonical particle swarm. (b) Fully particle
multiswarm. (c) Canonical particle multiswarm.
function evaluations in comparison with CPSO. Since multi-
swarm is capable of generating new subswarms and employing
them for grouping the new particles in the search space, there-
fore, new κ-disjoint paths are generated to improve the ability
of FMPSO and CPMSO to escape from local optimal solution
of the network connectivity.
2) QoS Parameters With Different Number of Connectivity:
We continue our investigation by optimizing the deployment of
sensors as a second scenario which requires information from
multisources at specific intervals. Therefore, our proposed
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
HASAN AND AL-RIZZO: OPTIMIZATION OF SENSOR DEPLOYMENT FOR IIoT USING MULTISWARM ALGORITHM 10359
(a)
(b)
(c)
Fig. 15. Total optimizing throughput with number of connectivity of devices
deployed IIoT. (a) Canonical particle swarm. (b) Fully particle multiswarm.
(c) Canonical particle multiswarm.
optimization algorithm is applied to transmit information
between sensors and Fog nodes while satisfying QoS parame-
ters. This leads to the generation of results for the total number
of connections between the sensors and Fog nodes deployed
during the execution of the CPSO, FPMSO, and CPMSO. We
consider messages exchanged for fault tolerance between the
Fog nodes and sensors as a metric to generate the topologies
in IIoT framework. In Figs. 13–15, we present the execution
of the algorithms to optimize the QoS parameters in terms of
energy consumption, average delay, and throughput where the
total number of connectivity, κ=3,4,5 relative to the number
(a)
(b)
(c)
Fig. 16. Performance evaluation of CPSO algorithm with different numbers
of connectivity of devices deployed IIoT. (a) Energy consumption. (b) Delay.
(c) Throughput.
of evaluations. Generally, we noticed that the performance of
the algorithm varies with increasing connectivity among the
sensors and Fog nodes. Since all information between the sen-
sors and Fog nodes is available, therefore, the topology created
with 4 ≤κ≤5 has all information shared, which tends to
explore more sensors and this in turn tends to create more
diverse paths. Meanwhile, in case 3 =κ≤2, the topology
does not have all information shared, i.e., only information
between two neighboring sensors, therefore, the topology
has less information to be shared among the sensors for a
predefined connectivity. This leads to exploring and creating
less diverse paths during the evaluation of objective functions.
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
10360 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 6, DECEMBER 2019
(a)
(b)
(c)
Fig. 17. Performance evaluation of fully multi-PSO algorithm with different
numbers of connectivity of devices deployed IIoT. (a) Energy consumption.
(b) Delay. (c) Throughput.
In Figs. 13(a)–(c), 14(a)–(c), and 15(a)–(c), we present the
optimized energy consumption, average delay, and through-
put, respectively, using the CPSO, FPMSO, and CPMSO with
κ=3,4,5. We observe that the performance of the proposed
multiswarm algorithm is better than canonical swarm because
the multiswarm tends to create more diverse paths from multi-
source nodes to a gateway. Both FPMSO and CPMSO require
less exchange of control messages between the sensors to
obtain the connectivity with each other and Fog nodes due
to the availability of all information between the sensors and
Fog nodes. In some cases, as seen in Figs. 13(b) and 14(b), we
have found that the optimal performance is obtained from the
(a)
(b)
(c)
Fig. 18. Performance evaluation of canonical multiswarm PSO algorithm
with different numbers of connectivity of devices deployed IIoT. (a) Energy
consumption. (b) Delay. (c) Throughput.
FMPSO with predefined connectivity, κ=4. This is because
of the behavior of some nodes in exchanging the information
which indeed can be beneficial in constructing and selecting
the optimal paths.
3) QoS Parameters With Increasing the Number of
Connectivity: We end our investigation on the performance
of the proposed algorithm by finding an optimal deploy-
ment topology pattern with increasing degree of connectivity
in order to achieve coverage and connectivity for a com-
plete range of transmission power range and sensing range
with minimum energy consumption and delay. We compare
three deployment patterns: 1) ring; 2) mesh; and 3) sub-
swarm through increasing the number of κas 5 ≤κ≤10
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
HASAN AND AL-RIZZO: OPTIMIZATION OF SENSOR DEPLOYMENT FOR IIoT USING MULTISWARM ALGORITHM 10361
In Fig. 16(a)–(c), we present the optimized energy consump-
tion, average delay, and throughput, respectively, using the
CPSO algorithm. Energy consumption and average delay are
presented using CPSO, FMPSO, and CMPSO in Figs. 16(a)
and (b), 17(a) and (b), and 18(a) and (b), respectively. We
observe that energy consumption and average delay are higher
when the degree of connectivity is increased in all deployment
patterns in terms of ring, mesh, and subswarm mesh of sen-
sor nodes. Throughput obtained using CPSO, FMPSO, and
CMPSO is shown in Figs. 16(c)–18(c), respectively. CPMSO
and FMPSO improve the throughput by approximately 95.23%
which represents better performance in finding optimal solu-
tion as compared to CPSO algorithm which is approximately
equal to 75.48%. Meanwhile, the energy consumption is min-
imized by 87.5% and the delay by 95.00% as compared
with CPSO which are approximately equal to 70.15% and
75.48%, respectively. In conclusion, it is observed that for
all deployment patterns in terms of ring, mesh, and sub-
swarm mesh, the optimal energy consumption, delay, and
throughput are achieved when the degree of connectivity
is increased. The ring and mesh deployment patterns using
CPSO as shown in Fig. 16(a)–(c) show good performance
when increasing the degree of connectivity compared to the
performance of FMPSO as shown in Fig. 17(a)–(c) and
CMSPO as shown in Fig. 18(a)–(c). In all deployment patterns,
information that requires control messages for fault tolerance
is available; for the ring pattern the information is shared
between two sensors, while with mesh and subswarm mesh
information is shared among a group of predefined size of
connectivity. Therefore, the latter deployment patterns have all
information to explore more κ-disjoint paths in their reachable
neighborhoods, consuming more energy and increasing delay
while transmitting along more diverse paths.
VII. CONCLUSION
Due to the requirements of IIoT applications and deploy-
ment environment, IIoT is vulnerable to large-scale discon-
necting devices. Therefore, connectivity and network topology
control can be classified under different objective functions
and constraints, such as energy efficiency and minimum num-
ber of sensor nodes that need to be deployed to satisfy IIoT’s
requirements. Swarm optimization strategy has many struc-
tural properties, such as the number of particles, clustering,
and degree of distribution. In this article, we investigated the
performance of multiswarm optimization strategy developed
to find the deployment pattern of the κ-disjoint path routes in
order to reduce energy consumption and delay. Our goal was to
dynamically evolve swarm techniques in IIoT according to the
environmental changes. The proposed algorithm offers robust
connectivity for addition and removal of sensors against con-
gestion or event-driven burst traffic, or the physical network
topology through inspiration by swarm network topology fea-
tures. We focused particularly on similarity between the swarm
topology structure of various CPSO to validate our proposed
strategy. We assessed an objective function taking into consid-
eration energy consumption, in-network delay, and throughput.
Our results show that using the characteristics of all personal-
best information is a valid strategy for the improvement of
the CPMSO performance. Moreover, we demonstrated that
the proposed algorithm outperforms existing algorithms which
optimize the energy consumption and average delay on the
explored paths.
ACKNOWLEDGMENT
The authors would like to thank the anonymous reviewers
for their very constructive and helpful comments and sugges-
tions which have resulted in significant improvement of the
content and quality of this article.
REFERENCES
[1] M. Aazam, S. Zeadally, and K. A. Harras, “Deploying Fog computing
in Industrial Internet of Things and Industry 4.0,” IEEE Trans. Ind.
Informat., vol. 14, no. 10, pp. 4674–4682, Oct. 2018.
[2] L. D. Xu, W. He, and S. Li, “Internet of Things in industries: A survey,”
IEEE Trans. Ind. Informat., vol. 10, no. 4, pp. 2233–2243, Nov. 2014.
[3] C. Gomez and J. Paradells, “Urban automation networks: Current and
emerging solutions for sensed data collection and actuation in smart
cities,” Sensors, vol. 15, no. 9, pp. 22874–22898, Sep. 2015.
[4] S. Rani, R. Talwar, J. Malhotra, S. H. Ahmed, M. Sarkar, and H. Song,
“A novel scheme for an energy efficient Internet of Things based on
wireless sensor networks,” Sensors, vol. 15, no. 11, pp. 28603–28626,
Nov. 2015.
[5] M. Chiang and T. Zhang, “Fog and IoT: An overview of research
opportunities,” IEEE Internet Things J., vol. 3, no. 6, pp. 854–864,
Dec. 2016.
[6] E. Ever, F. M. Al-Turjman, H. Zahmatkesh, and M. Riza, “Modelling
green HetNets in dynamic ultra large-scale applications: A case-study
for femtocells in smart-cities,” Comput. Netw., vol. 128, pp. 78–93,
Dec. 2017.
[7] J. Lin, W. Yu, N. Zhang, X. Yang, H. Zhang, and W. Zhao, “A sur-
vey on Internet of Things: Architecture, enabling technologies, security
and privacy, and applications,” IEEE Internet Things J., vol. 4, no. 5,
pp. 1125–1142, Oct. 2017.
[8] M. Z. Hasan and H. Al-Rizzo, “Task scheduling in Internet of
Things cloud environment using a robust particle swarm optimization,”
Concurrency Comput. Pract. Exp., vol. 30, no. 8, Jul. 2019,
Art. no. e5442.
[9] F. Al-Turjman, M. Z. Hasan, and H. Al-Rizzo, “Task scheduling
in cloud-based survivability applications using swarm optimization
in IoT,” Trans. Emerg. Telecommun. Technol., vol. 30, no. 8, 2018,
Art. no. e3539.
[10] Y. Huang, J.-F. Martínez, J. Sendra, and L. López, “Resilient wireless
sensor networks using topology control: A review,” Sensors, vol. 15,
no. 10, pp. 24735–24770, 2015.
[11] W. Dong and M. Zhou, “A supervised learning and control method to
improve particle swarm optimization algorithms,” IEEE Trans. Syst.,
Man, Cybern., Syst., vol. 47, no. 7, pp. 1135–1148, Jul. 2017.
[12] M. Z. Hasan, H. Al-Rizzo, and M. Günay, “Lifetime maximization by
partitioning approach in wireless sensor networks,” EURASIP J. Wireless
Commun. Netw., vol. 2017, no. 1, p. 15, Jan. 2017.
[13] M. Z. Hasan and F. Al-Turjman, “Optimizing multipath routing with
guaranteed fault tolerance in Internet of Things,” IEEE Sensors J.,
vol. 17, no. 19, pp. 6463–6473, Oct. 2017.
[14] M. Z. Hasan, H. Al-Rizzo, and F. Al-Turjman, “A survey on multipath
routing protocols for QoS assurances in real-time wireless multimedia
sensor networks,” IEEE Commun. Surveys Tuts., vol. 19, no. 3,
pp. 1424–1456, 3rd Quart., 2017.
[15] R. Haghighi and C.-C. Cheah, “Topology-based controllability problem
in network systems,” IEEE Trans. Syst., Man, Cybern., Syst., vol. 47,
no. 11, pp. 3077–3088, Nov. 2017.
[16] Z. Meng, Z. Wu, and J. Gray, “A collaboration-oriented M2M messaging
mechanism for the collaborative automation between machines in future
industrial networks,” Sensors, vol. 17, no. 11, Nov. 2017, Art. no. 2694.
[17] Z. Meng, Z. Wu, C. Muvianto, and J. Gray, “A data-oriented M2M
messaging mechanism for industrial IoT applications,” IEEE Internet
Things J., vol. 4, no. 1, pp. 236–246, Feb. 2017.
[18] S. Tang and L. Tan, “Reward rate maximization and optimal transmission
policy of EH device with temporal death in EH-WSNs,” IEEE Trans.
Wireless Commun., vol. 16, no. 2, pp. 1157–1167, Feb. 2017.
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
10362 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 6, DECEMBER 2019
[19] L. Tan and S. Tang, “Energy harvesting wireless sensor node with tempo-
ral death: Novel models and analyses,” IEEE/ACM Trans. Netw., vol. 25,
no. 2, pp. 896–909, Apr. 2017.
[20] A. Khalil, N. Mbarek, and O. Togni, “Self-configuring IoT service QoS
guarantee using QBAIoT,” Computers, vol. 7, no. 4, p. 64, 2018.
[21] S. Toyonaga, D. Kominami, and M. Murata, “Virtual wireless sensor
networks: Adaptive brain-inspired configuration for Internet of Things
applications,” Sensors, vol. 16, no. 8, 2016, Art. no. 1323.
[22] M. Handte, S. Foell, S. Wagner, G. Kortuem, and P. J. Marrón,
“An Internet-of-Things enabled connected navigation system for urban
bus riders,” IEEE Internet Things J., vol. 3, no. 5, pp. 735–744,
Oct. 2016.
[23] S. Fang et al., “An integrated system for regional environmental moni-
toring and management based on Internet of Things,” IEEE Trans. Ind.
Informat., vol. 10, no. 2, pp. 1596–1605, May 2014.
[24] E. Sisinni, A. Saifullah, S. Han, U. Jennehag, and M. Gidlund,
“Industrial Internet of Things: Challenges, opportunities, and direc-
tions,” IEEE Trans. Ind. Informat., vol. 14, no. 11, pp. 4724–4734,
Nov. 2018.
[25] A. H. Ngu, M. Gutierrez, V. Metsis, S. Nepal, and Q. Z. Sheng,
“IoT middleware: A survey on issues and enabling technologies,” IEEE
Internet Things J., vol. 4, no. 1, pp. 1–20, Feb. 2017.
[26] S. Savazzi, V. Rampa, and U. Spagnolini, “Wireless cloud networks for
the factory of things: Connectivity modeling and layout design,” IEEE
Internet Things J., vol. 1, no. 2, pp. 180–195, Apr. 2014.
[27] M. Z. Hasan, F. Al-Turjman, and H. Al-Rizzo, “Analysis of cross-layer
design of quality-of-service forward geographic wireless sensor network
routing strategies in green Internet of Things,” IEEE Access,vol.6,
pp. 20371–20389, 2018.
[28] M. T. Hajiaghayi, N. Immorlica, and V. S. Mirrokni, “Power
optimization in fault-tolerant topology control algorithms for wire-
less multi-hop networks,” IEEE/ACM Trans. Netw., vol. 15, no. 6,
pp. 1345–1358, Dec. 2007.
[29] T. Song, R. Li, B. Mei, J. Yu, X. Xing, and X. Cheng, “A
privacy preserving communication protocol for IoT applications in
smart homes,” IEEE Internet Things J., vol. 4, no. 6, pp. 1844–1852,
Dec. 2017.
[30] D. Garcia-Carrillo and R. Marin-Lopez, “Multihop bootstrapping with
EAP through CoAP intermediaries for IoT,” IEEE Internet Things J.,
vol. 5, no. 5, pp. 4003–4017, Oct. 2018.
[31] N. Ulltveit-Moe, H. Nergaard, L. Erdödi, T. Gjøsæter, E. Kolstad, and
P. Berg. (2016). Secure Information Sharing in an Industrial Internet of
Things. [Online]. Available: http://arxiv.org/abs/1601.04301
[32] M. Frey et al., “Security for the Industrial IoT: The case for information-
centric networking,” arXiv preprint arXiv:1810.04645, 2018.
[33] C. Vimalarani, R. Subramanian, and S. N. Sivanandam, “An enhanced
PSO-based clustering energy optimization algorithm for wireless sensor
network,” Sci. World J., vol. 2016, Dec. 2016, Art. no. 8658760.
[34] N. Li and J. C. Hou, “FLSS: A fault-tolerant topology control algorithm
for wireless networks,” in Proc. ACM 10th Annu. Int. Conf. Mobile
Comput. Netw., 2004, pp. 275–286.
[35] Y. C. Yee, S. W. Tan, H. S. Lim, and S. F. Chien, “Application of
particle swarm optimizer on load distribution for hybrid network selec-
tion scheme in heterogeneous wireless networks,” ISRN Commun. Netw.,
vol. 2012, pp. 1–7, Sep. 2012. doi: 10.5402/2012/340720.
[36] S. A. Fernandez, A. A. Juan, J. de Armas Adrián, D. G. e Silva, and
D. R. Terrén, “Metaheuristics in telecommunication systems: Network
design, routing, and allocation problems,” IEEE Syst. J., vol. 12, no. 4,
pp. 3948–3957, Dec. 2018.
[37] X. Han, X. Cao, E. L. Lloyd, and C.-C. Shen, “Fault-tolerant relay node
placement in heterogeneous wireless sensor networks,” IEEE Trans.
Mobile Comput., vol. 9, no. 5, pp. 643–656, May 2010.
[38] H.-L. Shieh, C.-C. Kuo, and C.-M. Chiang, “Modified particle swarm
optimization algorithm with simulated annealing behavior and its numer-
ical verification,” Appl. Math. Comput., vol. 218, no. 8, pp. 4365–4383,
2011.
[39] M. Z. Hasan, F. Al-Turjman, and H. Al-Rizzo, “Optimized multi-
constrained quality-of-service multipath routing approach for multimedia
sensor networks,” IEEE Sensors J., vol. 17, no. 7, pp. 2298–2309,
Apr. 2017.
[40] W. Yuan, Y. Liu, H. Wang, and Y. Cao, “A geometric structure-based
particle swarm optimization algorithm for multiobjective problems,”
IEEE Trans. Syst., Man, Cybern., Syst., vol. 47, no. 9, pp. 2516–2537,
Sep. 2017.
[41] M. P. Fanti, G. Faraut, J.-J. Lesage, and M. Roccotelli, “An integrated
framework for binary sensor placement and inhabitants location track-
ing,” IEEE Trans. Syst., Man, Cybern., Syst., vol. 48, no. 1, pp. 154–160,
Jan. 2018.
[42] X. Feng, Y. Wang, H. Yu, and F. Luo, “A novel intelligence algorithm
based on the social group optimization behaviors,” IEEE Trans. Syst.,
Man, Cybern., Syst., vol. 48, no. 1, pp. 65–76, Jan. 2018.
[43] M. A. M. de Oca and T. Stützle, “Convergence behavior of the fully
informed particle swarm optimization algorithm,” in Proc. ACM 10th
Annu. Conf. Genet. Evol. Comput., Atlanta, GA, USA, 2008, pp. 71–78.
[44] N. B. Long, H. Tran-Dang, and D.-S. Kim, “Energy-aware real-time
routing for large-scale industrial Internet of Things,” IEEE Internet
Things J., vol. 5, no. 3, pp. 2190–2199, Jun. 2018.
[45] M. Z. Hasan and F. Al-Turjman, “SWARM-based data delivery in social
Internet of Things,” Future Gener. Comput. Syst., vol. 92, pp. 821–836,
Mar. 2019.
Authorized licensed use limited to: University of Arkansas Little Rock. Downloaded on February 20,2020 at 02:44:28 UTC from IEEE Xplore. Restrictions apply.
