Optimizing the Sensor Deployment strategy for Large-Scale Internet of Things

(IoT) using Artificial Bee Colony

Author1, Author2, Author3 etc

1,2,3
UG Scholar, Department of IT, Kongu Engineering College, India
1
[email protected] (All authors mail ID)

Abstract:
In the world of Internet of Things (IoT), specifically in the environmental and nature monitoring strategy, recent
developments in the low-power and Long range (LoRa) communication systems have ignited up with new
possibilities. However, maintenance has been greatly challenged by the flexible climate and long distance travel.
Previous studies have shown that higher electronic failure rates are exponentially accelerated by temperature. The
cost of maintenance can be as high as 80% of the overall cost of maintenance. Expenses for deployment is extremely
high, if careful handling is not there. In this proposed paper, a problem with the deployment of sensors to
preventively mitigate maintenance costs thus maintaining tolerable efficiency of the sensing quality metrics as well
as the complete connectivity were formulated. A Cost model is proposed for Maintenance considering the
degradation for IoT networks and battery depletion in addition to thermal degradation. The spatial phenomenon
methodology is analyzed in order to adopt sensing quality metric based on shared results. Although problem
proposed is the conceptual one, it is brought out with the sparse nonlinear optimizer form to solve the problem. To
make the solution optimal, two algorithms namely CPSO (Canonical Particle Swarm Optimization) and ABC
(Artificial Bee Colony) are applied which are population based metaheuristics algorithms. Compared to the current
greedy heuristics, our meta-heuristics demonstrate good results for maintenance costs under the same appropriate
sensing efficiency.
Keywords: CPSO (Canonical Particle Swarm Optimization), ABC (Artificial Bee Colony) Optimization, Internet of
Things.

INTRODUCTION

With blooming technologies that make our lives smart, there is an underlying need to make these
technologies more efficient and affordable. These technologies are mainly based on WSN abbreviated as
Wireless Sensor Networks. There is an underlying need to reconfigure the sensors automatically based on
the operations to be performed. So, the emerging technology named as Long Range (LoRa) is developed
for the Internet of Things (IoT). It acquires considerable popularity in the manufacturing and research
communities in recent years. To provide data with long-life battery commitments, good transmission
distances, and huge node density, IoT uses chirp-based spectrum modulation [1]. LPWAN (Low Power
Wide Area Networks) was developed using LoRa and it has made large-scale WSN technology feasible.
LPWAN technology with a wider band based on spread spectrum technique. For smart city applications,
the most acceptable candidates are [2] are made by efficient costs, long range and energy efficacy of
LPWANs.
CPSO (Canonical Particle Swarm Optimization) and ABC (Artificial Bee Colony) has two goals upon its
implementation where the first one is to optimize the sensor node deployment and the other is to monitor
the environment and to reduce the maintenance cost. The Maintenance cost is mainly calculated by a
device's MTTF (Mean Time To Failure), i.e. the estimated time to fail and time for battery depletion. To
estimate the precise MTTF and battery depletion time, the well-known thermal ageing approach and power
models are integrated with the ever increasing temperature factor.
 RELATED WORK

Initially, the sensing model is assumed like a disc, whether it’s a linear or a modeling approach, they
intended to sense the targeted area with a minimal sensors count [3]. Subsequent research placed functional
restrictions on the initial formulation or converted the problem into multiple objective optimization [4]. The
original event model detection, however, is unable to stay upto date of the current frameworks of
environmental control that generate continuous reading strategy. The rigorous evidence on the problem
which was identified as NP completeness was given by Krause et al. [5]. They demonstrated that their
insightful and cost-effective position of polynomial-time heuristic padded sensor placements is at the most
constant factor worsen than the solution given as optimal.
GP model is utilized in many of the sensor related works and examined the sensor deployment strategy for
many industrial applications. The greedy shared knowledge maximization algorithm was specifically
applied by Wu et al. [8] to monitor the soil moisture. An extreme attempt is made by Du et al [9] to deploy
wind sensors and the simulation results produced at the most informative locations.
Recent studies by Boubrima et al [6] have examined the selection of the less count of grading points that to
provide the connectivity and the adequate accuracy of monitoring for non-rural air surveillance. The
measures are calculated through interpolation at positions not controlled.
The multiple objective of sensor deployment problem for the structural health monitoring was studied by
Bhuiyan et al [7], optimizing sensing accuracy, communication costs, and network lifetime. The authors
explicitly suggested a three-step heuristic positioning of the sensor with different phases for the high-end
nodes, low-end nodes, and finally the redundant nodes.

PROPOSED METHODOLOGY

A. Particle Swarm Optimization

Particle Swarm Optimization (PSO) is a search algorithm for stochastic and population-based intelligence
that had been shown to effectively solve the optimization problems. The earliest PSO strategy tried to
implement the idea of the social swarm activity, which ends in collection of particles distributing. The
determination of the fitness of the particle means fitness of the particle close to the global solution is
smaller than the non-connected particle. By integrating some features of the global solution and best-
fitness locations with one of the entire swarm participants, PSO strategy leads each and every particle to
travel via the search space in a direction.
The best global category of the PSO strategy determines the position of a particle that is stored as the best
previous location at its lowest cost. In addition, the global-best solution describes the location of the best
particle. For each particle, the algorithm decides which some other particle was its closest neighbour, then
allocates the velocity of the particle to another that interacts among others. The process of updating is
repeated until either the best global appropriate solutions is achieved or the maximum number of iteration
is reached [10].

The velocity for updating the forthcoming position of a particle is as follows

𝑉 𝑘+1 (1,𝑗) = 𝑉 𝑘 (1,𝑗) + 𝑐1 𝑟1 (𝑋𝑏𝑒𝑠𝑡 𝑘 (1,𝑗) − 𝑋 𝑘 (1,𝑗) ) + 𝑐2 𝑟2 (𝑋𝑔𝑏𝑒𝑠𝑡 𝑘 (𝑗) − 𝑋 𝑘 (1,𝑗) (1)

And

𝑋 𝑘+1 (1,𝑗) = 𝑋 𝑘 (1,𝑗) + 𝑉 𝑘+1 (1,𝑗) (2a)

1, 𝑗 = 1, 2, 3, … , 𝑁 + 𝑀 (2b)
 where 1, 𝑗 indicate the sensor’s index; 𝑋1 (1,𝑗) 𝑑 𝑘 (1,𝑗) , and 𝑉 𝑘 (1,𝑗) are the 𝑗𝑡ℎ component of the 1𝑡ℎ
sensor’s position, distance, and velocity vector respectively, in the 𝑘 𝑡ℎ iteration; 𝑟1 and 𝑟2 are the two
arbitrary numbers generated from a evenly dispersed function in the range [0,1]; 𝑋𝑏𝑒𝑠𝑡 𝑘 (1,𝑗) and
𝑋𝑔𝑏𝑒𝑠𝑡 𝑘 (𝑗) indicate the best and global positions observed so far by the 1𝑡ℎ sensor and whole swarm
topology, respectively; and 𝑐1 and 𝑐2 represent the confidence particle as in perception and social
behavior, respectively.

ALGORITHM - PSO
1. Input : Sensor data
2. Process :
3. Initialize inertia weight :
wx, wn
4. Initialize maximum iteration :
max _iter
5. Initialize population and velocity :
x, v
6. Update inertia weight :
(wx−wn) ∗ ite
w = wx − max _𝑖𝑡𝑒𝑟
7. Update position
𝑉(𝑖,𝑗) = w × (𝑉𝑖,𝑗 + 𝑐1 ∗ 𝑟𝑎𝑛𝑑() ∗ (𝑝𝑏𝑒𝑠𝑡(𝑖,𝑗) − 𝑋(𝑖,𝑗) ) …

+𝑐2 ∗ 𝑟𝑎𝑛𝑑() ∗ (𝑔𝑏𝑒𝑠𝑡(1,𝑗) − 𝑋(𝑖,𝑗) )

8. Evaluate fitness value
9. Update pbest and fitness
10. Update gbest
11. Repeat steps 3 to 8 until global best solution is found
12. Output : Best solution found

B. Canonical Particle Swarm Optimization

PSO swarm topologies will significantly affect the algorithm's efficiency. In addition, CPSO utilises a
fully related topology in which each and every sensors are neighbours. This guides to the direct relation
of a sensor to the world's best sensor and simultaneously affects it. With a set of benchmark functions, the
presented algorithm was evaluated and analysed both in terms of its efficacy, premature convergence and
the ability to prevent local optimum.
For continuous optimization problems, we concentrate on finding a stable, optimal solution. Updating
velocity means the best restricted neighbourhood sensor location to determine communication with other
neighbourhood sensor instead of the whole swarm topology. PSO imbalances the exploration versus
exploitation. To rectify that we propose CPSO. In addition, CPSO utilizes a fully related topology in which
each and every sensors are neighbours. This guides to directly link a sensor to the global best sensor.

Velocity Formulation

𝑉 𝑘+1 (1,𝑗) = x{𝑉 𝑘 (1,𝑗) + 𝑐1 𝑟1 (𝑋𝑏𝑒𝑠𝑡 𝑘 (1,𝑗) − 𝑋 𝑘 (1,𝑗) ) + 𝑐2 𝑟2 (𝑋𝑔𝑏𝑒𝑠𝑡 𝑘 (𝑗) − 𝑋 𝑘 (1,𝑗) )} (3)
Weight factor has been defined to attain and to maintain the exploration and exploitation
 2
𝑥=| | (4)
2−𝜙−√𝜙 2 −4𝜙

Where 𝜙 = 𝑐1 + 𝑐2 , 𝜙 > 4
ALGORITHM - CPSO
1. Input : Sensor data
2. Process :
3. Initialize inertia weight
wx, wn
4. Initialize maximum iteration
max _iter
5. Initialize population and velocity
x0, v
6. Update inertia weight
(wx−wn) ∗ ite
7. w = wx − max _𝑖𝑡𝑒𝑟
8. Update position
𝑉(𝑖,𝑗) = x × (𝑉𝑖,𝑗 + 𝑐1 ∗ 𝑟𝑎𝑛𝑑() ∗ (𝑝𝑏𝑒𝑠𝑡(𝑖,𝑗) − 𝑋(𝑖,𝑗) ) …
+𝑐2 ∗ 𝑟𝑎𝑛𝑑() ∗ (𝑔𝑏𝑒𝑠𝑡(1,𝑗) − 𝑋(𝑖,𝑗) )
2
x=
|(2−𝑝𝑖−𝑝𝑖2)|

𝑝𝑖 = 𝑐1 + 𝑐2 𝑝𝑖 > 4

𝑝𝑖2 = √𝑝𝑖 ∗ 𝑝𝑖 − 4 ∗ 𝑝𝑖
9. Evaluate fitness value
10. Update pbest and fitness
11. Update gbest
12. Repeat steps 3 to 8 until global best solution is found
13. Output : Best solution found

Fig (1) CPSO Graph

C. Artificial Bee Colony Optimization

The most recent based algorithms inspired by the intelligence etiquette of the honey bees is the Artificial
Bee Colony (ABC). There are three main aspects of ABC. Firstly Food source, these are possible solutions
to the desired problem, in our case, the deployment plan. A "profit" obtained from the fitness feature is
correlated with of food source. Secondly Bees employed, each bee employed is coupled with the source
of food. The Employed bees are required to explore the neighbor arbitrary food source of the present
exploiting bee in each iteration and return with the profitability details. Finally Unemployed Bees, Two
kinds of unemployed bees live, such as Onlookers and Scouts. According to the profitability details of
working bees, onlookers identified an established food supply. A more advantageous food sources will be
explored with the greater probability with the selection of roulette wheels [11]. On the other side, if its
associated food supply is "abandoned," an employed bees turn into a scout.
An optimization technique that intimates the searching etiquette of honey bees is the Artificial Bee Colony
(ABC) algorithm. Via social cooperation, a set of honey bees, called a swarm, can successfully carry out
tasks. There are three categories of bees in the ABC algorithm: Bees working, Onlooker bees, and Scout
bees. In their memory, the working bees hunt for food around the food source; meanwhile, they share the
information to the onlooker bees from these food sources. From those food sources found by the working
bees, Onlooker bees prefer to choose some good food sources.
The food supply that has larger quality (fitness) would have a greater opportunity for the onlooker bees to
be selected than the one of lower quality. The Scout Bees are transformed from set of bees working,
leaving their food sources and looking for the new ones. The first half of the swarm consists of the working
bees in the ABC algorithm, and the second half is the onlooker bees. The number of working bees or the
onlooker bees in the swarm is proportional to the number of solutions.

ALGORITHM - ABC
1. Initialize an arbitrary-related bee Population Places in the Solution Space.
2. Loop Commences
3. Shift the employed bees hired into their arbitrary food sources nearby the location
associated and assess the location Gain.
4. Shift the onlookers to their arbitrary sources of food near the similar one and evaluate the
gain.
5. When the food source is abandoned, shift the scouts for checking out new sources.
6. Update the best solution so far.
7. Exit loop if a condition (an adequately best fitness or a maximum number of iterations)
is satisfied.
8. Loop Stops

Fig (4) ABC Graph

CONCLUSION

A large-scale sensor was rigorously formulated, deployment issue in fixing hardware failures with the
target of minimal maintenance cost. The proposed quantitative maintenance cost models involve the
exponential temperature component, all contemplating electronics loss and battery exhaustion. In the
meantime, we combined the sensing efficiency metric with the statistical investigation of pre-
deployment data to determine the spatial informatively of the deployment.
We used two meta-heuristic algorithms to search effectively in the high-dimensional solution space
using our own fitness function, while the formulated optimization problem was non-convex and
nonlinear. The consistency of the meta-heuristics was contrasted with the present greedy heuristics.
The study shows that, compared to other state-of-the-art, the proposed CPSO (Canonical Particle Swarm
Optimization) and ABC (Artificial Bee Colony) substantially improve sensor deployment and save
maintenance costs to some extent. To have a greater impact we need to implement better strategy to
further improve sensor deployment and to further reduce maintenance cost.

REFERENCES
1. U. Noreen, A. Bounceur, and L. Clavier, (2017) “A study of LoRa low power and wide area
network technology,” in Proc. IEEE Int. Conf. Adv. Technol. Signal Image Process. (ATSIP),
Fez, Morocco, pp. 1–6.
2. J. C. Liando, A. Gamage, A. W. Tengourtius, and M. Li, (2019) “Known and unknown facts
of LoRa: Experiences from a large-scale measurement study,” ACM Trans. Sens. Netw., vol. 15,
no. 2, pp. 1–35.
3. F. Alduraibi, N. Lasla, and M. Younis, (2016) “Coverage-based node placement optimization
in wireless sensor network with linear topology,” in Proc. IEEE Int. Conf. Commun. (ICC),
Kuala Lumpur, Malaysia, pp. 1–6.
4. S. Sengupta, S. Das, M. Nasir, and B. K. Panigrahi, (2013) “Multi-objective node deployment
in WSNs: In search of an optimal trade-off among coverage, lifetime, energy consumption, and
connectivity,” Eng. Appl. Artif. Intell., vol. 26, no. 1, pp. 405–416.
5. A. Krause, A. Singh, and C. Guestrin, (2008) “Near-optimal sensor placements in Gaussian
processes: Theory, efficient algorithms and empirical studies,” J. Mach. Learn. Res., vol. 9,
pp. 235–284, Jun.
6. A. Boubrima, W. Bechkit, and H. Rivano, (2019) “On the optimization of WSN deployment for
sensing physical phenomena: Applications to urban air pollution monitoring,” in Mission-
Oriented Sensor Networks and Systems: Art and Science. Cham, Switzerland: Springer, pp. 99–
145.
7. M. Z. A. Bhuiyan, G. Wang, J. Cao, and J. Wu, (2014) “Sensor placement with multiple
objectives for structural health monitoring,” ACM Trans. Sens. Netw., vol. 10, no. 4, pp. 1–45.
8. X. Wu, M. Liu, and Y. Wu, (2012) “In-situ soil moisture sensing: Optimal sensor placement and
field estimation,” ACM Trans. Sens. Netw., vol. 8, no. 4, pp. 1–30.
9. W. Du, Z. Xing, M. Li, B. He, L. H. C. Chua, and H. Miao, (2015) “Sensor placement and
measurement of wind for water quality studies in urban reservoirs,” ACM Trans. Sens. Netw.,
vol. 11, no. 3, pp. 1–27.
10. Mohammed Zaki Hasan and Hussain Al-Rizzo, (2019) “Optimization of sensor deployment for
IIoT using a MultiSwarm algorithm” IEEE Internet of things journal, vol. 6, no. 6 Dec.
11. A. Lipowski and D. Lipowska, (2012) “Roulette-wheel selection via stochastic acceptance,”
Physica A, Stat. Mech. Appl., vol. 391, no. 6, pp. 2193–2196.