Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Engineering Applications of Artificial Intelligence 112 (2022) 104789

Contents lists available at ScienceDirect

Engineering Applications of Artificial Intelligence


journal homepage: www.elsevier.com/locate/engappai

An immune plasma algorithm with a modified treatment schema for UCAV


path planning
Selcuk Aslan
Department of Aeronautical Engineering, Erciyes University, Kayseri, Turkey

ARTICLE INFO ABSTRACT


Keywords: The success of a task carried out by an unmanned combat aerial vehicle for short UCAV is directly related with
IP algorithm the path in the battlefield or operation area. The path of the UCAV should ensure that the probability of being
Plasma extraction shot down by the enemy weapon systems and the consumption of fuel are optimized. Immune Plasma algorithm
Centrifuge
(IP algorithm or IPA) is one of the most recent optimization techniques. In this study, the path planning of
UCAV
UCAV was made based on a modified IP algorithm called as centrifuge IPA (centIPA). The performance of
Path planning
the centIPA was first investigated in detail by assigning different values to the control parameters and using
various battlefield scenarios. The results obtained by the centIPA were also compared with the results of
other metaheuristics and their improved variants under the same conditions. The comparative studies between
centIPA and other metaheuristics showed that newly introduced plasma extraction technique significantly
contributes to the solving capabilities of IPA and centIPA outperforms its competitors for the vast majority of
the test scenarios.

1. Introduction 2019), Biogeography based Optimization (BBO) (Ma et al., 2017),


Differential Evolution (DE) (Storn and Price, 1997), Evolutionary Strat-
The important roles played by the unmanned aerial vehicles (UAVs) egy (ES) (Beyer, 2001), Genetic algorithm (GA) (Katoch et al., 2021),
and unmanned combat aerial vehicles (UCAVs) increased the studies Population based Incremental Learning (PBIL) (Baluja, 1994), Particle
about the command-control systems, useful load capabilities, air-to-air Swarm Optimization (PSO) (Sengupta et al., 2019), and Stud Genetic
or air-to-ground weaponization, flight route or path planning methods algorithm (SGA) (Khatib and Fleming, 1998). Moreover, Wang et al.
for these modern aerial equipments (Schwarzrock et al., 2018). Path (2012b) changed the information exchange mechanism when updating
planning is one of the most important open issues for the UAV/UCAV or the light intensity of the Firefly algorithm (FA) and investigated their
other autonomous vehicle researches and can be described as a search
FA based approach on searching UCAV paths. The studies carried out
of optimal or near optimal route between the initial flight location
under the leadership of Wang also extend to the UCAV path planning
and previously determined target or destination within an operation
in three-dimensional environment. A hybrid metaheuristic approach
field (Pérez-Carabaza et al., 2019; Li et al., 2020a).
that manages the selection of the individuals belonging to the Cuckoo
Even though the problem description of the path planning is rel-
Search (CS) algorithm with the DE was modeled by Wang et al. (2012d)
atively simple, it still requires a straightforward mathematical model
that defines the optimal path and some constraints about the bat- and tested for path planning of UCAV operating in three-dimensional
tlefield (Mac et al., 2016; Ropero et al., 2019). One of the most combat environment. In another study, Wang et al. (2016) presented
versatile mathematical model for the UCAV path planning has been improved BA (IBA) that support the BA by using evolutionary operators
introduced by Xu et al. (2010) and allowed using different metaheuris- of the DE algorithm. Experimental studies proved that IBA is better
tic algorithms for solving the mentioned problem. In addition to the than standard BA on finding more safe paths in three-dimensional
mathematical modeling researches, they illustrated how a metaheuris- environment (Wang et al., 2016).
tic can be employed for UCAV path planning by using Artificial Bee Li and Duan (2012) executed a modified Gravitational Search al-
Colony (ABC) algorithm (Xu et al., 2010). Zhang et al. (2011) also used gorithm (GSA) for the same problem. An improved PSO algorithm
an ABC algorithm based approach called Fitness-Scaling Chaotic ABC was designed by Zhang et al. (2013) and experimental studies showed
(FSCABC) for the task of UCAV path planning. Wang et al. (2012a) that the newly proposed technique converges more quickly to optimal
proposed a UCAV path optimizer with Bat algorithm (BA) powered by path or paths. Chaotic number generation was utilized from Zhu and
a mutation operator and compared their method to various metaheuris- Duan (2014) for further improving the performance of the BBO and
tics including Ant Colony Optimization (ACO) (Dorigo and Stutzle, Chaotic Predator–Prey BBO for short CPPBBO was developed as a

E-mail address: selcukaslan@erciyes.edu.tr.

https://doi.org/10.1016/j.engappai.2022.104789
Received 17 October 2021; Received in revised form 24 December 2021; Accepted 23 February 2022
Available online xxxx
0952-1976/© 2022 Elsevier Ltd. All rights reserved.
S. Aslan Engineering Applications of Artificial Intelligence 112 (2022) 104789

UCAV path planner. They analyzed the capabilities of the CPPBBO 2. Mathematical background of UCAV path planning
on a battlefield with eight enemy threats and a set of comparison
between CPPBBO and other BBO variants was carried out (Zhu and In the mathematical model introduced by Xu et al. (2010), a UCAV
Duan, 2014). Li et al. (2014) introduced a method called Balance- starts flight from point 𝑃𝑠 = (𝑥𝑠 , 𝑦𝑠 ) for destroying the target positioned
Evolution Strategy (BES) and integrated it into the workflow of the at point 𝑃𝑡 = (𝑥𝑡 , 𝑦𝑡 ) after passing some enemy radars, missiles, anti-
ABC algorithm. Three different battlefield scenarios were used in the air artilleries or other threatening areas of the task region. The enemy
experiments and it was concluded that the UCAV paths found by the radars, missiles or anti-air artilleries are all represented in the form of
new ABC implementation are generally better than the paths found by circles (Xu et al., 2010). When a UCAV is inside of a circle, it can be shot
the standard ABC algorithm. Tang and Zhou (2015) analyzed solving down by the corresponding enemy threat. While the threatening level
capabilities of the Glowworm Swarm Optimization (GSO) algorithm on or possibility of being shot down by the enemy defence system increases
the UCAV path planning. The performance of the Grey Wolf Optimizer with the smaller distance between the UCAV and threat center, the
(GWO) was investigated by Zhang et al. (2016) for different battlefield vulnerability of the UCAV decreases with the longer distance between
scenarios. Luo et al. (2017) tried to strengthen the BA with quantum the UCAV and threat center (Xu et al., 2010).
encoding and solved path planning problem. In addition to all these For determining a path from 𝑃𝑠 to 𝑃𝑡 , a straight line between
algorithms, Symbiotic Organisms Search (SOS) (Miao et al., 2019), these two points is first generated and then divided into (𝐷 + 1)
Flower Pollination (FP) (Chen et al., 2021) and modified variants of equal segments or portions by determining 𝐷 different segment points
the BBO (Wang et al., 2012c), CS (Pan et al., 2019), PSO (Wu et al., appropriately. At each segment point, a new line that is vertical to the
2020), DE (Pan et al., 2020), and GWO (Qu et al., 2020) were also used line between 𝑃𝑠 and 𝑃𝑡 is drawn and a set of lines showed as 𝐿 =
{𝐿1 , 𝐿2 , … , 𝐿𝐷 } is obtained (Xu et al., 2010). The set of lines showed as
for planning the optimal paths of the UCAVs.
𝐿 = {𝐿1 , 𝐿2 , … , 𝐿𝐷 } has an important role on determining the path of
The metaheuristics that are capable of being successfully used on
the considered UCAV and its target. When a discrete point at each line
solving path planning problems for aerial vehicles cannot be limited
in the 𝐿 is selected and then selected points are connected sequentially,
with the algorithms or their variants introduced above (Li et al.,
a path from 𝑃𝑠 to 𝑃𝑡 can be described as {𝑃𝑠 , 𝑃𝐿1 , 𝑃𝐿2 , … , 𝑃𝐿𝐷 , 𝑃𝑡 }. As
2021). The tremendous richness of the nature still gives inspiration to
easily seen from the definition of the path, 𝐷 different points, (𝑥1 , 𝑦1 )
the researchers from different disciplines and new algorithms such as
for 𝐿1 , (𝑥2 , 𝑦2 ) for 𝐿2 and so on, should be searched or determined.
Monarch Butterfly Optimization (MBO) (Feng et al., 2021) modeling
In order to accelerate or simplify the search operations of the
the migration characteristics of the monarch butterflies, Moth Search required points, a coordinate transformation in which the original
(MS) algorithm (Wang, 2018) mimicking a complex feature of moths coordinate system is replaced with a new coordinate system can be
called phototaxis, Elephant Herding Optimization (EHO) (Li et al., utilized (Xu et al., 2010). In the new coordinate system, the horizontal
2020b) inspired by the herding habits of elephant groups, Krill Herd axis corresponds to the transformed straight line between 𝑃𝑠 and 𝑃𝑡 . For
(KH) (Wang et al., 2019) based on foraging motion of krill individuals the transformation of a point (𝑥𝑘 , 𝑦𝑘 ) in the original coordinate system
have been introduced. In addition to this richness of nature, the exis- into the suitable point (𝑥′𝑘 , 𝑦′𝑘 ) in the new coordinate system, the Eq. (1)
tence of the new coronavirus namely COVID-19 has gathered attentions where 𝜃 is the rotation angle is used Xu et al. (2010). When the required
to the details of new or conventional medical methods, diagnosis or transformation is completed, the point (𝑥1 , 𝑦1 ) of 𝐿1 is represented as
treatment techniques. They have been investigated in detail from the (0, 𝑦′1 ), the point (𝑥2 , 𝑦2 ) of 𝐿2 is represented as (0, 𝑦′2 ) and so on and
perspective of researchers in various disciplines including computer the whole path is simply described with a 𝐷-dimensional vector of
and information sciences when new metaheuristics are tried to be {𝑦′1 , 𝑦′2 , … , 𝑦′𝐷 } (Xu et al., 2010). In Fig. 1, generation of vertical lines,
designed. One of the most recent study that originates directly from transformation of them and subsequent operations are visualized. While
a medical treatment method was carried out by Aslan and Demirci 𝑥 and 𝑦 are used to represent the horizontal and vertical axes of the
(2020) and then Immune Plasma algorithm (IP algorithm or IPA) was original coordinate system, the horizontal and vertical axes of the new
introduced to the literature of metaheuristics. coordinate system are matched with 𝑥′ and 𝑦′ , respectively. Moreover,
IP algorithm is the first metaheuristic trying to model fundamen- the lines ranging from 𝐿1 to 𝐿𝐷 of the original coordinate system
tal operations of a medical method called convalescent or immune correspond to the lines ranging from 𝐿′1 to 𝐿′𝐷 of the transformed
plasma treatment and comparative studies showed its promising per- system.
formance (Aslan and Demirci, 2020). IP algorithm assumes that the [ ′] [ ] [ ]
𝑥𝑘 𝑐𝑜𝑠(𝜃) 𝑠𝑖𝑛(𝜃) 𝑥𝑘 − 𝑥𝑠 𝑦 − 𝑦𝑠
plasma taken from a donor can be directly represented with the same = × 𝑎𝑛𝑑 𝜃 = 𝑎𝑟𝑐𝑡𝑎𝑛( 𝑡 ) (1)
𝑦′𝑘 −𝑠𝑖𝑛(𝜃) 𝑐𝑜𝑠(𝜃) 𝑦𝑘 − 𝑦𝑠 𝑥𝑡 − 𝑥𝑠
donor. Even though the idea lying behind this type of assumption brings
implementation simplicity, the plasma corresponds to the antibody rich The path determined for the UCAV should be evaluated by consid-
part of the blood taken from the donor and obtaining it requires an ering the cost of enemy threats in the task field and the cost of fuel as
extra operation named centrifuge. In this study, the plasma extraction in the Eq. (2) (Xu et al., 2010). In Eq. (2), 𝐽 shows the sum of weighted
cost of enemy weapons and weighted cost of fuel. Also, 𝜆 corresponds
schema of the standard IP algorithm was modified by considering the
to the weighting coefficient and it is chosen between 0 and 1. While 𝑤𝑡
centrifuge process and a new variant called centrifuge IPA (centIPA)
is used on behalf of the cost of weapons changing with 𝑙, 𝑤𝑓 is used on
was introduced for solving the UCAV path planning problem. The
behalf of the cost of fuel changing with 𝑙 (Xu et al., 2010).
capabilities of the centIPA were examined by assigning different values
𝑙𝑒𝑛𝑔𝑡ℎ 𝑙𝑒𝑛𝑔𝑡ℎ
to the control parameters and using various battlefield scenarios. The
𝐽 = 𝜆𝐽𝑡 + (1 − 𝜆)𝐽𝑓 = 𝜆 𝑤𝑡 𝑑𝑙 + (1 − 𝜆) 𝑤𝑓 𝑑𝑙 (2)
comparative studies between centIPA, IPA, other metaheuristics includ- ∫0 ∫0
ing ABC, ACO, BBO, DE, ES, FA, GS, FA, PBIL, PSO, SGA, and their The integral calculations can be made by using some approxi-
variants informed that centIPA is capable of obtaining more robust and mations with the acceptable accuracy. For this purpose, the cost of
safe paths for most of the test scenarios. weapons or threats is tried to be calculated for each segment generated
The rest of the paper is organized as follows: Mathematical model- by connecting two subsequent points in the path. Assume that 𝐿𝑖𝑗 is
ing of the path planning is presented in Section 2. Details of the IPA and the length of segment obtained by connecting point 𝑖 and point 𝑗 of
newly proposed plasma extraction schema are given in Sections 3–4. the determined path. Also, it is assumed that the mentioned segment is
The Section 5 is devoted to the results of the experiments and com- divided equally into ten sub-segments and first, third, fifth, seventh and
parative studies. Finally, conclusion and future works are mentioned ninth inner division points are named as 1∕10 point, 3∕10 point, 5∕10
in Section 6. point and so on. If the distance between the center of the threat and

2
S. Aslan Engineering Applications of Artificial Intelligence 112 (2022) 104789

Fig. 1. Generation of the vertical lines (a), transformation of the original coordinate system (b), and corresponding segmentation points and paths for the UCAV (c).

the segment of length 𝐿𝑖𝑗 is less than the threat radius, the cost of the 3.1. Creation of initial population
threats for the segment of length 𝐿𝑖𝑗 is calculated by using Eq. (3) (Xu
et al., 2010). In Eq. (3), 𝑡𝑘 shows the level or grade of the 𝑘th enemy IP algorithm represents an individual of the population with a 𝐷
threat and 𝑁𝑡 shows the number of enemy threats in the battlefield. dimensional vector in which 𝐷 shows the number of parameters of
4
Moreover, 𝑑0.1,𝑖,𝑘 shows the distance between 1∕10 point on the segment the considered optimization problem (Aslan and Demirci, 2020). When
and the center of the considered enemy threat or weapon (Xu et al., the 𝑘th individual showed with 𝑥𝑘 in the population of size 𝑃 𝑆 is
2010). initialized, IPA utilizes from Eq. (4) for assessing a value to the 𝑗th
element of the same individual (Aslan and Demirci, 2020). In Eq. (4),
𝑁𝑡
𝐿𝑖𝑗 ∑ 1 1 1 1 1 𝑥𝑘𝑗 represents the 𝑗th element of 𝑥𝑘 . Also, 𝑥𝑙𝑜𝑤 and 𝑥ℎ𝑖𝑔ℎ define the
𝐽𝑡,(𝑖𝑗) = 𝑡𝑘 ( + + + + ) (3) 𝑗 𝑗
5 4
𝑑0.1,𝑖,𝑘 4
𝑑0.3,𝑖,𝑘 4
𝑑0.5,𝑖,𝑘 4
𝑑0.7,𝑖,𝑘 4
𝑑0.9,𝑖,𝑘 lower and upper bounds of the 𝑥𝑘𝑗 . Finally, 𝑟𝑎𝑛𝑑(0, 1) is used on behalf
𝑘=1
of a random number between 0 and 1.
For the calculation of the integral about the cost of fuel, another
approximation is used and 𝑤𝑓 is set to 1 (Xu et al., 2010). By setting 𝑥𝑘𝑗 = 𝑥𝑚𝑖𝑛
𝑗 + 𝑟𝑎𝑛𝑑(0, 1)(𝑥𝑚𝑎𝑥
𝑗 − 𝑥𝑚𝑖𝑛
𝑗 ) (4)
the 𝑤𝑓 to 1 or another constant, the cost of fuel becomes directly
proportional to the length of the path and UCAV path planning can 3.2. Distribution of infection
be modeled as a 𝐷-dimensional optimization problem that requires
minimization of Eq. (2) (Xu et al., 2010). The distribution of infection between individuals is managed by
IPA with Eq. (5) given below (Aslan and Demirci, 2020). In Eq. (5),
𝑥𝑖𝑛𝑓
𝑘𝑗
shows the randomly chosen 𝑗th parameters of the 𝑥𝑖𝑛𝑓
𝑘
individual.
3. Immune plasma algorithm
The 𝑥𝑖𝑛𝑓
𝑘
individual actually corresponds to the infectious 𝑥𝑘 individual
and it differs from 𝑥𝑘 by only the 𝑗th parameter. While 𝑥𝑚𝑗 shows
The usage of blood from an individual who infected and recov- the 𝑗th parameters of the 𝑥𝑚 that is also determined randomly from
ered previously for the treatment of critical patients suffering from the population and it must be different than the 𝑥𝑘 , 𝑥𝑘𝑗 presents the
the same infection is a creative medical method of a hundred years 𝑗th parameter of the 𝑥𝑘 . Moreover, 𝑟𝑎𝑛𝑑(−1, 1) is used on behalf of a
old (Marson et al., 2020). This medical method is called immune or random number generated between −1 and 1.
convalescent plasma treatment and some diseases stemmed from the
𝑥𝑖𝑛𝑓 = 𝑥𝑘𝑗 + 𝑟𝑎𝑛𝑑(−1, 1)(𝑥𝑘𝑗 − 𝑥𝑚𝑗 ) (5)
SARS, MERS, Ebola, and COVID-19 are tried to be treated with the 𝑘𝑗
mentioned approach (Shen et al., 2020). Even though the idea lying When the 𝑥𝑘 individual is infected, its immune response or antibody
behind the immune or convalescent plasma treatment is relatively amount synthesized for the infection is measured for a minimization
simple, it has a strong and evident biological background. The main problem with the objective function 𝑓 as 𝑓 (𝑥𝑖𝑛𝑓 𝑘
). If the 𝑓 (𝑥𝑖𝑛𝑓
𝑘
) is less
purpose of the plasma treatment is actually transferring the antibodies than the immune response of 𝑥𝑘 immediately before the infection calcu-
that are specifically synthesized proteins by the immune system cells, lated as 𝑓 (𝑥𝑘 ), it is understood that the immune system of 𝑥𝑘 is strongly
found in the blood and employed to inactivate the viruses or pathogens enough to handle and give a solid response to the infection (Aslan
causing infection (Hung et al., 2011; Parkin and Cohen, 2001). and Demirci, 2020). For the possible subsequent encountering with the
The fundamental steps of the mentioned treatment approach also same or similar infection, the 𝑗th parameter of the 𝑥𝑘 is replaced by
became a source of inspiration for Aslan and Demirci (2020) and a using the corresponding parameter of the 𝑥𝑖𝑛𝑓 𝑘
as in Eq. (6) (Aslan and
new metaheuristics known as IP algorithm or IPA has been intro- Demirci, 2020). Otherwise, the 𝑥𝑘 individual remains unchanged.
{ }
duced recently. IP algorithm matches the possible solutions of the 𝑥𝑖𝑛𝑓 , 𝑖𝑓 𝑓 (𝑥𝑖𝑛𝑓 ) < 𝑓 (𝑥𝑘 )
𝑥𝑘𝑗 = 𝑘𝑗 𝑘 (6)
optimization problem being considered with the individuals of the
𝑥𝑘𝑗 , 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
population (Aslan and Demirci, 2021). Moreover, the quality of an
individual is measured in terms of the objective function related with 3.3. Plasma transfer for receivers
the same problem. While some of the qualified individuals are selected
as blood or plasma donors, some of the individuals with the poor IPA continues its operations by determining how many individuals
qualities become receivers and they are tried to be treated by applying will be evaluated as critical and become receivers and how many
the plasma treatment (Aslan and Demirci, 2021). Until reaching the individuals will be evaluated as recovered and become donors after dis-
previously determined evaluation number, IPA solves the optimization tributing the infection between each individual of the population (Aslan
problem by infecting each individual, selecting donors and receivers, and Demirci, 2020). For determining how many individuals will be
applying plasma treatment for the receivers and finally modifying receivers and how many individuals will be donors, IPA utilizes from
antibody levels of the plasma donors with a quasi-deterministic schema. two control parameters named as 𝑁𝑜𝐷 and 𝑁𝑜𝑅 (Aslan and Demirci,
In the subsequent parts, the details of the IPA were given. 2020). While the best 𝑁𝑜𝐷 individuals from the population are chosen

3
S. Aslan Engineering Applications of Artificial Intelligence 112 (2022) 104789

as donors, the worst 𝑁𝑜𝑅 individuals from the population are chosen IP algorithm. Assume that 𝑥𝑏𝑒𝑠𝑡 is the best solution found by the IPA
as critical patients and become receivers (Aslan and Demirci, 2020). until the end of the current evaluation and 𝑥𝑑𝑛𝑟 is the best solution or
If 𝑥𝑟𝑐𝑣
𝑘
is the 𝑘th receiver in the set of receivers of size 𝑁𝑜𝑅 and donor individual chosen from the existing population. For extracting
𝑑𝑛𝑟
𝑥𝑚 is the randomly selected donor in the set of donors of size 𝑁𝑜𝐷, the plasma from 𝑥𝑑𝑛𝑟 donor of the population, the spin operations of
IP algorithm uses Eq. (7) for which 𝑗 ranges from 1 to 𝐷 (Aslan and the centrifuge device is modeled by using Eq. (9). In Eq. (9), 𝑥𝑝𝑙𝑠𝑚 𝑗
Demirci, 2020). In Eq. (7), 𝑥𝑟𝑐𝑣−𝑝 𝑘𝑗
shows the 𝑗th parameter of the 𝑥𝑘𝑟𝑐𝑣−𝑝 corresponds to the newly calculated 𝑗th parameter of the extracted
𝑟𝑐𝑣−𝑝
and 𝑥𝑘 𝑟𝑐𝑣
corresponds to the 𝑥𝑘 after plasma transfer is applied (Aslan plasma showed as 𝑥𝑝𝑙𝑠𝑚 . The initial values of the elements belonging
and Demirci, 2020). Also, 𝑥𝑟𝑐𝑣 𝑘𝑗
and 𝑥𝑑𝑛𝑟
𝑚𝑗 are matched with the 𝑗th
to 𝑥𝑝𝑙𝑠𝑚 are set to the values of the corresponding elements of 𝑥𝑑𝑛𝑟 .
parameters of the 𝑥𝑟𝑐𝑣 and 𝑥 𝑑𝑛𝑟 , respectively (Aslan and Demirci, 2020). Similarly, 𝑥𝑏𝑒𝑠𝑡
𝑗 and 𝑥𝑑𝑛𝑟
𝑗 are matched with the 𝑗th parameters of the
𝑘 𝑚
If the 𝑓 (𝑥𝑟𝑐𝑣−𝑝
𝑘
) is less than the 𝑓 (𝑥𝑑𝑛𝑟 𝑚 ), it is understood that plasma
𝑥 𝑏𝑒𝑠𝑡 𝑑𝑛𝑟
and 𝑥 , respectively.
treatment is useful for 𝑥𝑟𝑐𝑣 and 𝑥𝑟𝑐𝑣 is replaced with the 𝑥𝑟𝑐𝑣−𝑝 and then
𝑘 𝑘 𝑘 𝑥𝑝𝑙𝑠𝑚
𝑗 = 𝑥𝑏𝑒𝑠𝑡
𝑗 + 𝑟𝑎𝑛𝑑(−1, 1)(𝑥𝑑𝑛𝑟
𝑗 ) (9)
second dose of plasma is prepared. Otherwise, the plasma treatment
is completed for the 𝑥𝑟𝑐𝑣 𝑘
after changing 𝑥𝑟𝑐𝑣 𝑘
with the 𝑥𝑑𝑛𝑟
𝑚 in order to The plasma extraction from the 𝑥𝑑𝑛𝑟
will be completed after 𝐷 spins
guarantee that one dose of plasma is transferred. of the tube filled with the blood of 𝑥𝑑𝑛𝑟 . Each spin also indexed with
𝑗 tries to change the 𝑗th element of 𝑥𝑝𝑙𝑠𝑚 for further improving it by
𝑥𝑟𝑐𝑣−𝑝
𝑘𝑗
= 𝑥𝑟𝑐𝑣 𝑟𝑐𝑣 𝑑𝑛𝑟
𝑘𝑗 + 𝑟𝑎𝑛𝑑(−1, 1)(𝑥𝑘𝑗 − 𝑥𝑚𝑗 ) (7)
executing Eq. (9). If the 𝑗th spin of the tube improves the quality
When the IP algorithm decides to continue plasma treatment for 𝑥𝑟𝑐𝑣 of 𝑥𝑝𝑙𝑠𝑚 , in other words newly calculated 𝑓 (𝑥𝑝𝑙𝑠𝑚 ) is better than the
𝑘
with the second dose, Eq. (7) is used once more again. If the 𝑓 (𝑥𝑟𝑐𝑣−𝑝 𝑘
) previously calculated quality of 𝑥𝑝𝑙𝑠𝑚 , the 𝑗th parameter is updated with
calculated after transferring the second dose of plasma is less than the the new value and subsequent spin is started. The detailed operations
𝑓 (𝑥𝑟𝑐𝑣
𝑘
), 𝑥𝑟𝑐𝑣
𝑘
is replaced with the 𝑥𝑟𝑐𝑣−𝑝
𝑘
and then third dose of plasma is carried out for obtaining the 𝑥𝑝𝑙𝑠𝑚 and usage of it for the treatment of
prepared (Aslan and Demirci, 2020). Otherwise, the plasma treatment the most critical individual or the worst solution of the population are
is completed for the 𝑥𝑟𝑐𝑣 𝑘
. The usage of the second dose of plasma is given in Algorithm 1.
slightly different than the usage of the first dose of plasma. It should be
noticed that while the efficiency of the first dose of plasma is evaluated Algorithm 1 Obtaining plasma with centrifuge and its transfer
by comparing the 𝑓 (𝑥𝑟𝑐𝑣−𝑝𝑘
) with 𝑓 (𝑥𝑑𝑛𝑟
𝑚 ), the efficiency of the second or 1: //Extracting plasma with centrifuge by using selected donor
subsequent dose of plasma is evaluated by comparing the 𝑓 (𝑥𝑟𝑐𝑣−𝑝 𝑘
) with 2: 𝑥𝑝𝑙𝑠𝑚 ← 𝑥𝑑𝑛𝑟 and 𝑜𝑏𝑗𝑃 𝑙𝑠𝑚 ← 𝑓 (𝑥𝑑𝑛𝑟 )
𝑓 (𝑥𝑟𝑐𝑣 ) (Aslan and Demirci, 2020). 3: for 𝑗 ← 1 … 𝐷 do
𝑘 4: if 𝑡𝑐 < 𝑡𝑚𝑎𝑥 then
The completion of plasma treatment for each receiver carries the 5: 𝑡𝑐 ← 𝑡𝑐𝑟 + 1
𝑝𝑙𝑠𝑚
IPA into another stage in which the antibody levels of donor indi- 6: 𝑡𝑒𝑚𝑝 ← 𝑥𝑗
𝑝𝑙𝑠𝑚
viduals are configured by considering the ratio between 𝑡𝑐 and 𝑡𝑚𝑎𝑥 . 7: 𝑥𝑗 ← calculate the new 𝑗𝑡ℎ parameter of 𝑥𝑝𝑙𝑠𝑚 using Eq. (9)
The 𝑡𝑐 shows the current evaluation number. Also, the 𝑡𝑚𝑎𝑥 shows 8: if 𝑓 (𝑥𝑝𝑙𝑠𝑚 ) < 𝑜𝑏𝑗𝑃 𝑙𝑠𝑚 then
9: 𝑜𝑏𝑗𝑃 𝑙𝑠𝑚 ← 𝑓 (𝑥𝑝𝑙𝑠𝑚 )
the maximum evaluation number and IP algorithm terminates when
10: if 𝑜𝑏𝑗𝑃 𝑙𝑠𝑚 < 𝑓 (𝑥𝑏𝑒𝑠𝑡 ) then
𝑡𝑐 that is incremented one at each calculation of objective function 11: Update 𝑥𝑏𝑒𝑠𝑡 with 𝑥𝑝𝑙𝑠𝑚
reaches to 𝑡𝑚𝑎𝑥 . For each individual in the set of donors, IPA compares 12: end if
a random number generated between 0 and 1 with the 𝑡𝑐 ∕𝑡𝑚𝑎𝑥 . If the 13: else
𝑝𝑙𝑠𝑚
14: 𝑥𝑗 ← 𝑡𝑒𝑚𝑝
random number is higher than the 𝑡𝑐 ∕𝑡𝑚𝑎𝑥 for the 𝑥𝑑𝑛𝑟 𝑚 donor, 𝑥𝑚 is
𝑑𝑛𝑟
15: end if
re-initialized by using Eq. (4). Otherwise, 𝑥𝑑𝑛𝑟 𝑚 is changed by executing 16: end if
Eq. (8) in which 𝑥𝑑𝑛𝑟 𝑚𝑗 shows the 𝑗th parameter of 𝑥𝑚
𝑑𝑛𝑟 and 𝑗 index 17: end for
is selected sequentially from the set {1, 2, … , 𝐷}. When the value of 18: //Plasma transfer for the most critical individual
19: 𝑑𝑜𝑠𝑒𝐶𝑜𝑛𝑡𝑟𝑜𝑙 ← 1
𝑡𝑐 ∕𝑡𝑚𝑎𝑥 is investigated, it is easily seen that 𝑡𝑐 ∕𝑡𝑚𝑎𝑥 reaches to 1 and 20: 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝐶𝑜𝑛𝑡𝑟𝑜𝑙 ← 1
becomes equal to 1 at the end of execution. Because of this main 21: 𝑥𝑟𝑐𝑣 ← get the most critical individual
reason, the possibility of generating random number less than 𝑡𝑐 ∕𝑡𝑚𝑎𝑥 22: while 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝐶𝑜𝑛𝑡𝑟𝑜𝑙 == 1 do
23: if 𝑡𝑐𝑟 < 𝑡𝑚𝑎𝑥 then
and modifying donors with the Eq. (8) increases intrinsically. 24: 𝑡𝑐𝑟 ← 𝑡𝑐𝑟 + 1
25: 𝑥𝑟𝑐𝑣−𝑝 ← plasma treatment to 𝑥𝑟𝑐𝑣 with 𝑥𝑝𝑙𝑠𝑚 using Eq. (7)
𝑥𝑑𝑛𝑟 𝑑𝑛𝑟 𝑑𝑛𝑟
𝑚𝑗 = 𝑥𝑚𝑗 + 𝑟𝑎𝑛𝑑(−1, 1)𝑥𝑚𝑗 (8) 26: if 𝑑𝑜𝑠𝑒𝐶𝑜𝑛𝑡𝑟𝑜𝑙 == 1 then
27: if 𝑓 (𝑥𝑟𝑐𝑣−𝑝 ) < 𝑜𝑏𝑗𝑃 𝑙𝑠𝑚 then
28: Update 𝑥𝑟𝑐𝑣 with 𝑥𝑟𝑐𝑣−𝑝
4. Centrifuge based plasma extraction for IPA
29: 𝑑𝑜𝑠𝑒𝐶𝑜𝑛𝑡𝑟𝑜𝑙 ← 𝑑𝑜𝑠𝑒𝐶𝑜𝑛𝑡𝑟𝑜𝑙 + 1
30: else
As stated earlier, plasma can be described as the antibody rich part 31: Update 𝑥𝑟𝑐𝑣𝑘
with 𝑥𝑝𝑙𝑠𝑚
of the blood taken from the donor. However, extracting it requires a 32: Set 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝐶𝑜𝑛𝑡𝑟𝑜𝑙 to 0 and complete transfer
33: end if
set of operations that eliminate red blood cells, white blood cells and 34: else
platelets (Marson et al., 2020). For separating the plasma from the 35: if 𝑓 (𝑥𝑟𝑐𝑣−𝑝 ) < 𝑓 (𝑥𝑟𝑐𝑣 ) then
cells and platelets, the blood in a tube is processed by using a device 36: Update 𝑥𝑟𝑐𝑣 with 𝑥𝑟𝑐𝑣−𝑝
37: else
called centrifuge (Schenkel et al., 2013). The main motivation of the 38: Set 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝐶𝑜𝑛𝑡𝑟𝑜𝑙 to 0 and complete transfer
centrifuge is using centrifugal force that is achieved by spinning the 39: end if
tube filled with the blood and other liquid at high speed. When the 40: end if
41: if 𝑓 (𝑥𝑟𝑐𝑣 ) < 𝑓 (𝑥𝑏𝑒𝑠𝑡 ) then
blood cells fall to the bottom of the tube, plasma is then drawn of and 42: Update 𝑥𝑏𝑒𝑠𝑡 with 𝑥𝑟𝑐𝑣
collected separately (Schenkel et al., 2013). 43: end if
In the standard IP algorithm, the plasma is directly matched with 44: end if
45: end while
the selected donor and represented with a 𝐷-dimensional vector for
which its elements are equal to the elements of same donor. Even The workflow of the standard IPA can be supported with the cen-
though the representation of plasma with the corresponding donor trifuge based approach as in Algorithm 1. Moreover, newly proposed
simplifies the algorithm, it should be noted that getting plasma from plasma extraction approach requires only one donor individual and
blood requires further processing with a special device such as cen- obtained plasma will be used only for the most critical individual of
trifuge and an appropriate mechanism that mimics the working routine the population. By using this kind of mechanism, there is no need for
of the centrifuge device can be determined and integrated into the adjusting the 𝑁𝑜𝐷 and 𝑁𝑜𝑅 parameters separately and the chance

4
S. Aslan Engineering Applications of Artificial Intelligence 112 (2022) 104789

Table 1
Properties of the battlefields in the experiments.
Sc. Threat centers Threat radius Threat grade Start-Target point
(52,52),(32,40),(12,48),(36,26), 10,10,8,12, 2,10,1,2,
1 (11,11)–(75,75)
(80,60),(63,56),(50,42),(30,70) 9,7,10,10 3,5,2,4
(0,200),(200,0),(50,50),(95,95), 90,90,20,20, 7,7,5,5
2 (0,0)–(200,200)
(150,150),(95,50),(50,95),(140,105),(105,140) 20,20,20,20,20 5,6,5,6,5
(30,20),(50,15),(65,55),(0,24), 10,21,10,17, 5,5,5,5,
3 (0,0)–(100,100)
(50,80),(75,90),(100,70),(50,36) 27,10,20,11 5,5,5,5
(10,50),(20,20),(30,42),(30,80),(50,55), 10,9,8,10,10, 8,6,5,4,7,
4 (0,0)–(80,100)
(60,10),(60,80),(65,38),(75,65),(90,80) 10,10,12,8,10 6,7,6,8,10
5 (45,50),(12,40),(32,68),(36,26),(58,80) 10,10,8,12,9 2,10,1,2,3 (10,10)–(55,100)

of more effective plasma treatment is increased intrinsically. The IP Table 2


Results of centIPA with different 𝑃 𝑆 values for the first four scenarios.
algorithm that is modified with the proposed extraction technique will
be called as the centrifuge IPA or shortly centIPA in the subsequent Sc. 𝐷 𝑃𝑆

parts of this study. 20 30 40 50 75 100


Mean 48.4843 48.5362 48.5658 48.5165 48.6802 48.7913
5. Experimental studies 30 Best 48.1563 48.1555 48.1761 48.1554 48.1647 48.1839
Std. 0.2885 0.3620 0.3440 0.2707 0.3550 0.4494
1
The solving capabilities and effectiveness of the centIPA were tried Mean 51.9460 52.2912 50.1907 52.0747 52.4168 52.3589
50 Best 48.4093 48.3874 48.6319 48.6008 48.7798 48.7758
to be verified by using five different battlefields whose details are
Std. 7.3009 7.1014 2.6105 5.9654 5.9458 6.1131
summarized in Table 1. For the first three battlefield scenarios, the
Mean 150.1441 150.2401 150.1948 150.2774 150.7363 150.7408
number of division or segmentation points that is also matched with the 30 Best 149.6171 149.5441 149.6298 149.5603 149.6013 149.5562
number of parameters in the path planning problem and abbreviated as Std. 0.9446 1.0468 0.9737 1.0573 1.3520 1.3124
2
𝐷 was set to 30 and 50, respectively. For the fourth battlefield scenario, Mean 154.7972 155.0177 156.2449 156.1538 156.7025 155.9527
the number of segmentation points was set to 20 and 30. While the 50 Best 149.1759 149.2572 149.2421 149.2754 149.7532 149.5368
maximum evaluation number was equal to 30,000 for the first scenario, Std. 2.5747 2.3918 2.4402 2.5246 2.5619 2.6813
it was taken equal to 60,000 for the second and third scenarios and Mean 71.2516 71.2558 71.3520 71.3752 71.5070 71.6377
6000 for the fourth scenario. In order to evaluate how the qualities 30 Best 71.1750 71.1736 71.1762 71.1788 71.1854 71.2509
of the solutions obtained by the centIPA change with the population Std. 0.0757 0.0660 0.1640 0.1652 0.2499 0.3315
3
size, six different values such as 20, 30, 40, 50, 75, and 100 were used. Mean 72.7862 73.2893 74.5016 75.9453 77.1290 76.3917
50 Best 71.3695 71.3760 71.3664 71.4344 71.9730 71.7052
Each independent combination about the first four battlefield scenarios,
Std. 1.6377 1.7928 3.1869 3.5044 3.9636 4.3044
number of segmentation points and population size was tested 50 times
Mean 66.4339 66.4268 66.6530 66.6905 66.7785 66.4577
with random seeds by centIPA. The best solution found at each run or
20 Best 66.2605 66.3191 66.2731 66.4106 66.3810 66.3527
test was recorded with the corresponding objective function value. The Std. 0.1163 0.2275 0.3200 0.2798 0.3143 0.1187
4
best, mean best objective function values and standard deviations of 50
Mean 66.8121 67.1776 67.5618 67.3966 68.7725 68.4447
runs were presented in Table 2. 30 Best 66.4125 66.3715 66.9082 66.5817 66.9276 67.2245
The results given in Table 2 provide considerable information about Std. 0.2209 0.6156 0.5629 0.5746 3.4950 1.2064
the appropriate population size of the centIPA. For most of the test
cases, centIPA with population size taken equal to 20 obtains more
qualified UCAV paths compared to the paths of centIPA with population
that they are subtly adjusted by considering the enemy threats and
size taken equal to 30, 40, 50, 75, or 100. Only for the first scenario
total fuel consumption by the centIPA. The segments of the best paths
with 50 parameters and fourth scenario with 20 parameters, the popu-
found by the centIPA for all of four battlefield scenarios are not inside
lation size of the centIPA should be set to 40 and 30. By considering
the shooting range of any enemy threat. However, it should be noticed
the relatively close results given in Table 2, it can be generalized
that the segments that are not inside the shooting range of any enemy
that the population size can be determined between 20 and 40 for
the considered battlefields and termination criteria. The idea lying threat are not enough to decide the overall feasibility of the discovered
behind the proposed plasma extraction schema requires an intense path. While the segments are tried to be determined by considering the
search within the vicinity of the best solution found until the current enemy threats, the lengths of these segments or the fuel consumption
evaluation by utilizing from the selected donor. If the population size should also be minimized for an appropriate or feasible path. The
is chosen relatively small, centIPA extracts more plasma compared Figures 2–3 also allow to state that some segments of the best paths
to the running configurations with higher population sizes and treats are determined by the centIPA as tangents to the most outer circle
critical individuals quickly. Moreover, the increased intensity of the representing the maximum shooting range of the considered enemy
treated individuals in the population can help improving other critical threat for decreasing the fuel consumption of the UCAV.
individuals. The validation of the results found by the centIPA can be made
The UCAV path planning problem has a special attribute that in- after a comparison with the results of other metaheuristics including
creases its difficulty when compared to other real world or complex standard IPA. For this purpose, centIPA was first compared with the
engineering problems. Even though the quality or feasibility of a path IPA, ABC, and some ABC variants such as improved ABC (I-ABC),
is understood by considering the calculated objective function value, internal feedback ABC (IF-ABC), and balance-evolution ABC (BE-ABC)
a visual inspection should be made for the same path in order to algorithms over the first three scenarios. In order to guarantee that the
understand that whether a segment or part of the path is inside the comparison between centIPA, IPA, ABC, I-ABC, IF-ABC, and BE-ABC
shooting range of any enemy threat or not. In Figs. 2–3, the best and was carried out under the same conditions, the maximum evaluation
worst paths found by the centIPA with 𝑃 𝑆 equal to 20, 50, and 100 number was 30,000 for the first scenario and 60,000 for the second
for the first four battlefield scenarios are depicted. When the paths and third scenarios (Li et al., 2014). Moreover, the population size of
illustrated in the mentioned figures are controlled, it is clearly seen the centIPA and IPA was set to 30 and the number of food sources in

5
S. Aslan Engineering Applications of Artificial Intelligence 112 (2022) 104789

Fig. 2. The paths of centIPA by setting 𝑃 𝑆 to 20 (a), 50 (b), and 100 (c) for the Scenario-1 with 𝐷 equal to 30 and setting 𝑃 𝑆 to 20 (d), 50 (e), and 100 (f) for the Scenario-2
with 𝐷 equal to 30.

Table 3 and standard deviations obtained after 50 independent runs in Table 3


Comparison between centIPA and other techniques for the first three scenarios. are investigated, it is seen that centIPA is capable of outperforming
𝐷 centIPA IPA ABC I-ABC IF-ABC BE-ABC IPA, ABC, and other ABC variants for the first and second battlefield
Mean 48.5362 49.2894 52.9887 60.4591 51.9396 50.4567 scenarios. While the BE-ABC produces better solutions compared to the
30 Std. 0.3620 0.4657 1.4259 3.2883 1.3434 0.6726 other tested metaheuristics for the third scenario with 30 parameters,
Rank 1 2 5 6 4 3
IPA finds more qualified paths for the same battlefield scenario with 50
Mean 52.2912 59.3546 59.9722 80.1747 59.9569 54.9826
parameters. Even though the centIPA lags behind its competitors for the
50 Std. 7.1014 4.9141 2.9133 7.7575 2.2149 2.2310
Rank 1 3 5 6 4 2
third scenario, the difference between centIPA and winner algorithm is
relatively small and centIPA becomes the third best algorithm for the
Mean 150.2401 151.4998 154.9276 185.7927 153.5837 153.4103
30 Std. 1.0468 1.2856 2.5123 3.7547 0.4925 0.5549 mentioned battlefield scenario with 30 parameters and the second best
Rank 1 2 5 6 4 3 algorithm for the mentioned battlefield scenario with 50 parameters.
Mean 155.0177 157.3681 157.2044 164.9908 157.8147 153.5245 Another comparison has been carried out between centIPA and
50 Std. 2.3918 1.5963 3.6031 2.8118 1.9058 1.0479 other metaheuristics such as PSO, ABC, DE, and improved adaptive
Rank 1 4 3 6 5 2 DE variants such as JADE, CIJADE, and CIPDE (Pan et al., 2020). The
Mean 71.2558 71.2053 73.9346 74.8452 73.6928 70.0789 population size of PSO, DE, JADE, CIJADE, and CIPDE was taken equal
30 Std. 0.0660 0.0144 1.0134 0.8904 0.4083 0.4634 to 60. Also, the population size of IPA and centIPA and the number of
Rank 3 2 5 6 4 1 food sources in the ABC algorithm were set to 30. 𝑐1 and 𝑐2 coefficients
Mean 73.2893 72.5443 78.8720 81.8885 77.4828 75.8906 of PSO were equal to 2 and lower and upper bounds of inertia weights
50 Std. 1.7928 2.0778 3.1422 2.7718 2.0989 1.3950 were determined as 0.4 and 0.9 (Pan et al., 2020). While the 𝐹 and
Rank 2 1 5 6 4 3
𝑐𝑟 parameters of the DE were 0.5 and 0.1, the 𝑙𝑖𝑚𝑖𝑡 parameter of ABC
Average 1.5000 2.3333 4.6667 6.0000 4.1667 2.3333
was 100 (Pan et al., 2020). Finally, the 𝑁𝑜𝐷 and 𝑁𝑜𝑅 parameters of
Overall 1 2 5 6 4 2 the IPA were set to 1 (Aslan and Demirci, 2020). For guaranteeing that
the solutions of the algorithms are obtained under the same conditions,
the maximum evaluation number was 6000 for the fourth scenario (Pan
ABC, I-ABC, IF-ABC, and BE-ABC was taken equal to 30 (Li et al., 2014). et al., 2020). When the objective function values and standard devia-
tions obtained after 50 independent runs in Table 4 are investigated,
The 𝑙𝑖𝑚𝑖𝑡 was 30 for ABC and its tested variants (Li et al., 2014). Finally,
it is concluded that centIPA still protects its comparative performance
the 𝑁𝑜𝐷 and 𝑁𝑜𝑅 parameters of the IPA were determined as 1 (Aslan against tested metaheuristics. While the centIPA outperforms IPA, ABC,
and Demirci, 2020). When the mean best objective function values PSO, DE, and other DE based path planners for the fourth battlefield

6
S. Aslan Engineering Applications of Artificial Intelligence 112 (2022) 104789

Fig. 3. The paths of centIPA by setting 𝑃 𝑆 to 20 (a), 50 (b), and 100 (c) for the Scenario-3 with 𝐷 equal to 30 and setting 𝑃 𝑆 to 20 (d), 50 (e), and 100 (f) for the Scenario-4
with 𝐷 equal to 20.

Table 4
Comparison between centIPA and other techniques for the fourth scenario.
𝐷 centIPA IPA CIJADE PSO DE ABC JADE CIPDE
Best 66.3191 67.8385 66.2980 67.6630 66.7305 66.6316 66.3212 66.3310
Worst 67.0796 68.9444 66.6469 73.9605 72.5714 68.1843 67.0073 66.7515
20 Mean 66.4268 68.4550 66.4265 70.7947 67.3973 67.2604 66.5248 66.4696
Std. 0.2275 0.3919 0.0875 1.8699 1.2490 0.5129 0.1932 0.1267
Rank 2 7 1 8 6 5 4 3
Best 66.3715 74.1194 67.6224 73.7530 71.7664 67.8458 67.5219 68.1706
Worst 68.3358 79.8234 75.2568 85.5610 81.6450 76.7182 81.6311 78.2387
30 Mean 67.1776 75.6592 70.2976 78.9828 75.1973 72.6566 70.4078 71.0337
Std. 0.6156 1.4765 1.8130 3.8289 2.4347 2.1472 3.0261 2.5028
Rank 1 7 2 8 6 5 3 4
Average 1.5000 7.0000 1.5000 8.0000 6.0000 5.0000 3.5000 3.5000
Overall 1 7 1 8 6 5 3 3

with 20 parameters, it lags slightly behind the CIJADE and becomes the before the termination, it is accepted that algorithm is successful for
second best algorithm within other tested metaheuristics by considering this run. The ratio between the number of successful runs and total
the mean best objective function values. Even though the maximum number of runs is called success rate and it is abbreviated as 𝑆𝑟. When
evaluation number is chosen relatively small for the fourth battlefield an algorithm is successful for a run, the minimum number of evalua-
scenario compared to the previously used scenarios and centIPA termi- tions required to obtain a solution whose objective function value is
nates without operating changed treatment schema adequately, it still equal or less than a threshold is recorded. The average of recorded
protects promising performance against other path planners. values corresponds to the mean evaluations and it is abbreviated as
The newly proposed plasma extraction and treatment schema can 𝑀𝑒. The 𝑆𝑟 and 𝑀𝑒 metrics of the centIPA and IPA were calculated
change the convergence performance of the centIPA compared to the for the first four battlefield scenarios and summarized in Table 5. While
standard implementation of the same algorithm. For comparing the the threshold was 50 and 155 for the first and second scenarios, it was
convergence performances of the metaheuristics, success rate and mean determined as 75 for the third and fourth scenarios. The results given in
evaluations are two commonly used metrics (Gao et al., 2012; Feng Table 5 help to state that the convergence performance of the centIPA
and Wang, 2022). If an algorithm obtains a solution whose objective is better for the vast majority of the test cases compared to the IPA. The
function value is equal or less than the previously determined threshold centrifuge based plasma extraction model tries to support best solution

7
S. Aslan Engineering Applications of Artificial Intelligence 112 (2022) 104789

Table 5
𝑆𝑟 and 𝑀𝑒 values of centIPA and IPA for the first four scenarios.
Sc. 𝐷 centIPA IPA centIPA IPA
𝑃 𝑆 = 30 𝑃 𝑆 = 40
𝑆𝑟 𝑀𝑒 𝑆𝑟 𝑀𝑒 𝑆𝑟 𝑀𝑒 𝑆𝑟 𝑀𝑒
30 49.000 4660.837 48.000 22 362.542 50.000 6395.900 40.000 26 041.525
1
50 31.000 17 798.000 0.000 – 29.000 18 466.483 0.000 –
30 52.000 7166.940 50.000 24 079.920 50.000 8957.620 50.000 27 125.540
2
50 20.000 27 983.050 4.000 43 453.250 13.000 30 253.462 1.000 59 759.000
30 50.000 7872.080 50.000 9677.380 50.000 7061.620 50.000 11 919.420
3
50 41.000 37 153.195 43.000 27 025.465 31.000 32 709.258 39.000 37 153.692
20 53.000 303.480 50.000 2607.860 53.000 350.520 50.000 3234.280
4
30 50.000 727.400 16.000 5590.625 50.000 839.060 0.000 –
Sc. 𝐷 centIPA IPA centIPA IPA
𝑃 𝑆 = 50 𝑃 𝑆 = 100
𝑆𝑟 𝑀𝑒 𝑆𝑟 𝑀𝑒 𝑆𝑟 𝑀𝑒 𝑆𝑟 𝑀𝑒
30 50.000 6054.460 17.000 27 175.824 49.000 9202.959 0.000 –
1
50 23.000 16 605.826 0.000 – 16.000 21 190.688 0.000 –
30 54.000 9496.260 49.000 30 951.694 51.000 10 646.520 36.000 50 726.556
2
50 14.000 32 003.786 1.000 53 461.000 12.000 32 530.167 0.000 –
30 52.000 10 088.720 53.000 14 512.900 49.000 18 374.240 50.000 27 422.860
3
50 19.000 34 864.579 37.000 39 905.486 22.000 31 052.409 5.000 51 075.600
20 51.000 337.720 50.000 3409.560 50.000 265.920 18.000 4722.778
4
30 49.000 1020.540 0.000 – 52.000 1516.040 0.000 –

obtained by the algorithm until current evaluation with the donor. As Table 6
Average execution times of centIPA and IPA for the first four scenarios.
an expected result of this support, the chance of improving the best
solution is increased and convergence speed is accelerated. Sc. 𝐷 centIPA IPA centIPA IPA

Another important situation that should be considered when com- 𝑃 𝑆 = 30 𝑃 𝑆 = 40


paring a new variant of metaheuristic to its standard implementation is Mean Std. Mean Std. Mean Std. Mean Std.
related with the execution times of them. Even though the termination 30 1.513 0.014 1.505 0.022 1.512 0.015 1.497 0.016
1
criteria of both IPA and centIPA is same, plasma extraction mecha- 50 2.562 0.027 2.575 0.055 2.562 0.027 2.548 0.043
nism in centIPA changes the interactions between donor and receiver 30 3.362 0.065 3.313 0.037 3.409 0.073 3.313 0.069
2
individuals significantly compared to the standard IPA. In order to 50 5.712 0.045 5.592 0.071 5.739 0.051 5.521 0.047
understand that whether the centrifuge based plasma extraction brings 3
30 3.010 0.022 2.959 0.031 3.010 0.020 2.927 0.027
extra computational burden or not, the duration of each run in terms 50 5.107 0.082 4.961 0.071 5.118 0.061 4.948 0.067

of seconds is recorded when the IPA and centIPA are tested on the first 4
20 0.251 0.007 0.251 0.004 0.250 0.004 0.253 0.005
four battlefield scenarios. Both IPA and centIPA were coded in C pro- 30 0.387 0.007 0.380 0.008 0.388 0.008 0.383 0.009

gramming language and their runs were carried out on a machine under Sc. 𝐷 centIPA IPA centIPA IPA
Fedora 34 64-bit operating system with an Intel i5-10500 processor 𝑃 𝑆 = 50 𝑃 𝑆 = 100
and 16 GB of RAM.1 When the average execution times and standard Mean Std. Mean Std. Mean Std. Mean Std.
deviations calculated after 50 independent runs of IPA and centIPA in 30 1.533 0.016 1.514 0.019 1.534 0.027 1.517 0.019
1
Table 6 are investigated, it is seen that the difference between the 50 2.583 0.035 2.548 0.040 2.613 0.033 2.514 0.026
execution times of the algorithms is relatively small. The centrifuge 30 3.453 0.039 3.308 0.039 3.466 0.039 3.296 0.025
2
based plasma extraction depends on changing the selected donor with 50 5.822 0.090 5.533 0.080 5.851 0.074 5.585 0.054
the simple arithmetic operators by guiding the best solution found until 30 3.018 0.019 3.039 0.031 3.067 0.030 2.894 0.007
3
current treatment cycle. Because of the selection of donor and receiver 50 5.129 0.094 5.075 0.059 5.142 0.072 5.038 0.062
individuals and transferring plasma to receiver in the centIPA executes 20 0.254 0.006 0.254 0.009 0.251 0.006 0.258 0.006
4
similar dynamics with the standard IPA, the computational burden 30 0.391 0.008 0.386 0.011 0.394 0.009 0.389 0.008
originated from the centrifuge based method becomes negligible for
most of the test scenarios.
The comparative studies between centIPA and other metaheuristics was determined as 15, the population size of the remaining algorithms
were continued by using the fifth scenario. The metaheuristics that
was taken equal to 30 (Wang et al., 2012a,b; Tang and Zhou, 2015).
will be compared with the centIPA are IPA, ABC, BA (Wang et al.,
Each algorithm was tested 100 times for different values of 𝐷 by setting
2012a), ACO (Wang et al., 2012a), BBO (Wang et al., 2012a), DE (Wang
the maximum evaluation number to 6000 (Wang et al., 2012a,b; Tang
et al., 2012a), ES (Wang et al., 2012a), GA (Wang et al., 2012a),
PBIL (Wang et al., 2012a), PSO (Wang et al., 2012a), FA (Wang and Zhou, 2015). The best, worst, mean best objective function values
et al., 2012b), and some variants of them such as BAM (Wang et al., and standard deviations were recorded and summarized in Table 7.
2012a), MFA (Wang et al., 2012b), SGA (Wang et al., 2012a), and When the overall ranks given in Table 7 are controlled, it is decided
PGSO (Tang and Zhou, 2015). In order to decide that how the number that the centIPA with overall rank calculated as 1.625 outperforms
of segmentation points changes the performances of the algorithms, other tested algorithms for the fifth battlefield scenario. The mean best
eight different values including 5, 10, 15, 20, 25, 30, 35, and 40 were objective function values found by the centIPA for 𝐷 equal to 5, 10, 15,
assigned to 𝐷 of fifth scenario. While the number of food sources in ABC 30, and 40 are equal or better than the mean best objective function
values found by all of the other tested metaheuristics and its rank is
1
Source code is available upon request. determined as 1. Also, centIPA becomes the second best metaheuristic

8
S. Aslan Engineering Applications of Artificial Intelligence 112 (2022) 104789

Table 7
Comparison between centIPA and other techniques for the fifth scenario.
𝐷 centIPA IPA ABC BA BAM ACO BBO DE ES FA GA MFA PBIL PSO SGA PGSO
Best 50.384 50.384 50.384 60.690 54.357 61.372 60.330 54.357 59.590 54.359 55.247 54.357 59.763 55.167 55.654 53.380
Worst 50.385 50.385 50.385 345.255 60.240 63.320 171.500 62.200 112.260 65.740 61.600 62.419 72.250 66.071 61.200 60.630
5 Mean 50.384 50.384 50.384 106.483 59.054 61.520 72.730 58.596 80.720 58.750 60.470 59.167 66.139 59.906 60.501 53.669
Std. 0.001 0.001 0.001 – – – – 2.160 – 3.010 – 2.250 – 2.620 1.560 2.260
Rank 1 1 1 16 7 12 14 5 15 6 10 8 13 9 11 4
Best 50.369 50.376 50.370 52.360 51.395 60.228 52.947 51.395 57.420 51.399 51.607 51.397 83.112 52.207 51.549 50.649
Worst 50.384 50.457 50.406 108.738 60.7244 68.190 76.820 56.736 123.460 56.710 60.110 53.786 119.250 68.622 56.165 53.330
10 Mean 50.372 50.398 50.394 69.425 52.707 61.950 57.965 53.104 76.280 52.180 52.542 51.574 101.440 57.041 52.279 50.849
Std. 0.004 0.026 0.021 – – – – 2.600 – 2.370 – 1.730 – 2.250 1.430 1.870
Rank 1 3 2 14 9 13 12 10 15 6 8 5 16 11 7 4
Best 50.369 50.394 50.475 53.075 50.609 58.530 52.557 50.611 58.255 50.617 50.871 50.612 107.223 52.097 50.807 50.452
Worst 50.427 51.199 50.799 85.745 60.192 61.000 90.370 62.580 103.860 94.276 57.447 53.832 189.200 87.320 61.800 55.460
15 Mean 50.384 50.570 50.691 63.601 51.231 60.260 59.526 52.278 71.860 52.822 52.188 50.897 128.250 58.340 51.891 51.516
Std. 0.015 0.193 0.100 – – – – 3.730 – 4.250 – 1.340 – 4.010 2.450 1.490
Rank 1 2 3 14 5 13 12 9 15 10 8 4 16 11 7 6
Best 50.373 50.515 50.906 52.395 50.467 60.445 54.723 50.510 60.232 50.463 50.825 50.455 130.152 52.464 50.846 50.657
Worst 53.194 51.321 54.671 83.706 53.742 67.180 78.200 64.570 81.450 78.914 59.180 52.028 337.300 78.160 68.950 59.850
20 Mean 50.818 50.950 52.195 63.630 50.760 66.220 61.88 52.722 70.190 53.733 53.090 50.700 185.430 58.248 53.167 52.398
Std. 0.912 0.206 0.977 – – – – 3.710 – 7.580 – 1.020 – 6.950 3.990 1.560
Rank 3 4 5 13 2 14 12 7 15 10 8 1 16 11 9 6
Best 50.378 50.790 51.911 55.017 50.448 61.549 55.528 50.551 63.369 50.491 51.242 50.457 159.740 53.738 51.239 50.782
Worst 53.790 57.456 57.847 74.926 53.519 62.070 80.330 69.660 83.910 66.452 60.398 53.704 699.600 78.139 65.700 63.160
25 Mean 50.986 51.668 54.857 64.901 50.709 61.570 64.780 54.408 72.780 53.904 53.781 50.999 257.720 60.263 54.157 54.587
Std. 0.918 1.525 1.677 – – – – 4.120 – 8.660 – 0.810 – 7.550 4.060 2.380
Rank 2 4 10 14 1 12 13 8 15 6 5 3 16 11 7 9
Best 50.386 50.997 54.627 57.247 50.467 63.230 56.607 50.898 65.725 50.683 51.921 50.516 230.150 53.299 51.617 51.019
Worst 54.344 60.995 64.995 80.084 60.285 64.710 78.580 74.120 91.300 65.976 62.718 58.336 2396 93.695 64.710 75.320
30 Mean 51.092 51.799 59.866 66.616 51.106 63.950 67.870 59.988 74.780 54.962 55.008 51.357 395.540 62.385 54.521 56.891
Std. 0.935 2.305 2.346 – – – – 6.740 – 9.120 – 1.230 – 8.200 4.110 3.450
Rank 1 4 9 13 2 12 14 10 15 6 7 3 16 11 5 8
Best 50.438 51.455 57.579 57.448 50.479 66.960 63.021 52.537 66.745 51.083 52.311 50.471 270.330 55.503 51.633 54.136
Worst 55.300 97.051 74.505 82.737 58.819 68.720 93.850 84.440 88.76 83.887 74.479 55.883 6362 82.833 67.610 71.450
35 Mean 51.777 55.889 66.155 67.703 51.461 68.310 71.560 67.900 76.520 55.996 55.960 51.601 684.660 64.135 55.826 59.744
Std. 1.512 12.575 4.508 – – – – 9.150 – 9.550 – 1.650 – 8.650 4.120 4.010
Rank 3 5 10 11 1 13 14 12 15 7 6 2 16 9 4 8
Best 50.431 51.025 63.302 58.650 50.602 69.795 63.550 54.549 68.231 51.523 52.208 50.561 390.620 55.737 52.618 55.092
Worst 64.946 115.195 87.713 83.263 58.427 77.060 90.700 93.260 96.420 86.663 72.069 57.724 7103 84.730 67.870 72.650
40 Mean 51.857 55.984 75.495 69.973 51.876 74.580 74.850 77.620 80.260 57.856 57.493 52.198 1169 64.885 57.110 62.420
Std. 2.612 15.998 5.546 – – – – 10.900 – 10.430 – 2.380 – 9.410 4.550 4.540
Rank 1 4 13 10 2 11 12 14 15 7 6 3 16 9 5 8
Average 1.650 3.375 6.625 13.125 3.625 12.500 12.875 9.375 15.000 7.250 7.250 3.625 15.625 10.250 6.875 6.625
Overall 1 2 5 14 3 12 13 10 15 8 8 3 16 11 7 5

The results given in Table 7 provide sufficient clues about the


Table 8
comparative performance of the centIPA. However, mentioned perfor-
Statistical comparison between centIPA and other metaheuristics by using best objective
function values. mance of the centIPA should also be validated with a suitable statistical
centIPA vs. IPA vs. ABC vs. BA vs. BAM vs. ACO test. Wilcoxon signed rank test is one of the most commonly used
𝑍-val. 2.4175 2.4175 2.6600 2.6600 2.6600
statistical test for comparing the metaheuristics (Aslan et al., 2019).
𝜌-val. 0.0156 0.0156 0.0078 0.0078 0.0078 If the significance level (𝜌) is less than 0.05, it is understood that the
𝑊+ 0 0 0 0 0 difference between algorithms is enough for a statistical significance
𝑊− 28 28 36 36 36 in favor of one of them (Aslan et al., 2019). Otherwise, the difference
Sign. centIPA centIPA centIPA centIPA centIPA
between algorithms is not sufficient for making a decision about which
centIPA vs. BBO vs. DE vs. ES vs. FA vs. GA
one is statistically significant (Aslan et al., 2019). The Wilcoxon signed
𝑍-val. 2.6600 2.6600 2.6600 2.6600 2.6600 rank test results by considering the best, worst and mean best objective
𝜌-val. 0.0078 0.0078 0.0078 0.0078 0.0078
function values found by the algorithms for the fifth battlefield scenario
𝑊+ 0 0 0 0 0
𝑊− 36 36 36 36 36 were given in Tables 8–10. In these tables, 𝑍 corresponds to the test
Sign. centIPA centIPA centIPA centIPA centIPA statistics and 𝑊 + shows the sum of ranks for which centIPA is worse
centIPA vs. MFA vs. PBIL vs. PSO vs. SGA vs. PGSO than the compared algorithm and 𝑊 − shows the sum of ranks for
𝑍-val. 2.6600 2.6600 2.6600 2.6600 2.6600 which centIPA is better than the compared algorithm. From the results
𝜌-val. 0.0078 0.0078 0.0078 0.0078 0.0078 given in Tables 8–10, it is easily seen that the qualities of the paths
𝑊+ 0 0 0 0 0 found by the centIPA generate a statistical significance in favor of
𝑊− 36 36 36 36 36
the same algorithm. The significance is in favor of centIPA for all
Sign. centIPA centIPA centIPA centIPA centIPA
comparisons on the basis of the best objective function values. While
the statistical comparison between centIPA and ABC, BA, ACO, BBO,
DE, ES, FA, GA, PBIL, PSO, SGA, or PGSO generates 𝜌 less than 0.05
for the 𝐷 equal to 25 and the third best metaheuristic for the 𝐷 equal and proves the promising performance of the centIPA, the calculated 𝜌
to 20 and 35. values are not enough to state that the centIPA is statistically better

9
S. Aslan Engineering Applications of Artificial Intelligence 112 (2022) 104789

Table 9 Table 11
Statistical comparison between centIPA and other metaheuristics by using worst 𝑆𝑟 and 𝑀𝑒 values of centIPA and IPA for the fifth scenario.
objective function values. 𝐷 centIPA IPA ABC
centIPA vs. IPA vs. ABC vs. BA vs. BAM vs. ACO
𝑆𝑟 𝑀𝑒 𝑆𝑟 𝑀𝑒 𝑆𝑟 𝑀𝑒
𝑍-val. 1.7616 2.4175 2.6600 1.6010 2.6600
5 100.000 67.940 100.000 217.050 100.000 199.500
𝜌-val. 0.0781 0.0156 0.0078 0.1093 0.0078
10 100.000 499.730 100.000 1505.980 100.000 795.900
𝑊+ 3 0 0 6 0
15 100.000 415.100 100.000 2482.550 100.000 1920.900
𝑊− 25 28 36 30 36
20 100.000 735.310 100.000 3045.650 100.000 3755.400
Sign. – centIPA centIPA – centIPA
25 100.000 994.370 95.000 4323.516 60.000 5096.000
centIPA vs. BBO vs. DE vs. ES vs. FA vs. GA 30 100.000 1466.220 92.000 4015.663 5.000 5910.000
35 96.000 2041.969 85.000 4136.859 0.000 –
𝑍-val. 2.6600 2.6600 2.6600 2.6600 2.6600
40 97.000 1895.052 90.000 4416.678 0.000 –
𝜌-val. 0.0078 0.0078 0.0078 0.0078 0.0078
𝑊+ 0 0 0 0 0
𝑊− 36 36 36 36 36
Sign. centIPA centIPA centIPA centIPA centIPA
equal to 20 and 25. For a visual inspection how the centIPA converges,
centIPA vs. MFA vs. PBIL vs. PSO vs. SGA vs. PGSO
the curves given in Fig. 4 should be controlled.
𝑍-val. 0.8727 2.6600 2.6600 2.6600 2.6600
𝜌-val. 0.3828 0.0078 0.0078 0.0078 0.0078
𝑊+ 11 0 0 0 0 6. Conclusion
𝑊− 25 36 36 36 36
Sign. – centIPA centIPA centIPA centIPA
IP algorithm or IPA is the first metaheuristic that models a well-
defined medical method known as the immune or convalescent plasma
treatment. In this study, the plasma extraction method of IP algorithm
Table 10 was modified and a new variant called centIPA was proposed for solv-
Statistical comparison between centIPA and other metaheuristics by using mean ing the UCAV path planning problem. The path planning capabilities of
objective function values.
centIPA were analyzed in detail by assigning different values to algo-
centIPA vs. IPA vs. ABC vs. BA vs. BAM vs. ACO
rithm specific control parameters and using battlefields configured with
𝑍-val. 2.4175 2.4175 2.6600 0.7373 2.6600
various anti-air weapon systems. Moreover, the centIPA was compared
𝜌-val. 0.0156 0.0156 0.0078 0.4609 0.0078
𝑊+ 0 0 0 12 0 with a set of metaheuristics including IPA, ABC, BA, ACO, BBO, DE,
𝑊− 28 28 36 24 36 ES, FA, GA, PBIL, PSO, and some variants of them such as I-ABC, IF-
Sign. centIPA centIPA centIPA – centIPA ABC, BE-ABC, BAM, JADE, CIPDE, CIJADE, MFA, SGA, and PGSO. The
centIPA vs. BBO vs. DE vs. ES vs. FA vs. GA comparative studies showed that the newly proposed plasma extraction
𝑍-val. 2.6600 2.6600 2.6600 2.6600 2.6600 technique improves the performance of the IPA and helps centIPA to
𝜌-val. 0.0078 0.0078 0.0078 0.0078 0.0078 outperform other tested algorithms for most of the battlefield scenarios.
𝑊+ 0 0 0 0 0
The promising results of both IPA and centIPA also inform that IPA
𝑊− 36 36 36 36 36
Sign. centIPA centIPA centIPA centIPA centIPA has a potential for solving difficult real world problems similar to the
centIPA vs. MFA vs. PBIL vs. PSO vs. SGA vs. PGSO
UCAV path planning and can be further improved by remodeling some
stages or integrating new operational steps into the algorithm as in the
𝑍-val. 1.7616 2.6600 2.6600 2.6600 2.6600
𝜌-val. 0.0781 0.0078 0.0078 0.0078 0.0078 centrifuge based plasma extraction. In addition to these, some modern
𝑊+ 5 0 0 0 0 representatives of metaheuristics including EHO, MS, EWA, and MBO
𝑊− 31 36 36 36 36 can be investigated by considering their performances on UCAV path
Sign. centIPA centIPA centIPA centIPA centIPA
planning and hybridized with the IPA for further improving the local
and global search capabilities of the mentioned algorithms.

than IPA, BAM or MFA for the comparisons on the basis of worst CRediT authorship contribution statement
objective function values. However, it should be noticed that the 𝜌
values calculated by considering the mean objective function values
Selcuk Aslan: Conceptualization, Methodology, Software, Writing
prove the significance of the centIPA again compared to IPA and MFA.
– review & editing.
The comparative studies between centIPA and other metaheuristics
for the fifth battlefield scenario was concluded with the convergence
analysis. For this purpose, the 𝑆𝑟 and 𝑀𝑒 values were calculated for Declaration of competing interest
centIPA, IPA and ABC by setting the threshold to 55 and given in
Table 11. The 𝑆𝑟 and 𝑀𝑒 values showed in Table 11 inform that the The authors declare that they have no known competing finan-
centIPA whose convergence performance validated previously also con- cial interests or personal relationships that could have appeared to
verges more quickly compared to the IPA. While the centIPA converges influence the work reported in this paper.
at least 2 times faster than IPA for the fifth scenario with 𝐷 equal to
30, 35, and 40, it converges at least 3 times faster than IPA for the fifth Funding
scenario with 𝐷 equal to 5 and 10, at least 4 times faster than IPA for the
fifth scenario with 𝐷 equal to 20 and 25, at least 5 times faster than IPA
This study was not funded by any organization.
for the fifth scenario with 𝐷 equal to 15. Similar interpretations can also
be made for the comparison between the convergence performances
of the centIPA and ABC. While the centIPA converges at least 2 times Ethical approval
faster than ABC for the fifth scenario with 𝐷 equal to 5, it converges at
least 4 times faster than ABC for the fifth scenario with 𝐷 equal to 15 This article does not contain any studies with human participants
and 30, at least 5 times faster than ABC for the same scenario with 𝐷 or animals performed by any of the authors.

10
S. Aslan Engineering Applications of Artificial Intelligence 112 (2022) 104789

Fig. 4. Convergence curves of centIPA, IPA and ABC algorithms for the fifth scenario with 𝐷 equal to 15 (a), 20 (b), 25 (c), 30 (d), 35 (e), and 40 (f).

References Dorigo, M., Stutzle, T., 2019. Ant colony optimization: overview and recent advances.
Handb. Metaheuristics 311–351. http://dx.doi.org/10.1007/978-3-319-91086-4-
10.
Aslan, S., Badem, H., Karaboga, D., 2019. Improved quick artificial bee colony (iqabc) Feng, Y., Deb, S., Wang, G.-G., Alavi, A.H., 2021. Monarch butterfly optimization: A
algorithm for global optimization. Soft Comput. 23 (24), 13161–13182. http: comprehensive review. Expert Syst. Appl. 168, 114418. http://dx.doi.org/10.1016/
//dx.doi.org/10.1007/s00500-019-03858-y. j.eswa.2020.114418.
Aslan, S., Demirci, S., 2020. Immune plasma algorithm: A novel meta-heuristic for Feng, Y., Wang, G.-G., 2022. A binary moth search algorithm based on self-learning
optimization problems. IEEE Access 8, 220227–220245. http://dx.doi.org/10.1109/ for multidimensional knapsack problems. Future Gener. Comput. Syst. 126, 48–64.
ACCESS.2020.3043174. http://dx.doi.org/10.1016/j.future.2021.07.033.
Aslan, S., Demirci, S., 2021. Performance investigation of parallel immune plasma Gao, W., Liu, S., Huang, L., 2012. A global best artificial bee colony algorithm for
algorithm. In: 2021 International Conference on INnovations in Intelligent Systems global optimization. J. Comput. Appl. Math. 236 (11), 2741–2753. http://dx.doi.
and Applications (INISTA). pp. 1–5. http://dx.doi.org/10.1109/INISTA52262.2021. org/10.1016/j.cam.2012.01.013.
9548547. Hung, I.F., To, K.K., Lee, C.-K., Lee, K.-L., Chan, K., Yan, W.-W., Liu, R., Watt, C.-L.,
Chan, W.-M., Lai, K.-Y., 2011. Convalescent plasma treatment reduced mortality in
Baluja, S., 1994. Population-Based Incremental Learning: A Method for Integrating
patients with severe pandemic influenza a (h1n1) 2009 virus infection. Clin. Infect.
Genetic Search Based Function Optimization and Competitive Learning. Tech. Rep.,
Dis. 52 (4), 447–456. http://dx.doi.org/10.1093/cid/ciq106.
Carnegie Mellon University, USA.
Katoch, S., Chauhan, S.S., Kumar, V., 2021. A review on genetic algorithm: past,
Beyer, H.-G., 2001. The Theory of Evolution Strategies. Springer Science & Business present, and future. Multimedia Tools Appl. 80 (5), 8091–8126. http://dx.doi.org/
Media, http://dx.doi.org/10.1007/978-3-662-04378-3. 10.1007/s11042-020-10139-6.
Chen, Y., Pi, D., Xu, Y., 2021. Neighborhood global learning based flower pollination Khatib, W., Fleming, P.J., 1998. The stud ga: a mini revolution? In: International
algorithm and its application to unmanned aerial vehicle path planning. Expert Conference on Parallel Problem Solving from Nature. pp. 683–691. http://dx.doi.
Syst. Appl. 170, 114505. http://dx.doi.org/10.1016/j.eswa.2020.114505. org/10.1007/BFb0056910.

11
S. Aslan Engineering Applications of Artificial Intelligence 112 (2022) 104789

Li, P., Duan, H., 2012. Path planning of unmanned aerial vehicle based on improved Schwarzrock, J., Zacarias, I., Bazzan, A.L., de Araujo Fernandes, R.Q., Moreira, L.H.,
gravitational search algorithm. Sci. China Technol. Sci. 55 (10), 2712–2719. http: de Freitas, E.P., 2018. Solving task allocation problem in multi unmanned aerial
//dx.doi.org/10.1007/s11431-012-4890-x. vehicles systems using swarm intelligence. Eng. Appl. Artif. Intell. 72, 10–20.
Li, K., Ge, F., Han, Y., Xu, W., et al., 2020a. Path planning of multiple uavs with http://dx.doi.org/10.1016/j.engappai.2018.03.008.
online changing tasks by an orpfoa algorithm. Eng. Appl. Artif. Intell. 94, 103807. Sengupta, S., Basak, S., Peters, R.A., 2019. Particle swarm optimization: A survey of
http://dx.doi.org/10.1016/j.engappai.2020.103807. historical and recent developments with hybridization perspectives. Mach. Learn.
Li, B., Gong, L.-g., Yang, W.-l., 2014. An improved artificial bee colony algorithm based Knowl. Extr. 1 (1), 157–191. http://dx.doi.org/10.3390/make1010010.
on balance-evolution strategy for unmanned combat aerial vehicle path planning. Shen, C., Wang, Z., Zhao, F., Yang, Y., Li, J., Yuan, J., Wang, F., Li, D., Yang, M.,
Sci. World J. 2014, http://dx.doi.org/10.1155/2014/232704. Xing, L., 2020. Treatment of 5 critically ill patients with covid-19 with convalescent
Li, J., Lei, H., Alavi, A.H., Wang, G.-G., 2020b. Elephant herding optimization: variants, plasma. JAMA 323 (16), 1582–1589. http://dx.doi.org/10.1001/jama.2020.4783.
hybrids, and applications. Mathematics 8 (9), 1415. http://dx.doi.org/10.3390/ Storn, R., Price, K., 1997. Differential evolution–a simple and efficient heuristic for
global optimization over continuous spaces. J. Global Optim. 11 (4), 341–359.
math8091415.
http://dx.doi.org/10.1023/A:1008202821328.
Li, W., Wang, G.-G., Gandomi, A.H., 2021. A survey of learning-based intelligent
Tang, Z., Zhou, Y., 2015. A glowworm swarm optimization algorithm for uninhabited
optimization algorithms. Arch. Comput. Methods Eng. 28, 3781–3799. http://dx.
combat air vehicle path planning. J. Intell. Syst. 24 (1), 69–83. http://dx.doi.org/
doi.org/10.1007/s11831-021-09562-1.
10.1515/jisys-2013-0066.
Luo, Q., Li, L., Zhou, Y., 2017. A quantum encoding bat algorithm for uninhabited
Wang, G.-G., 2018. Moth search algorithm: a bio-inspired metaheuristic algorithm for
combat aerial vehicle path planning. Int. J. Innovative Comput. Appl. 8 (3),
global optimization problems. Memet. Comput. 10 (2), 151–164. http://dx.doi.org/
182–193. http://dx.doi.org/10.1504/IJICA.2017.086642.
10.1007/s12293-016-0212-3.
Ma, H., Simon, D., Siarry, P., Yang, Z., Fei, M., 2017. Biogeography-based optimization: Wang, G.-G., Chu, H.E., Mirjalili, S., 2016. Three-dimensional path planning for ucav
a 10-year review. IEEE Trans. Emerg. Top. Comput. Intell. 1 (5), 391–407. http: using an improved bat algorithm. Aerosp. Sci. Technol. 49, 231–238. http://dx.doi.
//dx.doi.org/10.1109/TETCI.2017.2739124. org/10.1016/j.ast.2015.11.040.
Mac, T.T., Copot, C., Tran, D.T., De Keyser, R., 2016. Heuristic approaches in robot Wang, G.-G., Gandomi, A.H., Alavi, A.H., Gong, D., 2019. A comprehensive review of
path planning: A survey. Robot. Auton. Syst. 86, 13–28. http://dx.doi.org/10.1016/ krill herd algorithm: variants, hybrids and applications. Artif. Intell. Rev. 51 (1),
j.robot.2016.08.001. 119–148. http://dx.doi.org/10.1007/s10462-017-9559-1.
Marson, P., Cozza, A., De Silvestro, G., 2020. The true historical origin of convalescent Wang, G.-G., Guo, L., Duan, H., Liu, L., Wang, H., 2012a. A bat algorithm with
plasma therapy. Transfus. Apher. Sci. 59 (5), 102847. http://dx.doi.org/10.1016/ mutation for ucav path planning. Sci. World J. 2012, http://dx.doi.org/10.1100/
j.transci.2020.102847. 2012/418946.
Miao, F., Zhou, Y., Luo, Q., 2019. A modified symbiotic organisms search algorithm Wang, G.-G., Guo, L., Duan, H., Liu, L., Wang, H., 2012b. A modified firefly algorithm
for unmanned combat aerial vehicle route planning problem. J. Oper. Res. Soc. 70 for ucav path planning. Int. J. Hybrid Inf. Technol. 5 (3), 123–144. http://dx.doi.
(1), 21–52. http://dx.doi.org/10.1080/01605682.2017.1418151. org/10.14257/ijhit.2012.5.3.11.
Pan, J.-S., Liu, N., Chu, S.-C., 2020. A hybrid differential evolution algorithm and Wang, G.-G., Guo, L., Duan, H., Liu, L., Wang, H., Shao, M., 2012c. Path planning for
its application in unmanned combat aerial vehicle path planning. IEEE Access 8, uninhabited combat aerial vehicle using hybrid meta-heuristic de/bbo algorithm.
17691–17712. http://dx.doi.org/10.1109/ACCESS.2020.2968119. Adv. Sci. Eng. Med. 4 (6), 550–564. http://dx.doi.org/10.1166/asem.2012.1223.
Pan, J.-S., Liu, J.-L., Hsiung, S.-C., 2019. Chaotic cuckoo search algorithm for solving Wang, G.-G., Guo, L., Duan, H., Wang, H., Liu, L., Shao, M., 2012d. A hybrid
unmanned combat aerial vehicle path planning problems. In: Proceedings of the metaheuristic de/cs algorithm for ucav three-dimension path planning. Sci. World
2019 11th International Conference on Machine Learning and Computing. pp. J. 2012, http://dx.doi.org/10.1100/2012/583973.
224–230. http://dx.doi.org/10.1145/3318299.3318310. Wu, P., Li, T., Song, G., 2020. Ucav path planning based on improved chaotic particle
Parkin, J., Cohen, B., 2001. An overview of the immune system. Lancet 357 (9270), swarm optimization. In: 2020 Chinese Automation Congress (CAC). IEEE, pp.
1777–1789. http://dx.doi.org/10.1016/S0140-6736(00)04904-7. 1069–1073. http://dx.doi.org/10.1109/CAC51589.2020.9326556.
Pérez-Carabaza, S., Scherer, J., Rinner, B., López-Orozco, J.A., Besada-Portas, E., Xu, C., Duan, H., Liu, F., 2010. Chaotic artificial bee colony approach to uninhabited
combat air vehicle (ucav) path planning. Aerosp. Sci. Technol. 14 (8), 535–541.
2019. Uav trajectory optimization for minimum time search with communication
http://dx.doi.org/10.1016/j.ast.2010.04.008.
constraints and collision avoidance. Eng. Appl. Artif. Intell. 85, 357–371. http:
Zhang, Y., Wu, L., Wang, S., 2011. Ucav path planning based on fscabc. Inf.- Int.
//dx.doi.org/10.1016/j.engappai.2019.06.002.
Interdiscip. J. 14 (3), 687–692.
Qu, C., Gai, W., Zhang, J., Zhong, M., 2020. A novel hybrid grey wolf optimizer
Zhang, Y., Wu, L., Wang, S., 2013. Ucav path planning by fitness-scaling adaptive
algorithm for unmanned aerial vehicle (UAV) path planning. Knowl.-Based Syst.
chaotic particle swarm optimization. Math. Probl. Eng. 2013, http://dx.doi.org/10.
194, 105530. http://dx.doi.org/10.1016/j.knosys.2020.105530.
1155/2013/705238.
Ropero, F., Muñoz, P., R-Moreno, M.D., 2019. Terra: A path planning algorithm
Zhang, S., Zhou, Y., Li, Z., Pan, W., 2016. Grey wolf optimizer for unmanned combat
for cooperative ugv–uav exploration. Eng. Appl. Artif. Intell. 78, 260–272. http: aerial vehicle path planning. Adv. Eng. Softw. 99, 121–136. http://dx.doi.org/10.
//dx.doi.org/10.1016/j.engappai.2018.11.008. 1016/j.advengsoft.2016.05.015.
Schenkel, A., Deville, M., Sawley, M., Hagmann, P., Rochat, J.-D., 2013. Flow Zhu, W., Duan, H., 2014. Chaotic predator–prey biogeography-based optimization
simulation and hemolysis modeling for a blood centrifuge device. Comput. & Fluids approach for ucav path planning. Aerosp. Sci. Technol. 32 (1), 153–161. http:
86, 185–198. http://dx.doi.org/10.1016/j.compfluid.2013.06.019. //dx.doi.org/10.1016/j.ast.2013.11.003.

12

You might also like