Appl Intell

https://doi.org/10.1007/s10489-017-1082-1

Modified collective decision optimization algorithm

with application in trajectory planning of UAV
Qingyang Zhang1 · Ronggui Wang1 · Juan Yang1 · Kai Ding1 · Yongfu Li1 ·
Jiangen Hu1

© Springer Science+Business Media, LLC 2017

Abstract Recently, a new heuristic method called col- Keywords Collective decision optimization algorithm ·
lective decision optimization algorithm (CDOA) was pro- Unmanned aerial vehicle · Path planning ·
posed. This paradigm is inspired from the decision-making Decision-making behaviour
behaviour of human beings, including the different fac-
tors influencing the decisions, such as experience, opin-
ion of others, group thinking, opinion of the leader, and 1 Introduction
innovation. However, the original version of the algorithm
concentrated only on a fixed evolution order. This study The path planning of unmanned aerial vehicles (UAVs) is
introduces an extended version of the CDOA (ECDOA) for a complicated global optimum task, which involves finding
the evolution mechanism without making any major con- optimal flight routes that have a high probability of avoiding
ceptual change to its architecture. In ECDOA, all the agents detection in complex combat field scenarios. In the field of
break the bonds of the original operator sequence. With military applications, UAVs are considered as a key element
the exception of the innovation operator, all the other oper- of modern aerial weapon equipment, owing to their capa-
ators are initially stored in an external archive, and then bility to perform dangerous and challenging assignments in
each agent combines at least one randomly selected opera- remote and hazardous situations. Path planning refers to the
tor from this archive with an innovation operator to create determination of a proper space trajectory between an initial
new update orders. This method not only provides more cal- location and the assigned destination, and it is an indis-
culation sequences in each iteration, but also generates more pensable component of mission planning systems. In fact,
promising candidate solutions. In addition, several operators the planning of an ideal flight path is very difficult, as it
are further modified to improve the optimization abilities. needs to satisfy certain important requirements such as opti-
A comprehensive set of modern benchmark functions and mality, completeness, and computational complexity [1].
UAV path planning are required to verify the effective- A series of conventional algorithms have been employed
ness of the ECDOA thoroughly. The results of the series- to solve this complex optimization task, including evolu-
simulation comparison, which simultaneously consider both tionary computation [2], genetic algorithm [3], ant colony
the convergence and accuracy, indicate that ECDOA is more algorithm [4], and neural network [5]. However, these meth-
effective and feasible than the other state-of-art optimization ods do not show a satisfactory performance in finding an
paradigms. optimal route, since they are easily trapped into a local opti-
mum. This study mainly focuses on trajectory planning in a
two-dimensional space.
Over the past decade, various optimization algorithms
 Qingyang Zhang
hgd qy@126.com
have been proposed to solve various optimization problems
in different fields. These techniques can be widely grouped
1 School of Computer and Information, Hefei University into two classes: gradient-based methods and stochastic
of Technology, Hefei 230009, China search methods [6]. The former aims to employ gradient
 Q. Zhang et al.

information to provide search guidance. For relatively sim- involves the use of five operators that are connected by
ple or ideal models, gradient-based approaches can generate a multi-step position-selected scheme, in which the new
promising results. However, in many complex optimiza- location generated by a previous movement is defined as
tion scenarios, these approaches face significant challenges the start point for the next movement. However, there is a
and difficulties, because the gradient information is either drawback for this method. Each search agent has to obey
not reliable or is not available. The potential limitations of a settled operator arrangement, which could result in pre-
existing gradient-based strategies have prompted an increas- mature convergence during evolution. To overcome these
ing number of scholars to utilize heuristic algorithms to drawbacks, the authors present an extended version of the
deal with highly complex optimization tasks that are char- CDOA (ECDOA), which overcomes the shortcomings with-
acterized as non-convex, non-differentiable, discontinuous, out requiring any major conceptual change to its structure.
etc. [7, 8]. Over the past decade, metaheuristic optimiza- In the ECDOA, all the operators other than the innova-
tion methods have become increasingly popular due to tion operator are initially stored in an external archive, and
certain obvious advantages such as simplicity, flexibility, each agent combines at least one randomly selected opera-
derivation-free mechanism, and local optima avoidance. tor from the archive with the innovation operator to create
These nature-inspired techniques can be divided into three new update orders. In this method, it is possible for each
groups: evolutionary algorithms (EAs), swarm intelligence agent to not only move its position freely but also to employ
paradigms (SIs), and physics-based methods (PBs). a new series for generating more number of promising solu-
Evolutionary methodologies are designed by simulating tions in each iteration. The search space can be explored
the laws of biological evolution processes, such as recom- or exploited effectively. It should be noted that the ran-
bination, mutation, and selection. Examples of evolutionary dom walk, which is a local search strategy, yields several
algorithms include Genetic Algorithm (GA) [9], which challenging points around the best individual and remains
imitates the theory of Darwinian evolution, Biogeography- unchanged from the CDOA. In order to achieve better per-
Based Optimizer (BBO), which is inspired from natural formance, some operators are further modified to maintain
biogeography [10], and Bird Mating Optimizer (BMO), a good balance between exploration and exploitation.
which is inspired from natural evolution [11]. The rest of this study is organized as follows. Section 2
The second class of nature-inspired algorithms includes introduces the basic principles of CDOA. Section 3 presents
the swarm intelligence methods, which involves the sim- the ECDOA. In Section 4, numerical simulation is car-
ulation of different kinds of swarm behaviours such as ried out on a comprehensive set of benchmark functions.
searching behaviour, mating behaviour, foraging behaviour, Section 5 investigates the effectiveness of the ECDOA
etc. Particle Swarm Optimization (PSO) [12] is inspired methodology in solving path planning of UAVs. Section 6
from the flocking behaviour of foraging birds, the Group concludes this study.
search optimizer (GSO) is inspired from animal scanning
mechanisms [13], Cuckoo Search (CS) is inspired from the
parasitic bio-interactions in living creatures [14], the Bat 2 Collective decision optimization algorithm
algorithm (BA) is inspired from the flight behaviour of bats (CDOA)
[15], the Whale Optimization Algorithm (WOA) is inspired
from the social behaviour of humpback whales [16], and To introduce this newly presented method clearly, this
the Firefly algorithm (FA) is inspired from the pollination section first describes the inspiration for CDOA [22]. Then,
process of flowers [17]. the basic calculation models and flowchart are provided in
The third class of nature-inspired algorithms includes detail.
physics-based paradigms. Unlike EAs and SIs, the PBs are
proposed by imitating the diverse physical characteristic 2.1 Inspiration
rules in nature. Simulated Annealing (SA) is inspired from
the process of metallurgic annealing [18], artificial raindrop The main principle of this new technique is inspired from
algorithm (ARA) is inspired from the natural rainfall phe- nature. It is based on the observation of the human decision-
nomenon [19], the Gravitational Search Algorithm (GSA) is making process and how people develop solutions for prob-
inspired from the law of gravity and mass interactions [20], lems or doubts in the real world. In this study, the act of
and the Multi-Verse Optimizer (MVO) is inspired from the holding a meeting (see Fig. 1) to develop a good strategy
theory of multiverse in physics [21]. or plan is chosen as an example to explain this inspira-
This study mainly focuses on the CDOA, which is tion. Each participant in the meeting is called a decider, and
a newly proposed algorithm inspired from the decision- each decider has a corresponding idea or plan. Besides, each
making behaviour of humans and has been successfully decider can present ideas and exchange their thoughts with
used to train neural networks [22]. The principle of CDOA the other participants through discussion, which promotes
Modified collective decision optimization algorithm...

2.2 Mathematical model and algorithm

2.2.1 Group generation

Similar to other heuristic methods, CDOA starts with an ini-

tial population (P op) filled with N feasible randomly gen-
erated solution vectors, and each agent has a corresponding
position defined as follows.
Xi (t) = (xi1 (t), xi2 (t), ..., xiD (t)), i = 1, 2, ..., N (1)

xik (t) = LB k + r × (U B k − LB k ), k = 1, 2, ..., D (2)

Fig. 1 the phenomenon of making plans in the meeting P op(t) = (X1 (t), X2 (t), ..., XN (t))
where N denotes the population scale, D is the dimension
of the unresolved task, t refers to the number of iterations,
r is a uniform random number, which varies in (0, 1), and
the formulation of a desirable plan or project. The best U B and LB indicate the upper and lower bounds of the
plan/project selected from the proposed plans/projects is variables, respectively.
considered as the optimal result.
On examining this action carefully, we find that develop- 2.2.2 Experience-based phase
ing a good plan is not a simple procedure, because deciders
have to consider or weigh some of the important factors In general, when we encounter a decision-making problem
in the decision-making process, as described below. Firstly, or some other task, we consider how to develop or utilize a
personal experience is typically the first option that is uti- good plan or project to solve these issues based on personal
lized for developing a good plan, since it originates from the experience. In this model, this is set as the best location
accumulation of daily life experiences and provides useful of the agent (ϕP ) so far, and its calculation formula is as
guidance. Therefore, some researchers firmly believe that follows.
personal experience is an important concept in the study
of collective decision-making behaviour [23, 24]. Secondly, newX = X (t) + −
i0 i
→τ × step (t) × d
0 size 0
deciders are more likely to consult the viewpoints of or d0 = ϕP − Xi (t) (3)
solicit ideas from others, who may introduce a new idea
as reference. As suggested in [25–27], this method may where − →
τ0 denotes a random vector uniformly distributed
encourage the decider to work out a better plan to some in the interval (0, 1), stepsize (t) is the step size in the ith
extent. Thirdly, since this action is a group behaviour, it is generation, and d0 is a direction vector.
easy to form a trend that is biased towards the opinion held
by most of the deciders. According to [28–30], this trend 2.2.3 Others’-based phase
is called group thinking, and it has the potential to affect
the opinions of the decider. Conformity behaviour is one of It is generally known that when facing a difficult problem,
the typical forms of collective thinking. Fourthly, the role people tend to consider the opinion of others, which is help-
of the leader should not be neglected, because the leader ful in making a good choice. In this scenario, the decider
determines the direction and result of the decision. All the can receive useful information by considering the opinion
planners should be appropriated to listen to the standpoints of others. In the model, another individual (Xj (t)) instead
of the leader, which would inspire them to carry out better of the current search agent(Xi (t)), is selected randomly
plans [31–33]. Finally, the deciders should promote orig- from the evolution population to generate new solutions
inality, inspire creativity, and encourage innovation. This according to the following equation.
approach may introduce some new methods or ideas for newX = newX + −
i1
→
τ × step (t) × d
i0 1 size 1
arriving at better results. Therefore, innovation is as impor-
d1 = beta1 × d0 + beta11 × (Xj (t) − Xi (t)) (4)
tant as the other factors described above for contributing to
the development of an ideal plan [34–36]. Undoubtedly, var- where j is a random integer between 1 and N. −
→
τ is a ran- 1
ious other factors may exist in the decision-making process; dom vector in the range (0, 1), d1 is the direction vector, and
however, for the sake of convenience, this study considers beta1 and beta11 are two random values distributed in the
only the five elements mentioned above. intervals (−1, 1) and (0, 2), respectively.
 Q. Zhang et al.

−
→
2.2.4 Group thinking-based phase where Wq refers to a random vector with each value gener-
ated between 0 and 1.
In a group scenario, all the members have the right to
express or change their ideas casually. As a result, it is not 2.2.6 Innovation-based phase
difficult to form a trend that is biased towards the opin-
ion held by most of the deciders. This trend is called group It is generally known that innovation exists across all walks
thinking, and it can influence a decider to change an idea of life, and the decision-making behaviour is no exception.
spontaneously. In the model, group thinking (ϕG ) is set as It not only brings us new or better things, but also helps us to
the geometric centre of all the members, and is described as break free from the bondage of obsolete things. Therefore,
follows: in the model, the one-dimensional mutation strategy, which
1 is expressed below, is designed to transform the variables
ϕG = (X1 (t), X2 (t), ..., XN (t)) slightly.
N
1  1 1  2 1  D
N N N
={ xi (t), xi (t), ..., xi (t)} (5) r1  MF
N N N
i=1 i=1 i=1
newXi4 = newXi3
The following formula is designed to generate new solu- p
newXi4 = LB(p) + r2 × (U B(p) − LB(p)) (9)
tions:

newXi2 = newXi1 + −
→
τ2 × stepsize (t) × d2 where r1 and r2 represent the random numbers between 0
d2 = beta2 × d1 + beta22 × (ϕG − Xi (t)) (6) and 1, p is a random integer in the interval [1, D], and MF
is the innovation factor.
where − →
τ2 refers to a random vector with each value uni- Another concern is the determination of how to update
formly distributed between 0 and 1, d2 denotes the direction the step size with the generations (t). To maintain a
vector, and beta2 and beta22 are the random values gener- good trade-off between exploration and exploitation, the
ated in the intervals (−1, 1) and (0, 2), respectively.

2.2.5 Leader-based phase

The leader, a pivotal decider, is generally considered to be a

highly learned person who shares his or her knowledge with
the deciders throughout the decision-making process. It is
obvious that the quality of a leader is very important, since
the leader not only influences the direction of the final deci-
sion, but also determines the final decision. In the model,
the best agent of the population based on fitness, is chosen
as the Leader (ϕL ).

newXi3 = newXi2 + −
→
τ3 × stepsize (t) × d3
d3 = beta3 × d2 + beta33 × (ϕL − Xi (t)) (7)

where − →
τ3 is a random vector with each value distributed
between 0 and 1, d3 is the direction vector, and beta3 and
beta33 are two random values generated in the intervals
(−1, 1) and (0, 2), respectively.
In addition, there is a hidden phenomenon that deter-
mines how the idea or plan of the leader is changed. Here,
we suppose that his thoughts can only be changed by him-
self. In the model, random walk technique is adopted to
modify the location of ϕL slightly. As a result, more solu-
tions that can be used to find the global optimum may be
found in the region around ϕL .
−
→
newXq = ϕL + Wq (q = 1, 2, 3, 4, 5) (8) Fig. 2 The flowchart of the CDOA algorithm
Modified collective decision optimization algorithm...

CDOA employs an adaptive updating mechanism, which is below. The latter limitation cases are described in detail in
described as follows: Section 3.2.5.
 
t −1
stepsize (t) = 2 − 1.7 (10)
T −1
where T is the maximum number of generations. Figure 2
provides the flowchart of the CDOA.

3 Extension collective decision optimization

algorithm

On inspecting the previous section carefully, it can be seen

that the entire evolution process of CDOA involves five 3.2 Some modified operators
main operators, namely experience-based, others’-based,
group thinking-based, leader-based, and innovation-based 3.2.1 Modified experience-based phase
phases, and a random walk strategy for searching within the
area around the best individuals. The multi-step position- In the modified model, this phase may not be the first oper-
selectable scheme is the main design scheme, and it utilizes ator based on the new calculation sequence. If this phase
these operators. is assigned as the first operator, the update is performed
It can be seen that there are two limiting cases during according to the original calculation formula (3). Other-
the evolution. On the one hand, each search agent obeys a wise, the operator should be modified slightly, as expressed
fixed iteration order ((3) −→ (4) −→ (6) −→ (7) −→ (9) below. However, personal experience is also defined as the
or (8)) in every generation. In other words, the population best position of the individual (ϕP ) so far.
species generate new solutions by following a fixed order.
This approach may result in premature convergence and
local optimum, because some areas with significant solution
information cannot be explored and exploited effectively.
On the other hand, the mutation factor is very large and the
mutation dimension is very small. It is thus possible to fine-
tune these shortages for improving the search performance
of the algorithm. The above-mentioned factors motivated
the authors to modify the original version of the algorithm
in order to enhance the optimization capability.

3.1 Construct evolution sequence

This section mainly reviews the strategy for constructing a Where z is the order indicator in the evolution sequence,
new evolution sequence. As described previously, the orig- −
→
τz is a random vector with each number uniformly gener-
inal version of the algorithm has five different operators, ated in the interval (0, 1), stepsize (t) denotes the step size
which are closely related to the multi-step position selection of the current iteration (t), dz is the new direction of move-
scheme. However, the original version has the following ment, and βz and φz are the random numbers in the ranges
limitations: invariable update sequence, large mutation fac- (−1, 1) and (0, 2), respectively.
tor, and small mutation dimension. In the first case, this
study introduces a new sequence generation method. Specif- 3.2.2 Modified others’-based phase
ically, the four distinct operators (excluding innovation-
based phase) are assigned a number and then stored in the As a significant part of the operator archive, this phase
external archive. For each search agent, a random integer may be arranged into two indexes: NO. 1 or others. The
distributed between 1 and 4 will be produced to deter- difference between the two indexes is the way in which
mine the number of operators involved in the new update the moving direction is defined (see the pseudo code). In
order. Then, a random integer vector is generated to confirm ECDOA, an individual (Xj (t)) is randomly selected from
the order and to determine the operator in the construc- the population. The calculation is performed according to
tion sequence. The implemented pseudo codes are described the following algorithm.
 Q. Zhang et al.

the moving direction (dz ) has little distinction with the

original version, which is provided below. In ECDOA,
Leader (ϕL ) is also defined as the fittest element among the
population.

Where z is the order indicator in the evolution sequence,

−
→
τz is a random vector with each number uniformly gener-
ated in the interval (0, 1), stepsize (t) denotes the step size Where z is the order indicator in the evolution sequence,
of the current iteration (t), dz is the new direction of move- −
→
τz is a random vector with each number uniformly gener-
ment, and βz and φz are the random numbers in the ranges ated in the interval (0, 1), stepsize (t) denotes the step size
(−1, 1) and (0, 2), respectively. of the current iteration (t), dz is the new direction of move-
ment, and βz and φz are the random numbers in the ranges
3.2.3 Modified group thinking-based phase
(−1, 1) and (0, 2), respectively.
The external archive has four different operators, which are
3.2.5 Modified innovation-based phase
used to establish a new evolution sequence. Each one may
be chosen as the first phase. The concern here is that the
It can be seen that the innovation-based phase is similar to
moving direction of the first operator is only slightly dif-
the mutation operator, which is widely adopted in evolu-
ferent from that of the other order operators; however, the
tionary algorithms. The innovation-based phase is capable
calculated equations of ϕG are the same as those used in the
of making small changes among the variables. However,
original version. The new position is generated according to
in the original version, the mutation control factor is very
the following pseudo codes.
large, which causes the position transformation frequency
to become very high. In addition, the mutation variable is
one-dimensional, which reduces the probability of finding
better solutions. The aforementioned shortcomings affect
the performance of the algorithm to some extent. In the new
modified model, the mutation factor is slightly degraded.
The mutation dimension is also slightly regulated.
δ1  MF
newXiz = newXiz−1

newXiz = LB() + δ2 . × (U B() − LB()) (11)
where z is the order indicator in the evolution sequence,
δ1 is a random value distributed between 0 and 1,  is a two-
Where z is the order indicator in the evolution sequence, dimensional or three-dimensional integral vector generated
−
→
τz is a random vector with each number uniformly gener- in the range [1, D], δ2 is a random vector in the interval
ated in the interval (0, 1), stepsize (t) denotes the step size (0, 1) with the same scale as that of , and MF is the
of the current iteration (t), dz is the new direction of move- mutation control factor.
ment, and βz and φz are the random numbers in the ranges
(−1, 1) and (0, 2), respectively. 3.3 Comparison between ECDOA and CDOA

3.2.4 Modified leader-based phase In this section, the differences and similarities between the
modified version and the original version are highlighted.
As four operators are stored in the archive simultaneously, The major difference is that the strategy of ECDOA is
each phase may be chosen as the first operator. Therefore, to construct an evolutionary sequence, which refers to the
Modified collective decision optimization algorithm...

creation of a new evolution order based on different opera- 4 Numerical simulation

tors and numbers. It assists each agent with a new evolution
sequence by randomly recombining the existing operators in In this section, the numerical simulations that were con-
each run. In other words, each search individual has differ- ducted to evaluate the performance of ECDOA practically
ent sequences for generating new solutions in each iteration. are presented. The numerical simulations were carried out
In this way, ECDOA has more randomness and can generate on 30 modern benchmark functions, which were selected
more new individuals (with high diversity) that are dis- from CEC2014 [37].
tributed in the solution space to overcome premature conver-
gence. Nevertheless, this concept does not appear in CDOA, 4.1 Benchmark functions
since the original version obeys a fixed generation pattern.
Moreover, some operators are further modified to improve A comprehensive set of 30 test functions that were designed
the optimization capability of the algorithm. For instance, for the special issue of CEC2014, which included the
the innovation factor and mutation dimension involved in competition on single objective and complex real param-
the innovation-based phase are changed appropriately. The eter numerical optimization problems, is adopted in this
enlargement of the mutation dimension makes it possible to section. Appendix provides brief descriptions of these
search for better solutions in a larger region, thus achieving a functions, which can be grouped into four main classes:
better performance. Although there exists many differences unimodal functions (f1 -f3 ), multi-modal functions (f4 -
between ECDOA and CDOA, they have the same design f16 ), hybrid functions (f17 -f22 ), and composition functions
machine, which is a multi-step position-selectable scheme. (f23 -f30 ).
These observations guarantee that the ECDOA is capable
of showing promising performance theoretically; however, 4.2 Simulation platform and parameter settings
it is necessary to investigate the performance of ECDOA
when it is solving actual problems. This is the motivation All the simulations were performed on the same machine,
of the next section, in which a set of benchmark functions which was equipped with Intel Xeon E3-1226 3.30 GHz
commonly used in literature are solved using the improved processor, 8 GB memory, Windows 10 operating system,
method. and Matlab R2009a.
As described earlier, the pseudo codes of the ECDOA can In this section, the mean value (fmean ) and standard
be summarized in Algorithm 6. deviation (std) of the function error ((f − f ∗ )) are set as
performance indicators for assessing the effectiveness of
each method. Here, f is the best fitness value and f ∗ is
the real global optimal solution of the benchmark problem.
To compare the ECDOA with its competitors statistically,
the Wilcoxon’s rank sum tests [38] at a significance level
of 0.05 was used to evaluate whether any statistical dif-
ference exists between the two methods for each task.
In addition, the maximum number of function evaluations
(MaxF ES) is regarded as the termination criterion. For
the purpose of displaying the convergence characteristics
of the method, the mean error values in 25 independent
runs at specified checkpoints (0.1, 1, 2, ..., D)×10000 were
recorded.
In this section, to establish the clear superiority of the
ECDOA methodology, the performance of ECDOA is com-
prehensively compared with eight original swarm intelli-
gence algorithms, namely CS [14], BA [15], FA [17], PSO
[39], GSO [13], GSA [20], WOA [16], and MVO [21],
and nine advanced variants of other heuristic algorithms,
namely CLPSO [39], ORCS [40], QPSO [41], AGGSA
[42], PSOFIPS [43], CEP [44], PSOcfLocal [45], NSTLBO
[46] and UPSO [47]. For ECDOA, there are two control
parameters: stepsize , which decays monotonically from 2 to
0.3 and MF = 0.45. For PSO, the inertia weight ω decreases
form 0.9 to 0.4 over the course of the iteration, and the
Table 1 Simulation results obtained by original swarm intelligence methods on CEC2014 unimodal functions

Fun Index CDOA CS BA FA PSO GSO GSA WOA MVO ECDOA

f1 fmean 1.4318e+03− 7.9488e+02− 4.7565e+06+ 6.4418e+04+ 4.1300e+05+ 4.5399e+05+ 2.7765e+06+ 3.9699e+06+ 1.0486e+05+ 3.6970e+03
std 1.5410e+03 3.3460e+02 5.9944e+06 5.5825e+04 5.9143e+05 5.6450e+05 9.2327e+05 3.8622e+06 5.3347e+04 3.4486e+03
p 3.8418e-04 4.9750e-08 1.4157e-09 4.4598e-08 1.0414e-08 1.4157e-09 1.4157e-09 1.4157e-09 1.8002e-09 –
h 1 1 1 1 1 1 1 1 1 –
f2 fmean 2.2896e+02= 2.0042e+02− 2.6476e+03+ 3.3379e+03+ 4.4582e+07+ 1.3709e+03+ 5.0465e+02+ 5.4275e+04+ 5.8465e+03+ 2.6456e+02
std 4.0944e+01 2.2397e-01 3.1518e+03 3.5163e+03 2.2290e+08 1.3679e+03 3.3265e+02 3.4238e+04 3.8098e+03 9.2730e+01
p 2.6874e-01 1.5967e-09 7.6757e-08 5.4418e-05 2.1441e-05 6.7488e-06 8.4600e-04 1.4157e-09 1.4157e-09 –
h 0 1 1 1 1 1 1 1 1 –
f3 fmean 3.0356e+02+ 3.0002e+02+ 5.1816e+04+ 3.3949e+02+ 8.2397e+02+ 2.3818e+03+ 1.5192e+04+ 3.3885e+04+ 3.2057e+02+ 3.0001e+02
std 4.1748e+00 1.4015e-03 3.6247e+04 7.5361e+01 7.6219e+02 2.6771e+03 2.7361e+03 1.9438e+04 1.5320e+01 9.3158e-03
p 1.4157e-09 1.4495e-02 1.4157e-09 1.9668e-05 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 –
h 1 1 1 1 1 1 1 1 1 –
+ 1 1 3 3 3 3 3 3 3 –
− 1 2 0 0 0 0 0 0 0 –
= 1 0 0 0 0 0 0 0 0 –

+, − and = indicate that the performance of ECDOA is better than, worse than and similar to that of the corresponding algorithm, respectively
Q. Zhang et al.
Table 2 Simulation results obtained by original swarm intelligence methods on CEC2014 simple multimodal functions

Fun Index CDOA CS BA FA PSO GSO GSA WOA MVO ECDOA

f4 fmean 4.3338e+02+ 4.0003e+02− 4.4231e+02+ 4.3340e+02+ 4.3172e+02+ 4.0937e+02= 4.4712e+02+ 4.3219e+02+ 4.2441e+02+ 4.2278e+02
std 6.9517e+00 3.9971e-02 2.3752e+01 6.9253e+00 6.3144e+00 1.4702e+01 5.4136e+00 2.3107e+01 1.5473e+01 1.6372e+01
p 6.4854e-03 2.0298e-05 8.2388e-06 7.4363e-09 3.5691e-03 3.9623e-01 7.4782e-10 8.5515e-04 6.9009e-04 –
h 1 1 1 1 1 0 1 1 1 –
f5 fmean 5.2015e+02+ 5.2016e+02+ 5.2005e+02= 5.2000e+02+ 5.2006e+02= 5.2000e+02+ 5.2001e+02+ 5.2008e+02+ 5.2003e+02+ 5.1765e+02
std 5.4058e-02 5.8726e-02 9.8586e-02 1.1878e-03 7.0382e-02 1.4626e-03 7.2168e-04 8.4200e-02 4.4634e-02 6.6541e+00
p 1.3054e-07 3.6950e-07 7.7424e-02 3.8920e-05 5.4418e-01 3.8920e-05 3.8920e-05 9.6904e-01 5.5223e-03 –
h 1 1 0 1 0 1 1 0 1 –
f6 fmean 6.0120e+02= 6.0379e+02+ 6.1289e+02+ 6.0019e+02= 6.0110e+02+ 6.0417e+02+ 6.0459e+02+ 6.0702e+02+ 6.0164e+02+ 6.0055e+02
std 1.8977e+00 6.5220e-01 9.8622e-01 4.8040e-01 1.5465e+00 1.5890e+00 6.4531e-01 1.9854e+00 1.2623e+00 7.8021e-01
Modified collective decision optimization algorithm...

p 1.3174e-01 1.4328e-07 1.2695e-09 4.9583e-01 2.7887e-02 2.9386e-09 1.2695e-09 1.2695e-09 2.9532e-03 –

h 0 1 1 0 1 1 1 1 1 –
f7 fmean 7.0002e+02− 7.0013e+02= 7.6048e+02+ 7.0004e+02− 7.0258e+02+ 7.0059e+02+ 7.0001e+02− 7.0081e+02+ 7.0028e+02+ 7.0011e+02
std 1.4914e-02 3.1751e-02 1.7663e+01 1.9956e-02 4.5843e+00 2.8050e-01 5.8374e-03 4.7767e-01 1.1072e-01 8.9782e-02
p 1.6011e-07 8.7738e-02 1.4157e-09 3.2928e-05 2.8526e-04 2.8695e-08 6.5445e-10 2.5742e-09 2.1982e-06 –
h 1 0 1 1 1 1 1 1 1 –
f8 fmean 8.0000e+02+ 8.0554e+02+ 8.5540e+02+ 8.0963e+02+ 8.0497e+02+ 8.0016e+02+ 8.3533e+02+ 8.3687e+02+ 8.0837e+02+ 8.0000e+02
std 6.0168e-07 1.6947e+00 2.3284e+01 5.3559e+00 3.2211e+00 3.7228e-01 6.8396e+00 1.2301e+01 2.7177e+01 0
p 8.1168e-02 9.7285e-11 9.7285e-11 9.7285e-11 9.3138e-11 9.7285e-11 9.6035e-11 9.7285e-11 9.7285e-11 –
h 0 1 1 1 1 1 1 1 1 –
f9 fmean 9.0256e+02− 9.1476e+02+ 9.5413e+02+ 9.0892e+02+ 9.1015e+02+ 9.1466e+02+ 9.3244e+02+ 9.4084e+02+ 9.1578e+02+ 9.0436e+02
std 1.3403e+00 5.6349e+00 2.2403e+01 3.9592e+00 3.8281e+00 7.2637e+00 7.1054e+00 1.3894e+01 1.0135e+01 1.5226e+00
p 6.1211e-04 2.4638e-09 1.3530e-09 2.5872e-06 2.4791e-08 3.9533e-09 1.3506e-09 1.3530e-09 2.7620e-08 –
h 1 1 1 1 1 1 1 1 1 –
f10 fmean 1.0434e+03+ 1.1099e+03+ 2.3438e+03+ 1.3538e+03+ 1.2880e+03+ 1.0040e+03= 1.9263e+03+ 1.4798e+03+ 1.3036e+03+ 1.0075e+03
std 1.5606e+01 6.3306e+01 3.4710e+02 2.4820e+02 1.9392e+02 3.0690e+00 2.0716e+02 2.1267e+02 1.8459e+02 6.7555e+00
p 3.2614e-09 1.5967e-09 1.4157e-09 2.2967e-08 2.5742e-09 1.2532e-02 1.4144e-09 1.4157e-09 5.2120e-09 –
h 1 1 1 1 1 0 1 1 1 –
f11 fmean 1.3328e+03+ 1.5543e+03+ 2.5641e+03+ 1.3697e+03= 1.4685e+03+ 1.6164e+03+ 2.0473e+03+ 2.1095e+03+ 1.7541e+03+ 1.1737e+03
std 1.8752e+02 1.6581e+02 3.6010e+02 1.8916e+02 2.0276e+02 1.9295e+02 2.4747e+02 3.2254e+02 2.8129e+02 8.3210e+01
p 4.1349e-04 1.3079e-08 1.4157e-09 5.4418e-05 1.1707e-06 7.3796e-09 1.4144e-09 1.4157e-09 3.2614e-09 –
h 1 1 1 1 1 1 1 1 1 –
Table 2 (continued)

Fun Index CDOA CS BA FA PSO GSO GSA WOA MVO ECDOA

f12 fmean 1.2005e+03+ 1.2015e+03+ 1.2010e+03+ 1.2000e+03− 1.2001e+03= 1.2002e+03= 1.2000e+03− 1.2006e+03+ 1.2001e+03+ 1.2001e+03
std 2.1034e-01 3.0536e-01 5.1696e-01 1.4108e-01 8.5653e-02 8.5547e-02 3.8776e-03 2.6605e-01 6.7807e-02 8.0427e-02
p 3.0175e-07 2.5490e-05 1.5967e-09 1.4157e-09 2.8590e-01 7.4151e-01 1.4157e-09 8.2805e-09 8.4194e-02 –
h 1 1 1 1 0 0 1 1 0 –
f13 fmean 1.3002e+03+ 1.3002e+03+ 1.3005e+03+ 1.3001e+03+ 1.3002e+03+ 1.3003e+03+ 1.3000e+03− 1.3004e+03+ 1.3002e+03+ 1.3001e+03
std 4.3155e-02 4.3791e-02 2.6606e-01 2.4772e-02 8.3002e-02 1.1314e-01 1.0281e-02 1.7044e-01 5.2789e-02 2.1980e-02
p 1.3090e-07 5.2120e-09 1.5967e-09 3.1856e-03 1.8141e-06 1.4157e-09 1.4157e-09 1.5967e-09 2.3388e-05 –
h 1 1 1 1 1 1 1 1 1 –
f14 fmean 1.4002e+03+ 1.4002e+03+ 1.4005e+03+ 1.4001e+03= 1.4003e+03= 1.4002e+03+ 1.4004e+03+ 1.4003e+03+ 1.4001e+03= 1.4001e+03
std 3.1482e-02 3.0306e-02 3.3727e-01 4.0693e-02 2.9533e-01 9.2426e-02 5.5919e-02 1.5947e-01 3.2498e-02 3.8856e-02
p 4.3602e-02 1.9668e-05 2.2857e-09 4.9707e-01 5.0032e-02 2.4543e-07 1.4157e-09 4.9750e-08 1.4561e-01 –
h 1 1 1 0 0 1 1 1 0 –
f15 fmean 1.5012e+03+ 1.5012e+03= 2.4944e+03+ 1.5007e+03− 1.5008e+03− 1.5025e+03+ 1.5012e+03+ 1.5064e+03+ 1.5013e+03= 1.5011e+03
std 3.7465e-01 2.7779e-01 3.2437e+02 1.8765e-01 2.4894e-01 1.1075e+00 4.2879e-01 3.7107e+00 7.5058e-01 2.6589e-01
p 8.8779e-05 2.6043e-01 1.4157e-09 9.6957e-06 1.6711e-04 1.7954e-07 5.3467e-01 1.4157e-09 9.6905e-01 –
h 1 0 1 1 1 1 0 1 0 –
f16 fmean 1.6009e+03= 1.6030e+03+ 1.6041e+03+ 1.6023e+03+ 1.6020e+03+ 1.6023e+03+ 1.6041e+03+ 1.6033e+03+ 1.6026e+03+ 1.6011e+03
std 5.5758e-01 2.1373e-01 3.3011e-01 5.3763e-01 6.6615e-01 4.7764e-01 1.7336e-01 3.2141e-01 6.2849e-01 5.5768e-01
p 2.2156e-01 1.4157e-09 1.4157e-09 1.0585e-07 1.5127e-05 1.4549e-07 1.4157e-09 1.4157e-09 1.6401e-08 –
h 0 1 1 1 1 1 1 1 1 –
+ 9 10 12 7 9 10 10 13 11 –
– 2 1 0 3 1 0 3 0 0 –
= 2 2 1 3 3 3 0 0 2 –

+, − and = indicate that the performance of ECDOA is better than, worse than and similar to that of the corresponding algorithm, respectively
Q. Zhang et al.
Table 3 Simulation results obtained by original swarm intelligence methods on CEC2014 Hybrid functions

Fun Index CDOA CS BA FA PSO GSO GSA WOA MVO ECDOA

f17 fmean 2.9134e+03+ 1.7983e+03− 3.0285e+04+ 3.4316e+03+ 3.8631e+03+ 7.8062e+03+ 4.1013e+05+ 1.6847e+05+ 5.6651e+03+ 2.5816e+03
std 7.5291e+02 2.5030e+01 5.9748e+04 1.5123e+03 2.2015e+03 4.6473e+03 1.0162e+05 3.4049e+05 4.1263e+03 1.5105e+03
p 1.5633e-03 8.8609e-06 2.8695e-08 1.5460e-04 1.9063e-03 7.6757e-08 1.4157e-09 8.2805e-09 3.2134e-06 –
h 1 1 1 1 1 1 1 1 1 –
f18 fmean 2.6401e+03= 1.8062e+03− 1.0502e+04+ 7.3530e+03= 1.1906e+04+ 9.7595e+03+ 9.7011e+03+ 1.2910e+04+ 1.2623e+04+ 4.5016e+03
std 1.3033e+03 1.6075e+00 9.3906e+03 6.4017e+03 9.4872e+03 8.2475e+03 1.9430e+03 1.2228e+04 1.2074e+04 3.0398e+03
p 1.3517e-01 2.5677e-08 9.0701e-04 8.4194e-03 2.9912e-03 1.9894e-02 1.4157e-09 5.5268e-03 9.8635e-03 –
h 0 1 1 0 1 1 1 1 1 –
Modified collective decision optimization algorithm...

f19 fmean 1.9008e+03+ 1.9017e+03+ 1.9479e+03+ 1.9021e+03+ 1.9030e+03+ 1.9016e+03+ 1.9038e+03+ 1.9057e+03+ 1.9019e+03+ 1.9003e+03
std 4.3458e-01 2.0955e-01 2.8728e+01 6.4244e-01 1.6611e+00 8.8464e-01 8.2302e-01 1.3754e+00 5.2936e-01 2.5003e-01
p 4.6094e-05 1.4157e-09 1.4157e-09 1.4157e-09 9.2880e-09 1.3079e-08 1.4950e-06 1.4157e-09 1.4157e-09 –
h 1 1 1 1 1 1 1 1 1 –
f20 fmean 2.0217e+03= 2.0029e+03− 1.7932e+04+ 2.2214e+03+ 2.7789e+03+ 3.2684e+03+ 1.1834e+04+ 7.3288e+03+ 2.0126e+03= 2.0645e+03
std 1.1996e+01 5.6852e-01 1.2006e+04 4.2277e+02 2.4741e+03 1.6792e+03 4.0787e+03 4.1059e+03 6.9773e+00 1.0899e+02
p 8.4615e-01 1.2215e-04 1.4157e-09 8.8779e-05 1.9505e-04 9.6957e-06 1.4157e-09 5.8547e-09 9.9098e-02 –
h 0 1 1 1 1 1 1 1 0 –
f21 fmean 2.1191e+03= 2.1123e+03= 6.8249e+03+ 4.0597e+03+ 2.2995e+03+ 3.7124e+03+ 1.5595e+05+ 2.9825e+04+ 2.9600e+03+ 2.1601e+03
std 1.3653e+01 5.6042e+00 5.9865e+03 1.8759e+03 1.6685e+02 1.7613e+03 1.0823e+05 2.6722e+04 8.8471e+02 1.2472e+02
p 2.2156e-01 7.8589e-01 1.4157e-09 2.2857e-09 8.1897e-05 4.5414e-07 1.4157e-09 1.4157e-09 3.5800e-08 –
h 0 0 1 1 1 1 1 1 1 –
f22 fmean 2.2127e+03+ 2.2199e+03+ 2.6151e+03+ 2.2684e+03+ 2.2841e+03+ 2.2517e+03+ 2.4030e+03+ 2.2594e+03+ 2.2814e+03+ 2.2004e+03
std 9.3827e+00 4.0541e+00 1.2965e+02 5.7369e+01 8.3971e+01 6.4158e+01 6.5012e+01 4.3711e+01 6.4152e+01 3.0715e-01
p 4.4598e-08 1.4157e-09 1.4157e-09 1.4157e-09 6.5743e-09 5.0255e-07 1.4157e-09 1.4157e-09 1.4157e-09 –
h 1 1 1 1 1 1 1 1 1 –
+ 3 2 6 5 6 6 6 6 5 –
– 0 3 0 0 0 0 0 0 0 –
= 3 1 0 1 0 0 0 0 1 –

+, − and = indicate that the performance of ECDOA is better than, worse than and similar to that of the corresponding algorithm, respectively
Table 4 Simulation results obtained by original swarm intelligence methods on CEC2014 Composition functions

Fun Index CDOA CS BA FA PSO GSO GSA WOA MVO ECDOA

f23 fmean 2.6295e+03= 2.5826e+03− 2.6308e+03+ 2.6295e+03+ 2.6313e+03+ 2.6295e+03+ 2.5673e+03− 2.6256e+03+ 2.6259e+03+ 2.6243e+03
std 4.6412e-13 9.1977e+01 2.5815e+00 7.5324e-08 5.0970e+00 1.7032e-05 6.6011e+01 2.6232e+01 2.3410e-04 2.5892e+01
p 3.3705e-01 3.7237e-03 1.3762e-10 1.3762e-10 1.8491e-04 1.3762e-10 9.8274e-04 3.3260e-09 1.3762e-10 –
h 0 1 1 1 1 1 1 1 1 –
f24 fmean 2.5100e+03− 2.5204e+03+ 2.5947e+03+ 2.5164e+03+ 2.5173e+03+ 2.5291e+03+ 2.6020e+03− 2.5652e+03+ 2.5229e+03+ 2.5125e+03
std 2.3364e+00 5.3254e+00 3.0813e+01 5.1499e+00 7.2403e+00 6.8611e+00 1.0097e+00 2.5871e+01 9.9335e+00 2.8746e+00
p 1.1151e-03 1.4950e-06 1.4157e-09 2.4762e-03 2.1720e-03 2.0288e-09 1.4157e-09 1.4157e-09 1.6471e-06 –
h 1 1 1 1 1 1 1 1 1 –
f25 fmean 2.6428e+03= 2.6664e+03+ 2.7013e+03+ 2.6742e+03+ 2.6912e+03+ 2.6718e+03+ 2.6995e+03+ 2.6929e+03+ 2.6911e+03+ 2.6215e+03
std 3.8029e+01 1.2591e+01 1.0310e+01 3.6527e+01 2.4765e+01 3.1998e+01 5.5649e-01 1.1226e+01 2.4335e+01 7.8910e+00
p 2.2156e-01 1.6401e-08 1.4157e-09 1.5127e-05 2.0536e-08 2.2967e-08 1.3941e-09 1.4144e-09 6.5743e-09 –
h 0 1 1 1 1 1 1 1 1 –
f26 fmean 2.7001e+03+ 2.7002e+03+ 2.7003e+03+ 2.7001e+03= 2.7002e+03+ 2.7003e+03+ 2.7059e+03+ 2.7004e+03+ 2.7001e+03+ 2.7001e+03
std 1.3433e-02 5.5534e-02 1.4793e-01 4.1439e-02 7.7022e-02 1.1188e-01 3.4550e+00 2.0271e-01 3.8456e-02 1.2058e-02
p 9.2880e-09 1.0414e-08 7.6757e-08 6.5289e-02 3.0746e-04 1.9933e-07 1.4157e-09 2.0288e-09 1.2780e-03 –
h 1 1 1 0 1 1 1 1 1 –
f27 fmean 2.8644e+03+ 2.7534e+03+ 3.2928e+03+ 2.9371e+03+ 3.0395e+03+ 2.9546e+03+ 3.2018e+03+ 3.0404e+03+ 3.0147e+03+ 2.7012e+03
std 1.7559e+02 1.3073e+02 1.5059e+02 1.3756e+02 1.5361e+02 1.9313e+02 1.7204e+02 1.9738e+02 1.2528e+02 4.0419e-01
p 7.8561e-03 1.4157e-09 1.4157e-09 4.9750e-08 1.3090e-07 5.2120e-09 1.4157e-09 1.4157e-09 7.3796e-09 –
h 1 1 1 1 1 1 1 1 1 –
f28 fmean 3.1896e+03= 3.1707e+03− 3.6887e+03+ 3.2073e+03= 3.3099e+03+ 3.3639e+03+ 3.8731e+03+ 3.3341e+03+ 3.2241e+03+ 3.2079e+03
std 4.2155e+01 4.2993e+00 2.5766e+02 1.5100e+02 7.4078e+01 1.1861e+02 3.4685e+02 1.0659e+02 6.3343e+01 4.9650e+01
p 1.3017e-01 1.4495e-02 1.4157e-09 9.5188e-02 7.3889e-06 1.9933e-07 2.5677e-08 9.1579e-07 5.0945e-01 –
h 0 1 1 0 1 1 1 1 0 –
f29 fmean 3.3369e+03+ 3.4346e+03+ 3.0922e+06− 3.1533e+05− 8.5299e+04+ 7.3676e+04+ 9.8022e+05+ 4.4222e+03+ 8.7446e+04+ 3.2359e+03
std 7.3456e+01 1.1882e+02 5.1609e+06 9.0837e+05 4.0894e+05 3.5154e+05 4.2585e+06 1.5063e+03 4.2016e+05 1.1227e+02
p 3.5681e-04 5.9096e-05 1.4157e-09 6.7951e-07 6.4154e-05 2.5659e-02 7.1158e-07 4.9750e-08 8.0950e-06 –
h 1 1 1 1 1 1 1 1 1 –
f30 fmean 3.5329e+03= 3.5054e+03= 1.3833e+04+ 3.8355e+03+ 4.1089e+03+ 4.1385e+03+ 6.0327e+03+ 4.7569e+03+ 3.9681e+03+ 3.5710e+03
std 8.7243e+01 3.1614e+01 2.9854e+04 3.3639e+02 4.9543e+02 3.3010e+02 4.6002e+02 7.6015e+02 4.5129e+02 1.5238e+02
p 6.2763e-01 7.2690e-01 1.4157e-09 2.7771e-05 2.6597e-06 7.6757e-08 1.4157e-09 3.2614e-09 3.5302e-06 –
h 0 0 1 1 1 1 1 1 1 –
+ 3 5 7 5 8 8 6 8 8 –
– 1 2 1 1 0 0 2 0 0 –
= 4 1 0 2 0 0 0 0 0 –

+, − and = indicate that the performance of ECDOA is better than, worse than and similar to that of the corresponding algorithm, respectively
Q. Zhang et al.
Table 5 Simulation results obtained by advanced variants of intelligence methods on CEC2014 unimodal functions

Fun Index CLPSO ORCS QPSO AGGSA PSOFIPS CEP PSOcfLocal NSTLBO UPSO ECDOA

f1 fmean 1.7003e+07+ 3.2728e+02− 1.6394e+04+ 1.3255e+06+ 4.1697e+08+ 2.7081e+08+ 5.1231e+08+ 1.1151e+04+ 3.9270e+08+ 3.6970e+03
std 9.6120e+06 1.5065e+02 1.5082e+04 7.9133e+05 1.9880e+08 2.0522e+08 3.4424e+08 1.1149e+04 2.4248e+08 3.4486e+03
Modified collective decision optimization algorithm...

p 1.4157e-09 1.4157e-09 4.6704e-06 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.0415e-03 1.4157e-09 –

h 1 1 1 1 1 1 1 1 1 –
f2 fmean 1.9610e+09+ 2.0011e+02− 1.3498e+03+ 7.4960e+02+ 1.5924e+10+ 1.1955e+10+ 1.5565e+10+ 8.1236e+02+ 1.4518e+10+ 2.6456e+02
std 7.2983e+08 5.2405e-02 1.5561e+03 7.5105e+02 4.7422e+09 3.6436e+09 4.7610e+09 6.5384e+02 5.6961e+09 9.2730e+01
p 1.4157e-09 1.4157e-09 1.9668e-05 2.8342e-02 1.4157e-09 1.4157e-09 1.4157e-09 3.3128e-04 1.4157e-09 –
h 1 1 1 1 1 1 1 1 1 –
f3 fmean 1.2395e+04+ 3.0001e+02+ 5.4290e+02+ 8.5736e+03+ 2.9751e+06+ 3.4780e+04+ 3.7899e+06+ 9.4522e+02+ 3.7249e+06+ 3.0001e+02
std 3.7939e+03 9.6875e-03 2.2390e+02 2.4942e+03 6.5629e+06 9.7155e+03 7.3810e+06 5.6976e+02 4.8760e+06 9.3158e-03
p 1.4157e-09 1.6177e-04 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 –
h 1 1 1 1 1 1 1 1 1 –
+ 3 1 3 3 3 3 3 3 3 –
– 0 2 0 0 0 0 0 0 0 –
= 0 0 0 0 0 0 0 0 0 –

+, − and = indicate that the performance of ECDOA is better than, worse than and similar to that of the corresponding algorithm, respectively
Table 6 Simulation results obtained by advanced variants of intelligence methods on CEC2014 simple multimodal functions

Fun Index CLPSO ORCS QPSO AGGSA PSOFIPS CEP PSOcfLocal NSTLBO UPSO ECDOA

f4 fmean 6.7385e+02+ 4.0019e+02− 4.2143e+02= 4.3927e+02+ 4.4222e+03+ 2.6417e+03= 3.4492e+03+ 4.2390e+02+ 3.3240e+03+ 4.2278e+02
std 1.4072e+02 8.6473e-01 1.6733e+01 8.5249e+00 1.6050e+03 1.2586e+03 1.5388e+03 1.5732e+01 1.6187e+03 1.6372e+01
p 7.4782e-10 1.3399e-06 6.0675e-01 5.2216e-09 7.4782e-10 7.4782e-10 7.4782e-10 3.0754e-03 7.4782e-10 –
h 1 1 0 1 1 1 1 1 1 –
f5 fmean 5.2015e+02+ 5.2018e+02+ 5.2034e+02+ 5.2000e+02+ 5.2105e+02+ 5.2039e+02+ 5.2111e+02+ 5.2034e+02+ 5.2109e+02+ 5.1765e+02
std 5.1620e-02 4.1470e-02 5.4714e-02 9.7007e-04 1.3844e-01 6.8817e-02 1.3013e-01 8.0070e-02 1.3376e-01 6.6541e+00
p 6.8711e-08 2.2777e-09 1.4106e-09 3.8920e-05 1.4106e-09 1.4106e-09 1.4106e-09 1.4106e-09 1.4106e-09 –
h 1 1 1 1 1 1 1 1 1 –
f6 fmean 6.0698e+02+ 6.0435e+02+ 6.0046e+02= 6.0038e+02= 6.1374e+02+ 6.1207e+02+ 6.1397e+02+ 6.0101e+02+ 6.1394e+02+ 6.0055e+02
std 8.6657e-01 4.8126e-01 6.4951e-01 6.8172e-01 1.0576e+00 1.1436e+00 9.6445e-01 6.0030e-01 1.0172e+00 7.8021e-01
p 1.2695e-09 1.2695e-09 8.5214e-01 7.7037e-01 1.2695e-09 1.2695e-09 1.2695e-09 4.2301e-03 1.2695e-09 –
h 1 1 0 0 1 1 1 1 1 –
f7 fmean 7.6289e+02+ 7.0004e+02− 7.0008e+02= 7.0000e+02− 9.4603e+02+ 9.0928e+02+ 9.3115e+02+ 7.0007e+02= 9.6355e+02+ 7.0011e+02
std 2.5623e+01 1.2808e-02 4.7745e-02 2.0479e-03 9.4707e+01 7.2843e+01 5.6624e+01 4.2646e-02 6.5410e+01 8.9782e-02
p 1.4157e-09 1.6711e-04 9.1402e-02 1.8733e-10 1.4157e-09 1.4157e-09 1.4157e-09 5.4746e-02 1.4157e-09 –
h 1 1 0 1 1 1 1 0 1 –
f8 fmean 8.4551e+02+ 8.0664e+02+ 8.0110e+02+ 8.0975e+02+ 9.3750e+02+ 8.8370e+02+ 9.3462e+02+ 8.0482e+02+ 9.2969e+02+ 8.0000e+02
std 8.8794e+00 1.0387e+00 9.2959e-01 3.5989e+00 1.6495e+01 2.1780e+01 1.4108e+01 2.1813e+00 2.2319e+01 0
p 9.7285e-11 9.7285e-11 1.4423e-08 9.5023e-11 9.7285e-11 9.7285e-11 9.7285e-11 9.7285e-11 9.7285e-11 –
h 1 1 1 1 1 1 1 1 1 –
f9 fmean 9.4407e+02+ 9.1800e+02+ 9.0642e+02+ 9.0856e+02+ 1.0318e+03+ 9.8527e+02+ 1.0393e+03+ 9.0605e+02+ 1.0430e+03+ 9.0436e+02
std 6.0147e+00 4.2476e+00 3.0446e+00 2.4371e+00 2.4338e+01 1.6314e+00 2.3286e+01 2.7442e+00 1.4867e+01 1.5226e+00
p 1.3530e-09 1.3530e-09 2.3608e-03 3.4481e-07 1.3530e-09 1.3530e-09 1.3530e-09 9.5359e-11 1.3530e-09 –
h 1 1 1 1 1 1 1 1 1 –
f10 fmean 2.0855e+03+ 1.1417e+03+ 1.0423e+03+ 1.6000e+03+ 3.5106e+03+ 2.5573e+03+ 3.6854e+03+ 1.1148e+03+ 3.5933e+03+ 1.0075e+03
std 1.7747e+02 5.9116e+01 7.1518e+01 2.3027e+02 3.0849e+02 2.7242e+02 3.0090e+02 1.1736e+02 2.4513e+02 6.7555e+00
p 1.4157e-09 1.4157e-09 1.5293e-02 1.4144e-09 1.4157e-09 1.4157e-09 1.4157e-09 4.1024e-07 1.4157e-09 –
h 1 1 1 1 1 1 1 1 1 –
f11 fmean 2.2243e+03+ 1.7022e+03+ 1.3449e+03+ 1.8061e+03+ 3.7617e+03+ 2.8136e+03+ 3.7593e+03+ 1.5792e+03+ 3.8201e+03+ 1.1737e+03
std 1.6583e+02 9.9654e+01 2.0402e+02 2.4041e+02 3.0291e+02 2.0467e+02 2.1820e+02 4.0058e+02 2.5898e+02 8.3210e+01
p 1.4157e-09 1.4157e-09 2.4528e-04 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.1596e-05 1.4157e-09 –
h 1 1 1 1 1 1 1 1 1 –
Q. Zhang et al.
Table 6 (continued)

Fun Index CLPSO ORCS QPSO AGGSA PSOFIPS CEP PSOcfLocal NSTLBO UPSO ECDOA

f12 fmean 1.2005e+03+ 1.2002e+03+ 1.2011e+03+ 1.2000e+03− 1.2051e+03+ 1.2012e+03+ 1.2050e+03+ 1.2010e+03+ 1.2050e+03+ 1.2001e+03
std 1.1466e-01 4.5424e-02 1.8382e-01 2.2835e-07 1.0642e-00 2.1732e-01 1.0655e+00 1.7653e-01 1.2682e+00 8.0427e-02
p 4.1259e-09 1.6516e-05 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 –
h 1 1 1 1 1 1 1 1 1 –
f13 fmean 1.3021e+03+ 1.3002e+03+ 1.3001e+03+ 1.3000e+03− 1.3055e+03+ 1.3048e+03+ 1.3055e+03+ 1.3002e+03+ 1.3056e+03+ 1.3001e+03
std 7.5052e-01 4.3996e-02 3.1611e-02 6.2040e-03 1.1715e+00 9.3084e-01 9.3348e-01 4.3791e-02 1.0969e+00 2.1980e-02
Modified collective decision optimization algorithm...

p 1.4157e-09 4.1259e-09 1.2315e-02 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 2.5742e-09 1.4157e-09 –

h 1 1 1 1 1 1 1 1 1 –
f14 fmean 1.4115e+03+ 1.4001e+03= 1.4001e+03= 1.4004e+03+ 1.4613e+03+ 1.4475e+03+ 1.4676e+03+ 1.4004e+03+ 1.4538e+03= 1.4001e+03
std 4.6133e+00 3.2873e-02 4.8707e-02 6.6035e-02 1.6095e+01 1.3647e+01 1.6557e+01 6.7837e-02 1.6855e+01 3.8856e-02
p 1.4157e-09 1.0415e-01 3.3197e-01 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 –
h 1 0 0 1 1 1 1 1 1 –
f15 fmean 1.5730e+03+ 1.5011e+03= 1.5011e+03= 1.5012e+03= 2.4273e+05+ 1.1201e+05+ 4.5315e+05+ 1.5011e+03+ 3.7804e+05+ 1.5011e+03
std 7.3528e+01 2.2588e-01 4.8193e-01 3.8476e-01 3.2602e+05 9.6583e+04 3.7668e+05 3.6199e-01 3.2955e+05 2.6589e-01
p 1.4157e-09 8.3100e-01 3.9326e-01 8.6137e-01 1.4157e-09 1.4157e-09 1.4157e-09 8.6127e-01 1.4157e-09 –
h 1 0 0 0 1 1 1 0 1 –
f16 fmean 1.6034e+03+ 1.6031e+03+ 1.6019e+03+ 1.6037e+03+ 1.6045e+03+ 1.6038e+03+ 1.6045e+03+ 1.6022e+03+ 1.6045e+03+ 1.6011e+03
std 2.1583e-01 2.0877e-01 5.8592e-01 2.0562e-01 2.2450e-01 2.3625e-01 1.6484e-01 3.8501e-01 1.8603e-01 5.5768e-01
p 1.4157e-09 3.0245e-05 3.0245e-05 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.0585e-07 1.4157e-09 –
h 1 1 1 1 1 1 1 1 1 –
+ 13 9 8 8 13 12 13 12 12 –
− 0 2 0 3 0 0 0 0 0 –
= 0 2 5 2 0 1 0 1 1 –

+, − and = indicate that the performance of ECDOA is better than, worse than and similar to that of the corresponding algorithm, respectively
Table 7 Simulation results obtained by advanced variants of intelligence methods on CEC2014 Hybrid functions

Fun Index CLPSO ORCS QPSO AGGSA PSOFIPS CEP PSOcfLocal NSTLBO UPSO ECDOA

f17 fmean 4.3773e+04+ 1.7625e+03− 3.9603e+03+ 2.2937e+05+ 1.7005e+07+ 2.4640e+06+ 1.9878e+07+ 2.6529e+03+ 2.1463e+07+ 2.5816e+03
std 4.4256e+04 1.7687e+01 2.3613e+03 1.0001e+05 1.0923e+07 3.0508e+06 1.7981e+07 5.9473e+02 1.6430e+07 1.5105e+03
p 3.2614e-09 3.3440e-07 1.0432e-02 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.9894e-02 1.4157e-09 –
h 1 1 1 1 1 1 1 1 1 –
f18 fmean 2.7128e+03= 1.8039e+03− 1.2733e+04+ 8.4218e+03+ 2.1240e+08+ 2.6968e+07+ 2.8372e+08+ 4.7210e+03= 2.5193e+08+ 4.5016e+03
std 7.9941e+02 8.7991e-01 7.5248e+03 2.2699e+03 2.0231e+08 8.2475e+03 1.9920e+08 2.7392e+03 2.5758e+08 3.0398e+03
p 1.7440e-01 2.2967e-08 5.9096e-05 1.9668e-05 1.4157e-09 1.4157e-09 1.4157e-09 4.6093e-01 1.4157e-09 –
h 0 1 1 1 1 1 1 0 1 –
f19 fmean 1.9067e+03+ 1.9019e+03+ 1.9004e+03= 1.9031e+03+ 2.0165e+03+ 1.9459e+03+ 2.0055e+03+ 1.9011e+03+ 2.0108e+03+ 1.9003e+03
std 2.3034e+00 2.2786e-01 4.5044e-01 6.7085e-01 5.8816e+01 3.4829e+01 5.5046e+01 4.4733e-01 5.7260e+01 2.5003e-01
p 1.4157e-09 1.4157e-09 2.0034e-01 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4549e-07 1.4157e-09 –
h 1 1 0 1 1 1 1 1 1 –
f20 fmean 4.0585e+03+ 2.0026e+03− 2.3502e+03+ 7.5347e+03+ 2.2632e+07+ 3.3961e+05+ 2.2401e+07+ 2.2380e+03+ 2.1420e+07= 2.0645e+03
std 2.1209e+03 6.0224e-01 5.8329e+02 1.8039e+03 2.6171e+07 1.3787e+06 3.0509e+07 1.4637e+02 3.0777e+07 1.0899e+02
p 3.6690e-09 3.8987e-05 7.3927e-06 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 2.1982e-06 1.4157e-09 –
h 1 1 1 1 1 1 1 1 1 –
f21 fmean 1.1636e+04+ 2.1095e+03= 2.1746e+03+ 1.6410e+04+ 8.3054e+06+ 2.0209e+05+ 5.5267e+05+ 2.2564e+03+ 1.0517e+07+ 2.1601e+03
std 9.0095e+03 5.7319e+00 7.2377e+01 3.2010e+04 8.2944e+06 2.8454e+05 5.9597e+05 6.1482e+01 1.1585e+07 1.2472e+02
p 1.4157e-09 4.1512e-01 3.1856e-03 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 2.9240e-06 1.4157e-09 –
h 1 0 1 1 1 1 1 1 1 –
f22 fmean 2.3067e+03+ 2.2201e+03+ 2.2148e+03+ 2.3481e+03+ 2.8854e+03+ 2.4376e+03+ 2.9988e+03+ 2.2184e+03+ 2.9949e+03+ 2.2004e+03
std 4.8590e+01 3.2451e+00 1.0163e+01 9.6967e+00 2.1543e+02 9.0564e+01 1.7953e+02 8.7531e+00 2.4482e+02 3.0715e-01
p 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 6.5743e-09 1.4157e-09 –
h 1 1 1 1 1 1 1 1 1 –
+ 5 2 5 6 6 6 6 5 5 –
− 0 3 0 0 0 0 0 0 0 –
= 1 1 1 0 0 0 0 1 1 –

+, − and = indicate that the performance of ECDOA is better than, worse than and similar to that of the corresponding algorithm, respectively
Q. Zhang et al.
Table 8 Simulation results obtained by advanced variants of intelligence methods on CEC2014 Composition functions

Fun Index CLPSO ORCS QPSO AGGSA PSOFIPS CEP PSOcfLocal NSTLBO UPSO ECDOA

f23 fmean 2.6512e+03+ 2.6043e+03= 2.6295e+03+ 2.6295e+03+ 2.9355e+03+ 2.8638e+03+ 3.0777e+03+ 2.6295e+03+ 2.9822e+03+ 2.6243e+03
std 9.7504e+00 7.0048e+01 5.4126e-13 5.8708e-13 1.7434e+02 1.3825e+02 2.0383e+02 9.7797e-13 1.5204e+02 2.5892e+01
p 1.3762e-10 5.0560e-02 4.7667e-04 1.3186e-02 1.3762e-10 1.3762e-10 1.3762e-10 7.8574e-11 1.3762e-10 –
h 1 0 1 1 1 1 1 1 1 –
f24 fmean 2.5593e+03+ 2.5225e+03+ 2.5130e+03= 2.5940e+03+ 2.6480e+03+ 2.6193e+03+ 2.6527e+03+ 2.5150e+03+ 2.6467e+03+ 2.5125e+03
std 1.1758e+01 4.4291e+00 3.4947e+00 2.3158e+01 1.7153e+01 1.3802e+01 1.3947e+01 3.8755e+00 1.4471e+01 2.8746e+00
p 1.4157e-09 5.8547e-09 5.7365e-01 1.4648e-08 1.4157e-09 1.4157e-09 1.4157e-09 2.2048e-02 1.4157e-09 –
h 1 1 0 1 1 1 1 1 1 –
f25 fmean 2.6841e+03+ 2.6504e+03+ 2.6741e+03+ 2.6980e+03+ 2.7243e+03+ 2.7102e+03+ 2.7274e+03+ 2.6487e+03+ 2.7274e+03+ 2.6215e+03
std 1.0923e+01 9.8402e+00 3.6561e+01 5.6678e+00 1.0926e+01 1.1280e+01 1.0449e+01 3.0616e+01 1.2927e+01 7.8910e+00
p 1.4157e-09 5.2120e-09 5.4418e-05 1.2661e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.8057e-04 1.4157e-09 –
h
Modified collective decision optimization algorithm...

1 1 1 1 1 1 1 1 1 –
f26 fmean 2.7024e+03+ 2.7002e+03+ 2.7001e+03= 2.7011e+03= 2.7083e+03+ 2.7065e+03+ 2.7125e+03+ 2.7001e+03+ 2.7142e+03+ 2.7001e+03
std 9.8818e-01 3.4822e-02 2.7916e-02 1.2847e+00 7.2537e+00 4.7688e+00 2.5168e+01 3.6333e-02 2.4048e+01 1.2058e-02
p 1.4157e-09 2.0288e-09 7.4151e-01 1.0585e-07 1.4157e-09 1.4157e-09 1.4157e-09 1.7925e-02 1.4157e-09 –
h 1 1 0 1 1 1 1 1 1 –
f27 fmean 3.0539e+03+ 2.7684e+03+ 2.9095e+03+ 3.0382e+03+ 3.3886e+03+ 3.2827e+03+ 3.4006e+03+ 2.9285e+03+ 3.3746e+03+ 2.7012e+03
std 1.3345e+02 1.4516e+02 1.7591e+02 4.3856e+01 1.1039e+02 1.5828e+02 1.1806e+02 1.5643e+02 1.2391e+02 4.0419e-01
p 1.4157e-09 1.4157e-09 9.1579e-07 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 –
h 1 1 1 1 1 1 1 1 1 –
f28 fmean 3.2041e+03= 3.1700e+03= 3.2433e+03+ 3.0631e+03− 4.6991e+03+ 4.2188e+03+ 4.6761e+03+ 3.1776e+03= 4.7003e+03+ 3.2079e+03
std 4.8631e+01 5.7547e+00 6.4087e+01 2.5171e+02 2.7272e+02 2.6125e+02 4.3356e+02 4.9942e+00 4.6409e+02 4.9650e+01
p 9.3814e-01 5.2345e-02 3.6126e-02 1.5293e-02 1.4157e-09 1.4157e-09 1.4157e-09 4.7282e-01 1.4157e-09 –
h 0 0 1 1 1 1 1 0 1 –
f29 fmean 3.4801e+03+ 3.1252e+03− 7.2263e+04+ 7.2892e+05+ 4.0223e+07+ 2.9387e+07+ 4.9073e+07+ 3.5037e+03+ 4.6912e+07+ 3.2359e+03
std 3.1163e+02 1.4263e+01 3.4483e+05 2.5173e+06 2.9529e+07 1.8964e+07 2.7236e+07 1.3541e+02 3.1275e+07 1.1227e+02
p 3.0746e-04 5.6183e-06 1.7926e-02 5.1446e-04 1.4157e-09 1.4157e-09 1.4157e-09 6.8909e-08 1.4157e-09 –
h 1 1 1 1 1 1 1 1 1 –
f30 fmean 3.8860e+03+ 3.5303e+03= 3.8453e+03+ 5.5470e+03+ 6.7659e+05+ 7.7554e+04+ 5.1243e+05+ 3.5444e+03+ 6.6921e+05+ 3.5710e+03
std 3.3245e+02 3.1321e+01 2.5143e+02 4.4915e+02 6.3124e+05 5.6294e+04 5.5809e+05 5.0571e+01 5.6455e+05 1.5238e+02
p 5.9096e-05 7.8590e-01 4.6704e-06 1.4157e-09 1.4157e-09 1.4157e-09 1.4157e-09 2.9475e-01 1.4157e-09 –
h 1 0 1 1 1 1 1 0 1 –
+ 7 5 6 6 8 8 8 7 8 –
− 0 1 0 1 0 0 0 0 0 –
= 1 2 2 1 0 0 0 1 0 –

+, − and = indicate that the performance of ECDOA is better than, worse than and similar to that of the corresponding algorithm, respectively
 Q. Zhang et al.

acceleration coefficients c1 and c2 are both set to 1.49445. ability compared to both CS and ORCS. In CS and ORCS,
The parameters of the other methods agree well with the the exploitation is performed using a local random walk
original literatures. with two parameters drawn from a Gaussian distribution and
a Levy distribution, which ensures that the move is not too
4.3 Simulation results far away and that it does not focus on the local search. The
modified method adopts the standard uniform distribution
In this section, Tables 1, 2, 3, 4, 5, 6, 7 and 8 summarize as the local search strategy. The comparison between these
the statistical results obtained using each method. The last methods shows that the former has a certain advantage. The
three rows of each table report the comparison results. In potential performance of the convergence rate and accuracy
addition, Figs. 3, 4, 5 and 6 list some convergence curves of can be observed in Fig. 3.
the test functions for comparing the convergence rate of the According to the results of Tables 2 and 6, it is appar-
improved methodology with its competitors. ent that ECDOA performs significantly better than the
For the three CEC2014 unimodal functions, the results, other techniques for most CEC2014 simple multimodal
which are recorded in Tables 1 and 5, provide information to functions. These results are attributed to the powerful explo-
statistically demonstrate that the ECDOA outperforms most ration capability of the technique. The multi-step position-
of the standard swarm algorithms as well as the advanced selection search mechanism, which is the main design strat-
variants of other heuristic paradigms in all the unimodal egy, is applied to ensure that each search agent can randomly
cases. However, for f1 , it performs slightly worse than move to other areas, and yield promising solutions that are
CDOA, CS, and ORCS. According to p-value, there is sufficiently diverse. In addition, one important concern is
no significant difference between CDOA and ECDOA on that ECDOA employs distinctive evolution orders, which
f2 . The main reason why CS and ORCS are superior to are formed by different operators that are randomly selected
ECDOA is that ECDOA has slightly weaker exploitation from an external archive in each generation. This not only

Fig. 3 Evolution curves of the Convergence process of f2 Convergence process of f3

mean function error values 11 6
derived from the algorithms CDOA CDOA
10 CS 5.5 CS
versus the number of FES on BA BA
unimodal functions 9 FA 5 FA
PSO PSO
8 GSO GSO
4.5
Mean value

Mean value

GSA GSA
7
WOA WOA
4
MVO MVO
6
ECDOA ECDOA
3.5
5
3
4

3 2.5

2 2
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
FES FES
Convergence process of f2 Convergence process of f3
11 7
CLPSO CLPSO
10 ORCS 6.5 ORCS
QPSO 6 QPSO
9 AGGSA AGGSA
PSOFIPS 5.5 PSOFIPS
8 CEP CEP
Mean value

Mean value

PSOofLocal 5 PSOofLocal
7
NSTLBO NSTLBO
4.5
UPSO UPSO
6
ECDOA 4 ECDOA
5
3.5
4 3
3 2.5

2 2
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
FES FES
Modified collective decision optimization algorithm...

Fig. 4 Evolution curves of the Convergence process of f4 Convergence process of f6

mean function error values 3.6 2.788
CDOA CDOA
derived from the algorithms 3.5 CS 2.787 CS
versus the number of FES on BA BA
3.4 2.786
multimodal functions FA FA
3.3 PSO 2.785 PSO
GSO GSO

Mean value

Mean value
3.2 GSA 2.784 GSA
WOA WOA
3.1 2.783
MVO MVO
3 ECDOA 2.782 ECDOA

2.9 2.781

2.8 2.78

2.7 2.779

2.6 2.778
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
FES FES
Convergence process of f10 Convergence process of f11
3.5 3.8
CDOA CDOA
3.45 CS CS
3.7
3.4 BA BA
FA 3.6 FA
3.35 PSO PSO
GSO 3.5 GSO
Mean value

Mean value
3.3 GSA GSA
WOA WOA
3.25 3.4
MVO MVO
3.2 ECDOA ECDOA
3.3
3.15
3.2
3.1
3.1
3.05

3 3
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
FES FES
Convergence process of f8 Convergence process of f10
2.99 3.8
CLPSO CLPSO
2.98 ORCS 3.7 ORCS
QPSO QPSO
2.97 AGGSA AGGSA
3.6
PSOFIPS PSOFIPS
2.96 CEP CEP
3.5
Mean value

Mean value

PSOofLocal PSOofLocal
2.95
NSTLBO NSTLBO
3.4
UPSO UPSO
2.94
ECDOA ECDOA
3.3
2.93
3.2
2.92

2.91 3.1

2.9 3
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
FES FES
Convergence process of f11 Convergence process of f16
3.8 3.2054
CLPSO CLPSO
ORCS 3.2053 ORCS
3.7
QPSO 3.2052 QPSO
3.6 AGGSA AGGSA
PSOFIPS 3.2051 PSOFIPS
3.5 CEP CEP
Mean value

Mean value

PSOofLocal 3.205 PSOofLocal

NSTLBO NSTLBO
3.4 3.2049
UPSO UPSO
ECDOA 3.2048 ECDOA
3.3
3.2047
3.2
3.2046
3.1
3.2045

3 3.2044
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
FES FES
 Q. Zhang et al.

Fig. 5 Evolution curves of the Convergence process of f17 Convergence process of f22
mean function error values 7 3.52
CDOA CDOA
derived from the algorithms CS 3.5 CS
6.5
versus the number of FES on BA BA
hybrid functions FA 3.48 FA
6
PSO PSO
GSO 3.46 GSO
5.5

Mean value

Mean value
GSA GSA
3.44
WOA WOA
5
MVO MVO
3.42
ECDOA ECDOA
4.5
3.4
4
3.38

3.5 3.36

3 3.34
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
FES FES
Convergence process of f21 Convergence process of f22
7.5 3.55
CLPSO CLPSO
7 ORCS ORCS
QPSO 3.5 QPSO
6.5 AGGSA AGGSA
PSOFIPS PSOFIPS
6 CEP CEP
Mean value

Mean value
PSOofLocal 3.45 PSOofLocal
5.5
NSTLBO NSTLBO
UPSO UPSO
5
ECDOA 3.4 ECDOA
4.5

4 3.35
3.5

3 3.3
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
FES FES

Fig. 6 Evolution curves of the Convergence process of f25 Convergence process of f27
mean function error values 3.436 3.6
CDOA CDOA
derived from the algorithms 3.434 CS 3.58 CS
versus the number of FES on BA BA
3.432 3.56
composition functions FA FA
PSO PSO
3.43 GSO 3.54 GSO
Mean value

Mean value

GSA GSA
3.428 3.52
WOA WOA
MVO MVO
3.426 3.5
ECDOA ECDOA
3.424 3.48

3.422 3.46

3.42 3.44

3.418 3.42
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
FES FES
Convergence process of f24 Convergence process of f25
3.425 3.436
CLPSO CLPSO
ORCS 3.434 ORCS
3.42 QPSO QPSO
AGGSA 3.432 AGGSA
PSOFIPS PSOFIPS
CEP 3.43 CEP
Mean value

Mean value

3.415 PSOofLocal PSOofLocal

3.428
NSTLBO NSTLBO
UPSO UPSO
3.426
3.41 ECDOA ECDOA
3.424

3.405 3.422

3.42

3.4 3.418
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
FES FES
Modified collective decision optimization algorithm...

A CS perform slightly better than ECDOA for f17 , f18 , and

f20 ; however, based on the other comparison paradigms,
Candidate node
the improved algorithm is statistically superior. It can be
concluded that ECDOA is able to provide competitive
results. The convergence behaviour of all the methods for
functions f17 , f21 , and f22 can be observed in Fig. 5.
Tables 4 and 8 list the results obtained by all method-
ologies for the composition functions. It is obvious that
ECDOA can generate satisfactory outcomes. CS produces
the best solutions for function f23 and f28 , while BA and
ORCS perform slightly better in solving f29 . In addition, the
results obtained by the original version for four test func-
UAV position
tions are almost similar to those obtained by the extended
B version. ECDOA is superior to CDOA for three out of
Fig. 7 Typical UAV combating field model in three-dimensional seven problems, because the construction of a new evolu-
scenario tion sequence strategy and the large mutation dimension
prevents it from being trapped in the local optima. From the
convergence curves of Fig. 6, it can be seen that ECDOA
provides more randomness and increases the number of is able to provide better results for composition functions
selectable solutions at each iteration, but also improves the compared to the other algorithms.
probability of escaping from poor local optima and locating
a good near-global optimum. From Fig. 4, it can be con-
cluded that ECDOA has a better convergence rate compared 5 Application of ECDOA to UAV trajectory
to its competitors for most typical functions. In addition, it planning
can be seen that FA and ORCS perform better than ECDOA
for f6 and f4 , respectively, and that AGGSA provides the The main purpose of trajectory planning is to find the opti-
best solutions for f7 , f12 , and f13 . mal flight path for an unmanned aerial vehicle (UAV). The
Unlike the two groups discussed earlier, hybrid func- calculated flight route not only ensures that the UAV avoids
tions are much more complex, because their variables are enemy threats, but also ensures that the predetermined mis-
randomly divided into subcomponents and different basic sion is completed successfully. The detailed mathematical
functions are used for different subcomponents. Therefore, model for path planning is described as follows [48].
it is difficult to find the global optimum of these func-
tions. However, Tables 3 and 7 clearly show that ECDOA 5.1 Region and path model
yields better results than the majority of comparable strate-
gies in most cases. Compared to the original CDOA, the As suggested in [49], the task region and threat circum-
modified version performs better in three cases and has stances must be modelled before planning the flight path
similar performance in three cases, which indicates that of the UAV. The typical UAV field combat model in a
the improved strategy has certain advantages. ORCS and three-dimensional scenario is shown in Fig. 7.

Fig. 8 The task area can be 100 100

divided into two-dimensional
mesh 90 90

80 80

70 70

60 60
Y
Y

50 50

40 40

30 30

20 20

10 10
0 20 40 60 80 0 20 40 60 80
X X
 Q. Zhang et al.

Table 12 The settings of threat environment with six threat points

9/10
Start point Goal point q Threat center Radius Grade
k -1
[10, 10] [55, 100] 0.5 [45, 50] 10 2
5/10
[12, 40] 10 10
3/10 [32, 68] 8 1
Threat k [36, 26] 12 2
7/10
[55, 80] 9 3
[27, 46] 6 4

In order to simplify the path-planning problem, the task

k+1 area can be divided into a two-dimensional mesh, as shown
1/10
in Fig. 8. The starting point is defined as A and the tar-
Fig. 9 Computation of threat cost get point is defined as B. The dots represent threats, such
as radar, artillery, missile etc. The scope of the threats is
depicted in the form of a circle. Within the circle, the UAV
will be susceptible to a threat with a certain probability rel-
ative to its distance from the threat centre. The objective of
path planning is to optimize the flight route between A(∗)
and B() while avoiding the threat areas. In this way, sup-
Table 9 The settings of threat environment with five threat points
pose that the segment AB is divided into m equal portions.
Start point Goal point q Threat center Radius Grade Then, the flight path from the starting point to the goal point
can be depicted by the equation given below.
[10, 10] [55, 100] 0.5 [45, 50] 10 2
[12, 40] 10 10
P ath = {A, L1 (x1 , y1 ), L2 (x2 , y2 ),
[32, 68] 8 1 ..., Lm (xm , ym ), B} (12)
[36, 26] 12 2 where Li (xi , yi ) is the vertical direction of the ith node.
[55, 80] 9 3
5.2 Threat price and evaluation model

The exact solution is to calculate the threat cost at several

positions along an edge. This study considers five loca-
tions along each edge to generate the threat cost, as shown

Table 10 Simulation results obtained by original swarm intelligence methods on UAV with five threat points

Index CDOA CS BA FA PSO GSO GSA WOA MVO ECDOA

Optimal length 51.0309 52.5822 52.8891 51.1884 52.2590 52.2730 50.7445 50.7598 52.9425 50.7225
std 0.40938 0.56677 2.79951 2.72567 2.74518 2.59820 0.16419 0.16927 2.71694 0.49259

Table 11 Simulation results obtained by advanced variants of intelligence methods on UAV with five threat points

Index CLPSO ORCS QPSO AGGSA PSOFIPS CEP PSOcfLocal NSTLBO UPSO ECDOA

Optimal length 57.2726 54.2939 56.9463 51.1493 71.8416 65.9376 68.1173 51.0000 69.8062 50.7225
std 3.63515 1.02756 9.33831 0.99566 4.70794 3.22001 4.65381 0.25699 4.87420 0.49259
Modified collective decision optimization algorithm...

Convergence process of UAV with five threat points in Fig. 9. The threat price of kth edge is expressed by the
1.86
CDOA
formula [50, 51].
1.84 CS
Li 
Nt
BA 1 1 1 1 1
1.82 FA W= tk ( 4 + 4 + 4 + 4 + 4 )
PSO 5 d0.1,i,k d0.3,i,k d0.5,i,k d0.7,i,k d0.9,i,k
k=1
1.8 GSO
(13)
Mean value

GSA
WOA
1.78
MVO where Nt denotes the number of threat areas, Li is the ith
ECDOA
1.76 sub-path length, d0.1,i,k is the distance from the 1/10 point
on the ith edge to the kth threat, and tk is the threat level of
1.74
the kth threat.
1.72 The path planning evaluation model is confirmed to min-
imize the cost incurred by the UAV for completing the task.
1.7
0 0.5 1 1.5 2 In this study, the total cost, which is the sum of the threat
FES 4
x 10 and fuel costs, is calculated using the following equation:
Convergence process of UAV with five threat points
1.88 MinJ = q × Jt + (1 − q) × Jf (14)
CLPSO
1.86 ORCS where q is a weighing coefficient between 0 and 1, Jt and
QPSO
1.84 AGGSA
Jf denote the threat price of each node and the fuel costs of
PSOFIPS the UAV, respectively.
1.82 CEP
Mean value

PSOofLocal
1.8
NSTLBO 5.3 Simulation results
1.78 UPSO
ECDOA
1.76
Twenty-five independent runs were conducted on the same
platform with 20,000 function evaluations as the termina-
1.74
tion criterion. The compared algorithms were required to
1.72 have a valid basis for comparison. The parameters of all the
1.7 techniques are the same as those mentioned in Section 4.
0 0.5 1 1.5 2
FES 4
The settings of the threat environment, which are based on
x 10
the suggestions in [52], are summarized in the following
Fig. 10 Convergence curves comparison between ECDOA and its Table 9.
peers From the simulation results recorded in Tables 10 and
11, it can be observed that ECDOA is able to generate the
best output (50.7225) in complicated combat circumstances.
Figure 10 compares the convergence curves of ECDOA with
Optimal flight route of UAV with five threat points those of some advanced version paradigms. From the con-
100 vergence graphs, it can be seen that WOA, AGGSA, and
90 GSA perform better than the other optimization techniques
before the half of the termination criteria. However, at the
80
end of the generation, ECDOA provides better solutions.
70 Therefore, it can be concluded that even though the modified
60 method shows weak performance in approximating better
Y

50
solutions at the initial stage, it provides satisfactory results
at the end of the termination conditions. Figure 11 shows
40
the optimal planning route in a two-dimensional space.
30 Table 12 summarizes the parameter settings of an envi-
20
ronment with six threats. Tables 13 and 14 list the statistical
ECDOA results, which include the shortest length and standard devi-
10
0 20 40 60 80 ation. The optimal length calculated by the ECDOA is
X 52.2518. Figures 12 and 13 illustrate the evolution curves
Fig. 11 Trajectory planning result by ECDOA and optimal flight path, respectively. It can be seen that
 Q. Zhang et al.

Table 13 Simulation results obtained by original swarm intelligence methods on UAV with six threat points

Index CDOA CS BA FA PSO GSO GSA WOA MVO ECDOA

Optimal length 53.3451 55.7276 56.7199 55.9527 55.9989 56.9879 55.6300 57.6526 55.9626 52.2518
std 2.69555 1.71078 4.94943 4.93407 4.14469 3.76562 4.76536 5.31323 4.23700 1.17719

Table 14 Simulation results obtained by advanced variants of intelligence methods on UAV with six threat points

Index CLPSO ORCS QPSO AGGSA PSOFIPS CEP PSOcfLocal NSTLBO UPSO ECDOA

Optimal length 69.1199 57.3281 62.2741 55.2964 79.7654 73.6836 78.5079 53.8561 81.2567 52.2518
std 5.78334 1.13481 12.6408 6.17991 7.08512 5.61289 8.24321 2.71115 9.63042 1.17719

ECDOA is able to determine flight routes that are more fea- These results suggest that the proposed strategy can pro-
sible faster than its peers, when all the threat regions are vide higher accuracy in real applications with unknown
considered. These results also demonstrate that the modified and challenging search spaces due to the following reasons.
methodology can be regarded as a new and promising tool Firstly, ECDOA intrinsically benefits from high exploration
for path planning in actual tasks. and local optima avoidance, because it is also a population-
based methodology. This ensures that the optimization
techniques will avoid a large number of local solutions. Sec-
ondly, ECDOA can smoothly achieve a good transforma-
Convergence process of UAV with six threat points
2 tion between exploration and exploitation on the multi-step
CDOA position-selected scheme and the new sequence generation
CS
1.95 BA strategy, which provides different operator orders for every
FA agent in each generation. This method is highly suitable
PSO
1.9 GSO
for exploring promising areas extensively and for achiev-
Mean value

GSA ing convergence to the most effective region in the final

WOA
1.85
MVO
stages of evolution. Finally, the best agent obtained thus
ECDOA far is obliged to update its position using the local random
1.8 walk and select the best one as the destination point. Hence,
the candidate solutions are always improved by trending
1.75 towards the best areas of the search spaces over the course
of generations.
1.7
0 0.5 1 1.5 2
FES x 10
4

Convergence process of UAV with six threat points

2
Optimal flight route of UAV with six threat points
CLPSO
100
ORCS
1.95 QPSO 90
AGGSA
PSOFIPS 80
1.9 CEP
Mean value

PSOofLocal 70
NSTLBO
1.85
UPSO 60
ECDOA
Y

1.8 50

40
1.75
30

1.7 20
0 0.5 1 1.5 2 ECDOA
FES 4 10
x 10 0 20 40 60 80
X
Fig. 12 Convergence curves comparison between ECDOA and its
peers Fig. 13 Trajectory planning result by ECDOA
Modified collective decision optimization algorithm...

6 Conclusion was performed on a comprehensive set of benchmark func-

tions. The route planning of the UAV indicates that the
This study proposes an extended version of a collective improved technique performs more effectively and accu-
decision optimization algorithm (ECDOA) to solve the rately than other state-of-the-art intelligent optimization
path-planning problem of UAVs. In ECDOA, four operators methods. Therefore, this method can be recommended as a
are initially stored in an external archive, and then each promising tool for performing different optimization tasks
search agent randomly selects at least one operator from in various real-world applications.
this archive and combines it with an innovation operator
to construct a new evolution congress or sequence using a Acknowledgments The authors express their sincere thanks to Dr.
multi-step position-selected scheme. In this approach, each Xin-she Yang for providing the codes. The authors would also like
particle can utilize different calculation orders by freely to thank Dr. Q.Y. Jiang and the anonymous reviewers for their free-
handed assistance and construction suggestions. This work is partly
organizing distinct operators in each iteration to gener-
supported by the National Natural Science Foundation of China under
ate more promising or diverse candidate solutions. These Project code(61075032)and the Anhui Provincial Natural Science
solutions are then applied to explore and exploit the solu- Foundation under Project code (J2014AKZR0055).
tion circumstance effectively. This improved idea not only
broadens horizons, but also creates a novel solution method-
ology or pathway for future work. In order to evaluate the Appendix: The brief descriptions of the CEC2014
performance of ECDOA, an extensive comparative study benchmark functions

Test function Types Optimum

f1 : Rotated High Conditioned Elliptic Function Unimodal 100

f2 : Rotated Bent Cigar Function 200
f3 : Rotated Discus Function 300
f4 : Shifted and Rotated Rosenbrock’s Function Simple Multimodal 400
f5 : Shifted and Rotated Ackley’s Function 500
f6 : Shifted and Rotated Weierstrass Function 600
f7 : Shifted and Rotated Griewanks Function 700
f8 : Shifted Rastrigins Function 800
f9 : Shifted and Rotated Rastrigins Function 900
f10 : Shifted Schwefels Function 1000
f11 : Shifted and Rotated Schwefels Function 1100
f12 : Shifted and Rotated Katsuura Function 1200
f13 : Shifted and Rotated HappyCat Function 1300
f14 : Shifted and Rotated HGBat Function 1400
f15 : Shifted and Rotated Expanded Griewank’s plus Rosenbrock’s Function 1500
f16 : Shifted and Rotated Expanded Scaffer’s F6 Function 1600
f17 : Hybrid Function 1 (N=3) Hybrid 1700
f18 : Hybrid Function 2 (N=3) 1800
f19 : Hybrid Function 3 (N=4) 1900
f20 : Hybrid Function 4 (N=4) 2000
f21 : Hybrid Function 5 (N=5) 2100
f22 : Hybrid Function 6 (N=5) 2200
f23 : Composition Function 1 (N=5) Composition 2300
f24 : Composition Function 2 (N=3) 2400
f25 : Composition Function 3 (N=3) 2500
f26 : Composition Function 4 (N=5) 2600
f27 : Composition Function 5 (N=5) 2700
f28 : Composition Function 6 (N=5) 2800
f29 : Composition Function 7 (N=3) 2900
f30 : Composition Function 8 (N=3) 3000
 Q. Zhang et al.

References 25. Nitzan S, Paroush J (1985) Collective decision making: an eco-

nomic outlook. Cambridge University press, UK, p 1985
26. Karotkin D, Nitzan S (1997) On two properties of the marginal
1. Ye W, Fan HD (2007) Research on mission planning system key
contribution of individual decisional skills. Math Soc Sci
techniques of UCAV. J Naval Aeronaut Eng Inst 22(2):201–207
34(1):29–36
2. Zheng CW, Li L, Xu FJ (2005) Evolutionary route planner for
27. Barron G, Erev I (2003) Small feedback-based decisions and their
unmanned air vehicles. IEEE T Robot 21(4):609–620
limited correspondence to description-based decisions. J Behav
3. Wang YX, Chen ZJ (1999) Genetic algorithms (GA) based flight
Decis Making 16(3):215–233
path planning with constraints. J Beijing Univ Aeronaut Astronaut
28. Aldag RJ, Fuller SR (1993) Beyond fiasco: a reappraisal of the
(China) 25(3):355–358
groupthink phenomenon and a new model of group decision
4. Ma GJ, Duan HB, Liu SQ (2007) Improved ant colony algorithm
processes. Psychol Bull 113(3):533–552
for global optimal trajectory planning of UAV under complex
29. Janis IL (1974) Victims of groupthink: a psychological study of
environment. Int J Comput Sci Appl 4(3):57–68
foreign-policy decision and fiascoes. J Polit 36(1):218–220
5. McLain TW, Beard RW (2000) Trajectory planning for coordi-
30. Mohamed A, Wiebe F (1996) Toward a process theory of group-
nated rendezvous of unmanned air vehicles. In: AIAA Guidance,
think. Small Group Res 27(3):416–430
navigation, and control conference and exhibit, pp 4369-4372
31. Nahavandi A (2006) The art and science of leadership. USA, p
6. Yang XS (2010) Nature-inspired metaheuristic algorithms, 2nd ed.
2006
Luniver Press
32. Hoy WK, Tarter CJ (2010) Swift and smart decision making:
7. Sadollah A, Bahreininejad A, Eskandar H, Hamdi M (2013) Mine
heuristics that work. Int J Educ Manag 24(4):351–358
blast algorithm: a new population based algorithm for solving con-
33. McHugh KA, Yammarino FJ, Dionne SD, Serban A, Sayama H,
strained engineering optimization problems. Appl Soft Comput
Chatterjee S (2016) Collective decision making, leadership, and
13(5):2592–2612
collective intelligence: tests with agent-based simulations and a
8. Lee KS, Geem ZW (2005) A new meta-heuristic algorithm for
Field study. Leadersh Q 27(2):218–241
continuous engineering optimization: harmony search theory and
34. Frederic A, Patrick B, Sven C, Patrick H (2006) Creativity and
practice. Comput Methods Appl Mech Eng 194(36):3902–3933
innovation in decision making and decision support. London, p
9. Holland JH (1975) Adaptation in natural and artificial systems.
2006
University of Michigan Press, The MIT Press, London, p 1975
35. Gutierrez E, Olundh Sandstrom G, Janhager J, Ritzen S (2008)
10. Mirjalili S, Mirjalili SM, Lewis A (2014) Let a biogeograph-based
Innovation and decision making: understanding selection and pri-
optimizer train your Multi-layer perceptron. Inform Sci 269:188–
oritization of development projects. In: The proceedings of the
209
2008 IEEE international conference on management of innovation
11. Askarzadeh A (2014) Bird mating optimizer: an optimization
and technology, ICMIT’08. IEEE, pp 333–338
algorithm inspired by bird mating strategies. Commun Nonlinear
36. Du J, Love James H, Roper S. (2007) The innovation decision: an
Sci Numer Simul 19(4):1213–1228
economic analysis. Technovation 27:766–773
12. Eberhart RC, Kennedy J (1995) A new optimizer using particle
37. Liang J, Qu BY, Suganthan PN (2013) Problem definitions
swarm theory. In: Proceedings of the sixth international sympo-
and evaluation criteria for the CEC2014 special session and
sium on micro machine and human science. IEEE, pp 39–43
competition on single objective real-parameter numerical opti-
13. He S, Wu QH, Saunders JR (2009) Group search optimizer: an
mization Zhengzhou University, Zhengzhou, China and Nanyang
optimization algorithm inspired by animal search behavior. IEEE
Technological University, Singapore Special Session Competition
Trans Evol Comput 13(5):973–990
on Real-Parameter Single Objective, (Expensive) Optimization.
14. Yang XS, Deb S (2010) Engineering optimization by cuckoo
Technical Report
search. Int J Math Modell Numer Optim 1(4):330–343
38. Wang Y, Cai ZX, Zhang QF (2011) Differential evolution with
15. Yang XS (2012) Bat algorithm: a novel approach for global
composite trial vector generation strategies and control parame-
engineering optimization. Eng Comput 29(5):464–493
ters. IEEE Trans Evol Comput 15(1):55–66
16. Mirjalili S, Lewis A (2016) The Whale optimization algorithm.
39. Liang JJ, Qin AK, Suganthan PN, Baskar S (2006) Comprehen-
Adv Eng softw 95:51–67
sive learning particle swarm optimizer for global optimization of
17. Fister I, Jr IF, Yang XS, Brest J (2013) A comprehensivereviewof
multimodal functions. IEEE Trans Evol Comput 10(3):281–295
firefly algorithms. Swarm Evol Comput 13:34–46
40. Ahmed ST, Amr AB, Ibrahim FAR (2013) One rank cuckoo
18. Kirkpatrick S, Gelatt CD (1983) Optimization by simmulated
search algorithm with application to algorithmic trading systems
annealing. Science 220(4598):671–680
19. Jiang QY, Wang L, Hei XH (2015) Parameter identification of optimization. Int J Comput Appl 4(6):30–37
chaotic systems using artificial raindrop algorithm. J Comput Sci 41. Sun J, Xu WB, Feng B (2004) A global search strategy of
8:20–31 quantum-behaved particle swarm optimization. In: IEEE Confer-
20. Rashedi E, Nezamabadi-Pour H, Saryazdi S (2009) GSA: a gravi- ence on cybernetics and intelligent systems. IEEE, pp 111–116
tational search algorithm. Inform Sci 179(13):2232–2248 42. Mirjalili S, Lewis A (2014) Adaptive gbest-guided gravitational
21. Mirjalili S, Mirjalili SM, Hatamlou A (2016) Multi-verse opti- search algorithm. Neural Comput Appl 25(7):1569–1584
mzer: a nature-inspired algorithm for global optimization. Neural 43. Mendes R, Kennedy J, Neves J (2004) The fully in formed par-
Comput Appl 27(2):495–513 ticle swarm: simpler, maybe better. IEEE Trans Evol Comput
22. Zhang QY, Wang RG, Yang J, Ding K, Li YF, Hu JG (2017) 8(3):204–210
Collective decision optimization algorithm: a new heuristic opti- 44. Yao X, Liu Y, Lin G (1999) Evolutionary programming made
mization method. Neurocomputing 221(19):123–137 faster. IEEE Trans Evol Comput 3(2):82–102
23. Hogarth RM, Soyer E (2011) Sequentially simulated outcomes: 45. Kennedy J, Mendes R (2002) Population structure and particle
kind experience versus nontransparent description. J Exp Psychol swarm performance. In: IEEE Congress on evolutionary computa-
Gen 140(3):434–463 tion. IEEE, pp 1671-1676
24. Koritzky G, Yechiam E (2010) On the robustness of description 46. Wang L, Zou F, Hei XH et al. (2014) An improved teaching-
and experience based decision tasks to social desirability. J Behav learning-basedoptimization with neighborhood search for applica-
Decis Making 23(1):83–99 tions of ANN. Neurocomputing 143(2):231–247
Modified collective decision optimization algorithm...

47. Parsopoulos KE, Vrahatis MN (2004) UPSO ła unified particle Kai Ding received the B.S.
swarm optimization. In: Lecture series on computational sciences, degree in computer science
pp 868-873 from Hainan Normal Uni-
48. Duan HB, Li P (2014) Bio-inspired computation in unmanned versity, China, in 2014. He
aerial vehicles. Springer-Verlag, Berlin, p 2014 received the M.S. degree
49. Duan HB, Liu SQ, Wu J (2009) Novel intelligent water drops opti- at School of Computer and
mization approach to single UCAV smooth trajectory planning. Information, Hefei University
Aerosp Sci Technol 13(8):442–449 of Technology in 2017. His
50. Duan HB, Yu YX, Zhang XY, Shao S (2010) Three-dimension research interests include digi-
path planning for UCAV using hybrid meta-heuristic ACO-DE tal image processing, artificial
algorithm. Simul Model Prac Th 18(8):1104–1115 intelligence and data mining.
51. Xu CF, Duan HB, Liu F (2010) Chaotic artificial bee colony
approach to uninhabited combat air vehicle (UCAV) path plan-
ning. Aerosp Sci Technol 14(8):535–541
52. Duan HB, Zhang XY, Xu CF (2011) Bio-inspired computing.
Science Press, Beijing, p 2011 Yongfu Li received the B.S.
Qingyang Zhang received degrees in mathematics from
the B.S. degree in mathemat- Anhui university of Science
ics from Beihua University, and Technology, Huainan,
Jilin, China, in 2011, the M.S. China, in 2014. His research
degree in applied mathemat- interests include image
ics from Beifang University processing, artificial intelli-
of Nationalities, Yinchuan, gence and intelligent visual
China, in 2015. He is currently surveillance.
pursuing the Ph.D. degree at
the School of Computer and
Information, Hefei University
of Technology, Hefei, China.
His current research interests
include meta-heuristics, evo-
lutionary computation and Jiangen Hu received the B.S.
image processing. degree in mathematics from
Anhui architecture university,
Ronggui Wang received the China, in 2014. He received
M.S. degree in mathematics the M.S. degree at School
from Anhui University, China, of Computer and Information,
in 1997, and the Ph.D. degree Hefei University of Technol-
in computer science from ogy in 2017. His research
Hefei University of Technol- interests include Image pro-
ogy, Hefei, China, in 2005. cessing, artificial intelligence
He is currently a Professor and video analysis.
with School of Computer
and Information, Hefei Uni-
versity of Technology. His
research interests include digi-
tal image processing, artificial
intelligence and data mining.

Juan Yang received the B.S.

and M.S. degrees in math-
ematics from Hefei Univer-
sity of Technology, Hefei,
China, in 2004 and 2008,
respectively. She received the
Ph.D. degree with school of
Computer and Information,
Hefei University of Technol-
ogy. She is currently a lecturer
with school of Computer and
Information, Hefei University
of Technology. His research
interests include image pro-
cessing and intelligent visual
surveillance.