Received October 19, 2021, accepted November 11, 2021, date of publication November 16, 2021,

date of current version November 22, 2021.

Digital Object Identifier 10.1109/ACCESS.2021.3128433

A Novel Sparrow Search Algorithm for the

Traveling Salesman Problem
CHANGYOU WU 1, XISONG FU 1, JUNKE PEI 1, AND ZHIGUI DONG 2
1 School of Management Science and Engineering, Shandong Institute of Business and Technology, Yantai 264005, China
2 Liaoning Institute of Science and Technology, Benxi 117004, China

Corresponding author: Changyou Wu ([email protected])

This work was supported in part by the National Natural Science Foundation of China under Grant 41601593, in part by the National Key
Research and Development under Project 2018YFD0300105, and in part by the Startup Foundation for Doctors of Liaoning Institute of
Science and Technology College under Grant 1910B04.

ABSTRACT The sparrow search algorithm (SSA) tends to fall into local optima and to have insufficient
stagnation when applied to the traveling salesman problem (TSP). To address this issue, we propose a
novel greedy genetic sparrow search algorithm based on a sine and cosine search strategy (GGSC-SSA).
First, the greedy algorithm is introduced to initialize the population and to increase the diversity of the
population. Second, genetic operators are used to update the population, balancing global search and local
development capabilities. Finally, the adaptive weight is introduced in the producer update to increase the
adaptability of the algorithm and to optimize the quality of the solution, and a sin-cosine search strategy is
introduced to update the scroungers. In addition, the GGSC-SSA is compared with the genetic algorithm
(GA), simulated annealing (SA), particle swarm optimization (PSO), grey wolf optimization (GWO), ant
colony optimization (ACO) and the artificial fish (AF) algorithm on TSP datasets for performance testing.
We also compare it with some recently improved algorithms. The results of the simulations are encouraging;
the GGSC-SSA significantly enhances the solution precision, optimization speed and robustness.

INDEX TERMS Sparrow search algorithm, traveling salesman problem, greedy algorithm, genetic opera-
tors, sin-cosine search strategy, combinatorial optimization.

I. INTRODUCTION efficiency. Since proposed, it has attracted the attention of lots

The core idea of swarm intelligence algorithms is to find of scholars [11], [12], [12]–[18].
optimal solutions by simulating the living habits and behav- Swarm intelligence algorithms are widely used in engi-
ior rules of creatures in nature and by searching for the neering optimization problems such as the knapsack prob-
spatial distribution of solutions in a limited space. Domes- lem [19], path planning [20]–[22], robot control [23], data
tic and foreign scholars have proposed a large number of mining [24], [25] and other issues [26]–[29]. The population
swarm intelligence algorithms through the swarm behavior characteristics and behavior of the heuristic algorithm are
of various swarms of intelligent creatures such as ants, bees, conducive to solving the discretization problem [30], [31].
birds, wolves, fireflies, sailfish, and sparrows, such as Particle The traveling salesman problem (TSP) is a classic combi-
Swarm Optimization (PSO) [1], Firefly Algorithm (FA) [2], natorial optimization problem [32], [33]. It is one of the
Ant Colony Optimization (ACO) [3], Grey Wolf Optimiza- standard test problems used in the performance analysis of
tion (GWO) [4], Sailfish Algorithm (SFO) [5] and Sparrow swarm intelligence algorithms and has NP-hard characteris-
Search Algorithm (SSA) [6] and so on [7]–[10]. Among tics. There are other practical problems that can be solved
them, the sparrow search algorithm was proposed by Jiankai in real life by abstracting and extending the TSP. For this
Xue and Bo Shen in 2020. Compared with other intelligent reason, the TSP remains a popular topic in current research
algorithms, the SSA has the advantages of simple imple- on new and different heuristic strategies, and it is significant
mentation, strong scalability, robustness, and high solution in both theory and practice. The continuous development
of swarm intelligence algorithms has shed new light on
NP-hard problems. An increasing number of algorithms have
The associate editor coordinating the review of this manuscript and been successfully applied to TSPs, including PSO [34], the
approving it for publication was Ugur Guvenc . GA [35], SA [36] and the SSA.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
153456 VOLUME 9, 2021
C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

In basic PSO, there are few parameters that need to be presented. In Section III, some basic knowledge is briefly pre-
adjusted, and the algorithm is easy to implement. However, sented. In Section IV, the proposed GGSC-SSA is described
the accuracy of solving the high-dimensional test function in detail, including the greedy strategy initializing the popu-
set is slightly insufficient. The GA has strong parallelism and lation, genetic variation strategy, and adaptive inertia weight
global search capabilities. However, it is easily falls into local investigated. In Section V, a series of TSP instances are simu-
optima. SA has a strong ability to jump out of local optima lated, and the results of the experiments are analyzed. Finally,
but has many parameters that need to be adjusted, and the a summary of the paper with conclusions and directions for
cooling time directly affects the efficiency of the algorithm. future improvement is presented in Section VI.
The advantages of the SSA mainly include strong robustness,
simple implementation and few parameters. However, due II. THE RELATED WORK
to the random generation problem of early producer spar- The TSP is of great significance in the history of operations
rows, the algorithm falls into local optima. A good SSA for research. In 1952, Danzig and others successfully solved the
solving the TSP should have the following characteristics: TSP examples of 48 cities in different states in the United
(1) Each improvement strategy should be adjusted according States and 49 cities in the District of Columbia, introducing
to the size of the TSP. (2) The use of loop sentences should the problem to more people for the first time. The significance
be reduced in the improved SSA to increase the speed of of combinatorial optimization research has also improved
the algorithm. (3) In the entire solution process, a balance the accuracy in solving discrete problems. With the rapid
between exploration and development should be achieved. development of heuristic algorithms, an increasing number
(4) For small or large TSP instances, the algorithm should of scholars have tried to apply different heuristic algorithms
be able to converge to the global optimal solution with high to solve the TSP. The TSP is an NP-hard problem; thus,
accuracy. In response to the above problems, this article there are no algorithms that can find the optimal solution in
proposes an improved SSA (the greedy genetic sine cosine polynomial time, so it is very important to study the swarm
sparrow search algorithm (GGSC-SSA)). intelligence algorithm of the TSP. A large number of
In this paper, the main effort is to improve the convergence new meta-heuristic algorithms are produced. As a result,
speed and the solution accuracy on TSP instances of different this natural-inspired algorithm design method has been
sizes. The main contributions of this work to research on the widely criticized. How to design an improved algorithm
TSP are as follows: for solving practical problems in your own domain is very
• We propose an improved SSA. The GGSC-SSA offers important [37], [38]. This paper chooses the SSA to solve
three main improvements over the basic SSA: the TSP, which is a very large challenge, because the SSA
1) The greedy algorithm is introduced into the SSA. has just recently been proposed and applied to the TSP for
First, the greedy algorithm is a simpler and faster the first time and has not been widely used.
design technique for finding a higher-quality TSP Most methods that can be used to solve the TSP are usu-
solution set, and it is used when the SSA initializes ally divided into two categories: (1) heuristic algorithms and
the population. Then, the top-down, iterative method (2) exact algorithms. Although there are some accurate
is used to make successive greedy choices, and each methods for solving the TSP with priority constraints in
time a greedy choice is made, the problem is reduced the literature, such as branching and shearing and dynamic
to a smaller subproblem, increasing the ability of the programming, exact solutions cannot be obtained as the scale
initial SSA to jump out of local optima. of the TSP continues to grow, and exact methods can solve
2) We apply genetic crossover and mutation strategies to only a small part of this problem. In recent years, due to the
update the SSA population. complexity of the TSP, metaheuristic algorithms such as the
3) We introduce dynamic adaptive weights to update the tabu search algorithm, simulated annealing (SA) algorithm
position of the producers. and genetic algorithm (GA) have been proposed in the liter-
4) We introduce the sine and cosine search strategy ature to solve this problem. This article briefly reviews the
to expand the search range of the scrounger, effec- related literature published in recent years.
tively preventing the algorithm from prematurely In 2019, Al et al. [22] presented a parallel version of the
converging. 2-opt algorithm based on Optical Transpose Interconnection
• The GGSC-SSA and other classic algorithms proposed System (OTIS) to solve the TSP. Reference [39] proposed
in the literature are tested on the TSPLIB test set. It is a novel Artificial Bee Colony(ACO) algorithm based on
found that the GGSC-SSA is superior to other algo- a swap sequence. The experimental results show that the
rithms in terms of solution time and solution accuracy. improved ACO has a good performance on the TSPLIB test
• We compare the proposed GGSC-SSA with other set, although it has insufficient solution time. In [40], Kim and
improved algorithms that have recently been presented Moon proposed a traveling salesman problem with a drone
and show that the proposed GGSC-SSA has great advan- stations(TSP-DS) based on the characteristics of UAV system
tages in terms of solution quality. delivery services. Zhukova et al. [41] developed a hybrid and
The remainder of this report is structured in the following accurate algorithm for solving the asymmetric traveling sales-
manner. In Section II, the latest SSA and TSP studies are man problem(ATSP). The key technology is to predict the

VOLUME 9, 2021 153457

 C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

solution time of the exact solution based on the combination one-dimensional extended case. This method can effectively
of branch and bound method and approximate algorithm. The reduce the computational complexity of the traveling sales-
experimental results show that the proposed algorithm solves man problem. Cavaleri et al. [54] proposed a method of dis-
the asymmetric traveling salesman problem more effectively. tance balance diagram which effectively solves the traveling
Zhu et al. [42] proposed a novel ant colony optimization salesman problem. Reference [55] introduced a metaheuristic
based on pearson correlation coefficient. A large number of approach, namly (Ib/ub)Alg, which was successfully applied
simulations in TSPLIB show that the proposed improved to the close-enough traveling salesman problem. In [56],
ant colony algorithm algorithm can obtain a better solution a strategy to consider the size of the time window is proposed,
for small, medium and large-scale TSP. Zhong et al. [43] which effectively improves the efficiency of solving the
introduced a discrete Pigeon-inspired optimization (DPIO) traveling salesman problem with a time window(TSPTW).
algorithm which uses the Metropolis acceptance criterion of Reference [57] proposed an agglomerative greedy brain
simulated annealing algorithm to solve the TSP problem. storm optimization algorithm(AGBSO) for solving
Zhao et al. [44] converted the energy-related mission plan the TSP.
into a dynamic traveling salesman problem, and proposed In 2021, several scholars have been proposed various
a hybrid method combining the Gaussian pseudospectral methods for TSP such as Pan et al. [58] proposed a novel Ant
method and the genetic algorithm (GPM-GA). The experi- Colony Optimization based Pheromone refactoring mech-
mental results show the effectiveness of GPM-GA in terms of anism. This algorithm effectively solves the problem of
energy efficiency, computational efficiency and smoothing of large-scale TSP falling into local optimality and slower con-
the attitude trajectory. vergence speed. Yang et al. [45] introduced an improved ACO
In 2020, Yang et al. [45] proposed a novel game-based for symmetric TSP problem based on Long Short-Term Mem-
ACO(NACO) that includes two ant colony systems and ory network and adaptive Tanimoto communication strategy.
introduces mean filtering to process pheromone distribution, Zhang et al. [59] proposed an improved whale optimization
which effectively solves the problem that basic ACO is easy to algorithm based on the adaptive weight, Gaussian
fall into local optimum. In [46], ABC and Greedy Algorithm disturbance, and variable neighborhood search strategy.
were combined in a novel manner to form an improved ABC, Experimental results show that, compared with recent related
which was successfully applied to multi-objective traveling algorithms, this algorithm has better optimization perfor-
salesman problem. In [47], a modified version of social mance and higher efficiency. Zelinka et al. [60] introduced
group optimization (SGO) has more competitive results when a gamesourcing approach to replace ACO. The algorithm is
solving TSP, and its convergence speed is better than GA in the form of a maze, TSP nodes move within the maze, and
and discrete particle swarm optimization. The efficiency then the performance of the algorithm is evaluated and com-
of solving large-scale TSP problems has also been proved. pared with some well-known versions of ACO. Experiments
Tu-san et al. [48] introduced a novel variant of the TSP, show that this method achieves better results on well-known
called the intermittent travelling salesman problem (ITSP), NP-hard optimization problems such as TSP. Yousefikhosh-
and proposed a branch and bound method to solve the bakht [61] provided an improved particle swarm algorithm
optimality problem of ITSP. Reference [49] proposed a new that shifts the particles to the best particles. This method
ACO based on dynamic adaptive method. In addition, the takes into account the concept of randomness and prevents
experiment of the variant ant colony algorithm tested on the premature convergence of the algorithm. Wu et al. [62]
TSPLIB instance shows that this method has better algorithm combined k-means, top-layer ACS, and bottom-layer ACS
performance. Tran et al. [50] designed a UAV trajectory that to solve large-scale TSP. The experimental results show
reduces energy consumption based on the traveling salesman that the solution efficiency of the algorithm is effective.
problem, and proposed a new heuristic search and dynamic Vasquez et al. [63] studied the Traveling Salesman Problem
programming (DP) method. The results show that the with Drone(TSP-D), which is a variant of the TSP prob-
DP algorithm is close to exhaustion with significantly lem, and proposed a mixed-integer programming Eq. and a
reduced complexity. Cinar et al. [86] proposed an improved Benders-type precise algorithm. Finally, the proposed method
Tree Seed algorithm to solve TSP. Experimental results show has been empirically tested in a randomly generated example
that DTSA is another qualified and competitive solver on to prove its effectiveness. Sun et al. [64] studied the gener-
discrete optimization. In [51], a novel heuristics mathematical alization ability of the machine learning model, which can
Eq.tion is proposed, which is based ACO to minimize travel effectively solve the problem of the classic traveling salesman
costs. In [52], an analog electronic computing system has problem (TSP). Experiments have proved that the model can
been successfully applied to the traveling salesman problem. find a better solution from the optimization problem. Even
The system spontaneously and dynamically simulates the if tested on different TSP problem variants, the model can
effective foraging behavior of similar organisms, and realizes still make useful predictions and improve the solution quality
the flexibility and flexibility of high problem mapping, and of the TSP problem. Stieber et al. [65] introduced a new
has high application potential. Popescu et al. [53] have type of dynamic model to solve the problem of multiple
successfully developed a novel approach to approximate the traveling salesmen(MTSP)with moving targets in an accurate
Shapley value of Euclidean TSG, which is inspired by the way. Compared with other mathematical models and swarm

153458 VOLUME 9, 2021

C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

intelligence algorithms on randomly generated large-scale Algorithm 1 Pseudocode of the Basic SSA
examples, the results show that the proposed model has Input: MaxIter: the maximum number of iterations
strong solution efficiency and robustness. Silva et al. [66] N : the number of sparrows
invented a new technique for parallel computing called Multi PD: the number of producers
Improvement (MI). In addition, three dynamic programming ST : the safety value
algorithms for solving the Maximum Multi Improvement SD: the number of sparrows that perceive danger
Problem (MMIP) are developed, and the effective solution of R: the alarm value
the traveling salesman problem is given. In order to solve the Output: Xb : the global optimal individual
inefficiency of the dynamic programming method in solving Fb : the best fitness value
the sequential order problem (SOP), Salii et al. [67] proposed 1: Initialize a population of N sparrows and the parameters.
a new dynamic programming method with lower bound 2: g = 1(Record the number of iterations).
heuristic parameters. The scheme is tested on an example 3: while g < MaxIter do
of TSPLIB, and the effectiveness of the proposed method is 4: Calculate the fitness values of all individuals.
proved. 5: Sort the individual fitness values, and mark the best
Overall, a large number of scholars have contributed to the individual and the worst individual.
solution of the TSP, and it has been proven that the TSP has 6: R = rand(1),
value in real life. Although scholars have made some progress 7: for i = 1 : PD do
in combinatorial optimization theory and application, there 8: Use Eq.4 to update the producer locations.
are still some key issues. The improved algorithms proposed 9: end for
in [57], [59] and [68] can reach an ideal state when solving 10: for j = (PD + 1) : N do
small-scale symmetric TSPs. The improved algorithms pro- 11: Use Eq.5 to update the scrounger locations.
posed in [34], [41], [45] and [69] have achieved satisfactory 12: end for
solution accuracy when solving symmetric TSPs, but the 13: for J = 1 : SD do
solution time is very unsatisfactory. References [40], [70] 14: Use Eq.6 to update the warning locations.
and others have converted practical problems into TSPs well, 15: end for
but the solving efficiency of the algorithm is an obvious 16: Get the current new locations.
disadvantage. Note that the cited in this section are repre- 17: Compare the new and old individuals.
sentative of only a small portion of the related work on the 18: end while
TSP. Due to the increasing number of studies on the TSP, 19: return Xb and Fb
it is difficult to summarize all the related work. Therefore,
to further understand the solutions related to the TSP and
its variants, it is recommended that readers study the work be expressed as:
introduced in [71], [72] and [73]. On the other hand, readers n
n X
who want to know more about the possible applications of the
X
min Z = dij xij
SSA can refer to [11]–[14], [16]–[18]. i=1 j=1
 Xn
III. THE BASIC PROBLEM DESCRIPTION  xij = 1, i∈V

 j=1
A. TRAVELING SALESMAN PROBLEM (TSP) 
 Xn
xij = 1, j∈V


The TSP is a classical combinatorial optimization problem.

i=1
s.t (1)
In this problem, a businessman needs to visit several cities and
X X

 xij 6 |S| − 1, ∀S ∈ V
then return to the starting city; each city can be visited only

 i∈S

 j∈S
once, and the shortest path needs to be determined. Although


xij ∈ {0, 1}
the constraints are simple, it is extremely complicated to solve
as the number of places increases. Currently, the solutions In Eq. 1, dij represent the distance between each vertex,
to this combinatorial optimization problem can be roughly xij represent the decision variable, xij = 1 is on the loop,
divided into two categories: precise optimization algorithms xij = 0 is not on the loop. Corresponding to the Hamiltonian
and metaheuristic algorithms. The existing research shows cycle, G = (V , E), V is the vertex set, E is the edge set, the
that precise optimization algorithms can effectively solve third set of constraints are sub-tour elimination constraints.
small-scale TSPs, while metaheuristic algorithms are more
suitable for medium- and large-scale TSPs. Precise optimiza- B. SPARROW SEARCH ALGORITHM (SSA)
tion algorithms include the gradient descent method, Newton The inspiration of the SSA comes from the foraging behav-
method, dynamic programming, enumeration method and ior of sparrows in nature. Sparrows have excellent flying
others. Meta-heuristic algorithms include SA, PSO, the SSA, ability and strong vigilance and are resident birds that like
etc. The SSA employs the ideas of evolution and a flex- to live with humans. According to the different foraging
ible behavior strategy and provides a novel method for behaviors, sparrows are divided into two types: producers and
solving the TSP. The mathematical model of the TSP can scroungers. Producers are responsible for finding food and

VOLUME 9, 2021 153459

 C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

have higher energy reserves. Scroungers follow and monitor where XOP is the best location of the producers.
producers and have low energy reserves; some scroungers XWorst denotes the worst position of the scroungers in the
compete with producers for food. When predators (natural current iteration. When i > n/2, the fitness value of the
enemies of sparrows) appear in a foraging area, a sparrow ith scrounger is low, and it is unable to access enough
recognizes the danger and immediately enters alert mode. The food.
basic SSA pseudocode is shown in Algorithm 1. When the sparrow population forages at the feeding source,
In the SSA, it is necessary to simulate the sparrow foraging 10%-20% of the sparrows perform early warning work to
process to find the solution of the target problem. The position prevent being attacked by predators. The updated position
of the sparrows is denoted as follows: of sparrows with early warning capability can be shown as
follows:
x11 x12 · · · x1d
 

x21 x22 · · · x2d  X t + β · Xi,j
t − Xt if fi 6= fb
X = . .. .. ..   Best Best

(2)
..

. . . t+1
Xi,j = t −X t
Xi,j
!
(6)
Worst
xn1 xn2 · · · xnd 
 t
Xi,j + H · (fi −fw )+δ if fi = fb
where n represents the population size of sparrows and d is
the dimension of the variables to be optimized. The fitness where XBest is the current global optimal position. fi rep-
values of all sparrows are expressed by the following matrix: resents the fitness value of the current sparrows. fb and fw
express the current best and worst fitness values, respectively.
f ([x11 x12 · · · x1d ])
 
β represents standard normally distributed random numbers
 f ([x21 x22 · · · x2d ]) 
FX =  .. .. .. .. .. (3) with an average value of 0 and a variance of 1. H is a
.....

random number in the interval [0,1], and controls the moving
f ([xn1 xn2 · · · xnd ]) direction of sparrows and the adjustment of the step size. δ is
Each value in FX represents the value of the individual. the minimum constant,which prevents the situation where the
The higher the fitness value of a sparrow, the easier it is for denominator is 0 and the fitness value of the current sparrow
it to obtain food during foraging. Additionally, they can act is the global worst. When fi 6= fb , the sparrows are at the edge
as producers that are responsible for the food search of the of the foraging area and are vulnerable to predators. fi = fb
whole population and can find food outside the search space. shows that the sparrow at the center of the population is aware
According to Eq. 2 and 3, in each iteration, the producers of the danger and needs to quickly approach other sparrows
update the position, and the formula of the position update to readjust the foraging strategy.
is as follows:
(   IV. IMPROVED SSA FOR THE TSP
t+1
t · exp
Xi,j −i
α·Itermax if W < ST A. THE GREEDY ALGORITHM INITIALIZES THE
Xi,j = (4) POPULATION
t +Q·L
Xi,j if W > ST
Dynamic programming algorithms usually give solutions
where t represents the current iteration. Xi,jt represents the with a bottom-up method, while greedy algorithms, in con-
value of t iterations of the ith sparrow in the jth dimension. trast, use the method of constructing the optimal solution
α is the random number in the interval [0, 1]. Itermax rep- stepwise with top-down method and make greedy choices
resents the maximum number of iterations of the current in an iterative fashion. Every time a greedy choice is made,
population. W (W ∈ [0, 1]) and ST (ST ∈ [0.5, 1.0]) repre- the optimization problem is simplified to a smaller subprob-
sent the alarm threshold and safety threshold respectively. lem [74]–[76]. The nature of the greedy algorithm means that
Q is a random number that follows the normal distribution. the global optimal solution of the problem can be achieved
L shows a 1 × d matrix in which each element inside is 1. through a series of local optimal choices, that is, greedy
When W < ST , the sparrow population is in a safe state and choices. Greedy algorithms have been widely used in path
continues to forage, while the producers search for food in planning [77], job-shop scheduling [78] and other issues [79].
a large range. If W > ST , then predators are present in the In this paper, the greedy algorithm is used to replace the
sparrow population, and all sparrows need to immediately fly random generation of the population in the original SSA,
to a safe area. which not only maintains the diversity of the population,
Except for the producers, all the sparrows in the popula- but also improves the efficiency of the algorithm. Note
tion are scroungers, and their positions are updated with the that greedy algorithm has some shortcomings, such as the
following formula: inability to guarantee that the final solution is the optimal
t+1 solution.
Xi,j
  t t
 B. THE GENETIC OPERATORS UPDATE THE POPULATION
Q · exp XWorst2−Xi,j

if i > n
1) OX CROSSOVER OPERATOR
i 2
= To increase the diversity of sparrow population, OX crossover
X t+1 + X t − X t+1 · AT AAT −1 · L


otherwise

OP i,j OP operation in the GA is used after population initializa-
(5) tion [80]. Assume that the parent individuals are as follows:

153460 VOLUME 9, 2021

C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

Algorithm 2 Pseudocode of the GGSC-SSA of GA is introduced to update the position of the producer
Input: MaxIter: the maximum number of iterations and the exchange operation. The introduction of a mutation
N : the number of sparrows strategy improves the search efficiency and global optimiza-
PD: the number of producers tion ability of the producers. Dynamic adaptive weights are
ST : the safety value imported to update the position of the producers [83]. The
SD: the number of sparrows that perceive danger location of the producers are updated with the weight coeffi-
R: the alarm value cient [84], and the formula is as follows:
Output: Xb : the global optimal individual
λmin − (λmax −λ min )·(fi −fw )
(
Fb : the best fitness value if fi < favg
λ= favg −fw
(7)
1: Structure coding method. Initializes the sparrow popula- λmax otherwise
tion N and the parameters with the greedy algorithm. (
t · exp −i
Xi,j


if W < ST
2: g = 1(Record the number of iterations).
t+1
Xi,j = α·t (8)
t +λ·L
Xi,j if W > ST
3: while g < MaxIter do
4: Calculate the fitness value of all individuals. where λ is a random number in the interval [0,1]. Two random
5: Update the population with the OX crossover operator. numbers are generated, and after comparison, λm in and λm ax
6: R = rand(1) are obtained; favg is the average of the current global optimal
7: for i = 1 : PD do and worst fitness values. Through the introduction of the
8: Use Eq.8 to update the producer location. dynamic weight coefficient, the adaptation of the algorithm
9: end for is effectively increased, and the coefficient is adjusted with
10: for j = (PD + 1) : N do the number of iterations to better perform a global search.
11: Use Eq.5 to update the producer’s location.
12: Use the sine and cosine search strategy to enhance C. SINE AND COSINE SEARCH STRATEGY
global ability using Eq.9 In the SSA population update, the scrounger location update
13: end for is mainly guided by the producers, which results in a rough
14: for J = 1 : SD do optimization effect. To further improve the convergence accu-
15: Use Eq.6 to update the producer’s location. racy and optimization effect of the algorithm, and to balance
16: end for the local development and global search abilites, a sine and
17: Get the current new location. cosine search strategy is introduced [85]. After the scroungers
18: Compare the new and old individuals. are updated according to the location of the producer, a sine
19: end while and cosine search is carried out to obtain the optimal feasible
20: return Xb and Fb solution. The mathematical expression of the sine and cosine
search strategy is as follows:

X t + r1 · sin (r2 ) · r3 · Pt − X t r4 < 0.5
t+1 i,j i,j i,j
Xi,j = (9)
X t + r1 · cos (r2 ) · r3 · Pt − X t r4 > 0.5
i,j i,j i,j

We randomly select two crossover positions 3 and 6, then t

r1 = a − a · (10)
move the crossover segment of parent 2 to the front of MaxIter
parent 1, and the crossover segment of parent 1 to the front of where r1 will increase with the increase of iteration times
parent 2 and delete the duplicate individuals in turn to form MaxIter, a is a constant, the value in experiment is 2,
two offspring individuals. This is expressed as follows: r2 is a random number the interval [0, 2π ], r3 is the random
number between [0,2], and r4 is the uniformly distributed
random number on [0,1]. The function of sine and cosine
search is to make the algorithm effectively prevent premature
convergence and improve the convergence accuracy to a cer-
tain extent, so as to improve the efficiency of each iteration.
2) MUTATION STRATEGY
The purpose of the GA mutation operator is twofold: first, D. CONSTRUCTION CODING MODE
to give the GA local random searching abilities [80], [81]. Since the basic sparrow search algorithm cannot directly
When the GA is close to the optimal solution neighbor- solve some discrete optimization problems like TSP, it is
hood through the crossover operator, the local random search necessary to reconstruct the search space of the algorithm
ability of the mutation operator can accelerate the conver- and redefine the objective function according to the actual
gence to the optimal solution [82]. In the SSA, the producer problem [86], [87]. Therefore, the solution range of the
approaches the global optimal solution from the beginning sparrow search algorithm needs to be transformed into a
iterations, and the search range is restricted, so it is easily two-dimensional continuous space. Only by defining the
trapped in local optima. Therefore, the variation strategy value range of the independent variable and the objective

VOLUME 9, 2021 153461

 C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

function expression, the optimal solution and the correspond- According to the above re-encoding settings for the spar-
ing independent variable value can be obtained. row, the improvement process of the sparrow search algo-
After introducing the TSP problem into the SSA, it can be rithm can be abstracted into a combined optimization model
defined as the sparrow population size as N and the number on a continuous space, and the sparrow search algorithm can
of cities as D. In the D dimensional city search space, the be applied to the TSP problem.
position Xi of the i-th sparrow is defined as a set of different
positive integer sequences, and the N sparrows search for E. GGSC-SSA FOR THE TSP
prey in the D dimensional space, that is, the search space The GGSC-SSA is based on the initial SSA and introduces
domain K is an entity matrix, the formula is as follows: a greedy algorithm to initialize the population. When the
 1 producers and scroungers are updateds, the crossover and
X12 · · · X1D

X1 variation strategies in the GA are embedded to optimize the
results of sparrow traversal. Finally, global optimization is
 1
X22 · · · X2D 

X
 2
K = (11) carried out according to the position of the early warning

 .. .. .. .. 

 . . . .  sparrow. The steps for solving the TSP with a combined
XN1 XN2 · · · XND a greedy genetic strategy and sine and cosine SSA are as
follows:
The first row of the matrix indicates the position sequence 1) Initialize the TSP city information and GGSC-SSA
of the first sparrow in the search space, and the last row parameters, and discretize the algorithm;
indicates the position sequence of the N th sparrows in the 2) Initialize the sparrow population with the greedy
search space. algorithm;
After constructing the sparrow population search space 3) Calculate the fitness values of all sparrows in the popu-
matrix and the sparrow position sequence expression method, lation, and find the sparrow with the best fitness value;
another important problem is to solve the distance matrix L 4) Select part of sparrows that have the highest fitness value
in the TSP problem. For the TSP with the number of cities D, as producers, and update their positions according to
the distance matrix L formed by the distance d(i,j) between Eq. 8. Update the positions of the remaining sparrows
the ith city and the jth city can be expressed as: as scroungers according to Eq.5;
  5) Randomly select early warning sparrows from the pop-
d(1,1) · · · d(1,N ) ulation, and update the positions based on Eq.6;
L =  ... ..
.
.. 
.  (12) 6) Use the sine and cosine search strategy through the


d(N ,1) ··· d(N ,N ) updated warning sparrows to prevent the algorithm from
falling into local convergence;
The distance of d(N ,N ) is 0. Through the above description, 7) According to the current state of the sparrow population,
the relationship expression between the search space and the update the optimal position and fitness of the entire
objective function can be constructed as follows: population, as well as the worst position and fitness;
X  8) Judge whether all individuals are traversed, if so, pro-
C = min K d (i, j) ceeding to the next step, otherwise, jumping to step 3;
 9) Judging whether the maximum iteration times have been
 (i, j) 6 D
 reached, if so, proceed to the next step, otherwise, go
s.t (i, j) ∈ N (13) to step 2;
10) Introduce the program operation, output the optimal

i 6= j

result.
Among them, i and j represent the city number, The detailed process of the GGSC-SSA algorithm for solv-
min K represents the optimal populationP position, and the ing the TSP is shown in Figure1. The GGSC-SSA solves TSP
distance matrix sum corresponding to d (i, j). The function pseudocode as shown in Algorithm 2.
of this function is to read the cumulative sum of the dis-
tance between the position sequence in each sparrow position V. EXPERIMENTAL STUDIES
matrix K and the corresponding distance matrix L. In this section, the experiments conducted on the TSP to test
Random initialization of different sparrow individuals will the improved SSA are introduced in detail. First, we ana-
generate different solution vectors, and calculate the distance lyze the parameters related to the algorithm and the related
between different solution vectors. The optimization of objec- components of the improved SSA. For the TSP dataset,
tive function is determined through function. If the solution the 36 examples used in this article are from the TSPLIB
obtained is better than the previous one, it will be replaced benchmark. In these 36 instances, the number of city nodes
with a better solution and used as the optimal solution for ranges from 22 to 1291. For each instance, the algorithm
the current iteration of this sparrow. Otherwise, it remains described in this article is run 50 times. To evaluate the perfor-
unchanged, and the next line of judgment is continued until all mance of the GGSC-SSA, we compare it with the following
the sparrow solution vector optimizations are all completed. intelligent optimization methods: (1) traditional intelligent

153462 VOLUME 9, 2021

C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

TABLE 1. Algorithm related parameter settings.

TABLE 2. Component comparison test results.

in Table 1. Note that the parameters in the sine and cosine

search strategy are set according to the literature [85].
To test the performance of each component of the
GGSC-SSA and to analyze the impact on the basic SSA,
the three components are added to the SSA and renamed.
Additionally, comparative experiments are carried out on the
Ulysses22, Eil51, Berlin52, Rat99 and Ch130 datasets. The
results are shown in Table 2. The algorithm variants for
different components are defined as follows:
1) The SSA introduced with the greedy algorithm is repre-
sented by SSA1.
2) The SSA introduced with the genetic operator is repre-
sented by SSA2.
FIGURE 1. The GGSC-SSA algorithm. 3) The SSA that introduces the sine and cosine search
strategy is represented by SSA3.
Additionally, we obtain the characteristics of each com-
ponent through experiments, and the results are as follows:
algorithms, such as the GA, SA, PSO, GWO, ACO, and
1) The time spent by SSA1 on the tested dataset is
AF [88] and (2) other improved intelligent algorithms [6],
significantly shorter than that of the original SSA, but
[45], [57], [61], [89]–[93]. The simulation experiment is run
the solution accuracy is not sufficient, and the global
on a computer equipped with an Intel Core i7-10600 proces-
search capability is poor. When SSA1 is used to solve
sor, the program is created in a Windows 10 environment, and
the TSP examples ulysses22, eil51, berlin52, rat99 and
the programming software used is MATLAB2018a.
ch130, the solution times (in seconds) are 1.32, 3.69, 10.24,
12.16 and 15.73, respectively. However, the solution accu-
A. COMPONENT TESTING AND PARAMETER SETTINGS racy is the worst, at 85.63, 460.97, 7956.23, 1398.3 and
For many intelligent optimization algorithms, parameter tun- 7026.34. The optimized value is lower than that of the
ing is the key to algorithm optimization performance. After original SSA.
analysis of the existing literature related to the SSA [6], [11], 2) The solution accuracy of SSA2 is better than that of the
[12], [18], [94], [95], the parameters of the GGSC-SSA are original SSA and weaker than that of SSA3, and the solution
determined through repeated and in-depth revisions, as shown time is significantly longer than those of the original SSA

VOLUME 9, 2021 153463

 C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

TABLE 3. Results of the proposed SSA algorithm and the basic SSA algorithm.

and SSA1. SSA2 has a good local search ability and easily on test sets berlin52 and rat99 are 39.56 and 67.63 seconds,
falls into local optima. The optimal values of SSA2 on the respectively, which are 14.84 and 12.93 seconds longer than
datasets eil51 and ch130 are 428.36 and 6416.38, and the the initial SSA solution times. In terms of solution accuracy,
times are 12.23 and 49.63. Compared with the optimal values the deviation rate (calculated as shown in Eq.14) of SSA3
of the initial SSA, the values are 13.45 and 425.01 higher, are 0.08% and 0.68%, which are significantly higher than the
and the solution times are 6.04 and 13.54 seconds longer, 2.27% and 7.21% of the original SSA.
respectively.
BV − KV
3) SSA3 performs relatively well in terms of solution accu- Dev = × 100% (14)
racy, significantly better than the basic SSA, SSA1 and SSA2, KV
but it takes more time to reach a solution. SSA3 has excellent where Dev represents the deviation rate, BV represents the
global optimization capabilities that effectively prevent the best solution value of the algorithm, and KV represents the
algorithm from falling into local optima. The times spent best known value.

153464 VOLUME 9, 2021

C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

FIGURE 2. Dynamic average convergence curves of GGSC-SSAand the original SSA algorithm.

B. COMPARISON OF THE ORIGINAL AND IMPROVED SSA Based on the results shown in Table 3, an obvious con-
In this subsection, a comprehensive comparison between clusion can be drawn. For 36 TSP instances, the improved
the GGSC-SSA and SSA is made to prove that the per- SSA is significantly better than the basic SSA. We conduct
formance of the improved SSA is better than that of the further analysis to prove the validity of the conclusion. Of all
original SSA. The experimental results of these two algo- the TSP instances, six instances of the GGSC-SSA reach
rithms are listed in Table 3, where ‘‘optimal’’ represents the known optimal solution, namely, on ulysses22, eil51,
the known optimal solution of the instance, ‘‘best’’ repre- berlin52, rat99, kroA100 and tsp225. Note that the optimal
sents the average value after running the instance 50 times, values of ulysses22, kroA100 and tsp225 are lower than
and ‘‘time’’ represents the average time after running the known optimal solutions, which are 75.24, 20989.04
the instance 50 times. A total of 36 TSP instances are and 3900.93, respectively. Supporting the authenticity of
tested. the results obtained, many studies related to TSPs report

VOLUME 9, 2021 153465

 C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

FIGURE 3. Some examples of optimal paths obtained by the improved SSA.

FIGURE 4. The dynamic average convergence curves.

solutions better than the known optimal solution for TSP than 1%. When the TSP scale continues to increase, the
examples. For example, in [96], using the improved ant solution performance of the GGSC-SSA is much better than
colony algorithm to solve ulysses22, the optimal value of that of the SSA. For the TSP instance tsp225, the best value
75.31 found is lower than the known optimal value of 75.67; obtained by the GGSC-SSA is 3900.93, which is 698.83
in [97], the metaheuristic hybrid algorithm is used on the lower than the best value obtained by the SSA, and the
examples ulysses22, att488 and berlin52, and the optimal solution time is 59.92 seconds shorter than the 91.15 sec-
values are 56.52, 13908.4 and 5970.83, which are all lower onds of the SSA. For TSP instances rat783 and pr1002, the
than the known best values; in [57], when using the improved standard deviation rates of the GGSC-SSA are 3.26% and
brainstorming algorithm to solve ulysses22 and kroA100, the 4.96%, respectively, which are significantly better than the
best values are 75.24 and 21070.09, which are both lower SSA values of 24.60% and 16.16%.
than the known best values. In 29 experimental results, the Through the above detailed analysis, it is clear that the
standard deviation is less than 3%, accounting for 75% of GGSC-SSA is superior to the original SSA in terms of solu-
all examples. When the number of city nodes is less than or tion accuracy, solution time, and stability. The six examples
equal to 100, only the standard deviation of att48 is greater of ulysses22, eil51, berlin52, rat99, kroA100 and tsp225 are

153466 VOLUME 9, 2021

C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

TABLE 4. Experimental results of the SSA algorithm and other classic heuristic algorithms.

GGSC-SSA can find a better value in the first iteration.

The greedy algorithm is used for the population initializa-
tion of the SSA, which greatly enhances the optimization
ability of the SSA. To further illustrate the advantages of
the GGSC-SSA in terms of the optimization efficiency and
solution time, the optimal value found by each algorithm and
the specific time spent are introduced in detail. The experi-
mental results are shown in Table 4. In this table, Dev, MeanV
and MeanT represent the standard deviation rate, average
running time and average running time, respectively. Taking
the eil51 dataset as an example, the GGSC-SSA reaches the
known optimal value, exhibiting great advantages compared
with other algorithms. In particular, compared with GA, PSO
FIGURE 5. Run time visualization. and GWO, the deviation rate is 97.16%, 78.53% and 94.55%
lower, respectively. Note that the SA algorithm performs
better in terms of solution time, but as the TSP scale increases,
selected to visually analyze the performance of the improved the solution accuracy continues to decrease. Figure 5 shows
SSA and the basic SSA. The dynamic convergence curve is the detailed solution time of the eight algorithms used to
shown in Figure 2. Figure 3 shows the experimental results solve the three TSP instances. When solving small-scale
of TSP instances where the GGSC-SSA achieves a known TSPs, GGSC-SSA is better than the other algorithms in terms
optimal solution. of optimizing speed, deviation rate and stability. Compara-
tive experiments with traditional heuristic algorithms further
C. EXPERIMENTATION WITH THE GGSC-SSA AND show that the improved algorithm has better optimization
CLASSICAL INTELLIGENT ALGORITHM capabilities and greater robustness.
To prove the efficiency of the GGSC-SSA, we compare the
improved algorithm with six classic heuristic algorithms, GA,
SA, PSO, GWO, ACO and AF, on the ulysses22, eil51 and D. COMPARISON WITH OTHER IMPROVED ALGORITHMS
berlin52 datasets. To ensure the validity and fairness of the To further comprehensively verify the efficiency of the
experiment, for the same TSP instance, all algorithms are GGSC-SSA, it is compared with a series of recently improved
tested under the same hardware environment. Figure 4 shows intelligent optimization methods. The nine improved algo-
the optimization process of the eight algorithms used to test rithms participating in the comparison are as follows:
the TSP instances ulysses22, eil51 and berlin52. (1) the novel discrete water cycle algorithm (DWCA) [90];
The information in Figure 4 clearly shows that for (2) the improved ant colony optimization (IACO) [98];
the tested TSP examples, the solution accuracy of the (3) the agglomerative greedy brain storm optimization
GGSC-SSA and the optimal number of iterations are bet- (AG-BSO) [57]; (4) the parallel ant colony optimization
ter than those of the other seven algorithms. Note that and 3-opt (PACO-3OPT) algorithm [99]; (5) the genetic ant

VOLUME 9, 2021 153467

 C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

TABLE 5. Statistical results of GGSC-SSA and nine other algorithms used to solve TSP instances.

colony optimization (GACO) [91]; (6) the adaptive brain 1) The greedy algorithm is introduced into the SSA to ini-
storm optimization (ABSO) [100]; (7) the Pearson correlation tialize the population to enhance the solving efficiency
coefficient ant colony optimization (PCCACO) [91]; (8) the of the algorithm.
novel ant colony optimization (NACO) [45]; (9) the multi- 2) The crossover operation of the GA is used to update
strategy discrete brain storm optimization (MDBSO) [101]. the population to enhance the global search ability. The
The statistical results of the GGSC-SSA and other mutation operation is used to update producers and to
improved algorithms are shown in Table 5, and ‘‘-’’ indi- enhance the local search ability of the algorithm.
cates that the method is not tested in its article. The con- 3) Sine and cosine search strategies are used through early
clusions drawn from Table 5 are similar to the previous warning sparrows to prevent premature convergence
conclusions. In the small TSP examples eil51, berlin52, st70, and to enhance the global optimization ability of the
eil76 and kroA100, good results are obtained. In addition, algorithm.
for the large-scale TSP instance pr439, the performance of On the basis of a thorough and comprehensive theoret-
the GGSC-SSA proposed in this paper is significantly better ical study of the original SSA and the TSP, a novel SSA
than that of the other improved algorithms. In contrast, the (GGSC-SSA) is introduced for the first time to solve the
GGSC-SSA maintains good adaptability to all TSP instances. TSP in this research. To demonstrate that the proposed
In summary, the experimental data show that this method GGSC-SSA is an effective algorithm for solving the TSP,
has strong competitiveness. As the complexity of the prob- we compare its performance with the basic SSA on 36 TSP
lem increases, the GGSC-SSA can jump faster out of local instances. Furthermore, we analyze the GGSC-SSA in
optimal solutions, thereby improving the global optimization detail through comparative experiments with six classical
capability. This effect is mainly derived from the greedy algorithms and eight existing improved algorithms. The sim-
algorithm to obtain a solution close to the global optimal ulation results validate the effectiveness of the proposed
value. The sine-cosine search strategy enhances the global algorithm. The GGSC-SSA shows excellent performance in
optimization capability, thereby improving the convergence solving TSP cases on large and small scales and is better
of the algorithm. The GGSC-SSA proposed in this paper can than other improved algorithms in most cases. In future work,
obtain a better and more stable solution when solving TSPs, other intelligent algorithms will be introduced into the SSA
which is more obvious in large-scale examples. to explore the new intergroup communication model and to
improve the robustness and adaptability of the SSA.
VI. CONCLUSION
To address the issues of the SSA having insufficient conver- ACKNOWLEDGMENT
gence ability and efficiency in solving the TSP, an improved The authors would like to thank the anonymous reviewers
SSA named the GGSC-SSA is introduced in this paper. and Academic Editor for their valuable and constructive com-
The three key improvements of the GGSC-SSA are as ments, which greatly improved the quality and integrity of
follows: this manuscript.

153468 VOLUME 9, 2021

C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

REFERENCES [24] P. F. Robbins, Y. C. Lu, M. El-Gamil, Y. F. Li, C. Gross, J. Gartner,

J. C. Lin, J. K. Teer, P. Cliften, E. Tycksen, and Y. Samuels, ‘‘Mining
[1] M. Clerc and J. Kennedy, ‘‘The particle swarm—Explosion, stability, and
exomic sequencing data to identify mutated antigens recognized by adop-
convergence in a multidimensional complex space,’’ IEEE Trans. Evol.
tively transferred tumor-reactive T cells,’’ Nature Med., vol. 19, no. 6,
Comput., vol. 6, no. 1, pp. 58–73, Aug. 2002.
p. 747, 2013.
[2] X. Yang, ‘‘Firefly algorithms for multimodal optimization,’’ in Proc. Int.
[25] X. Wu, X. Zhu, G.-Q. Wu, and W. Ding, ‘‘Data mining with big data,’’
Conf. Stochastic Algorithms Found. Appl., 2009, pp. 169–178.
IEEE Trans. Knowl. Data Eng., vol. 26, no. 1, pp. 97–107, Jan. 2014.
[3] M. Dorigo, V. Maniezzo, and A. Colorni, ‘‘Ant system: Optimization by
[26] D. I. Arkhipov, D. Wu, T. Wu, and A. C. Regan, ‘‘A parallel genetic algo-
a colony of cooperating agents,’’ IEEE Trans. Syst., Man, Cybern. B,
rithm framework for transportation planning and logistics management,’’
Cybern., vol. 26, no. 1, pp. 29–41, Feb. 1996.
IEEE Access, vol. 8, pp. 106506–106515, 2020.
[4] S. Mirjalili, S. Saremi, S. M. Mirjalili, and L. D. S. Coelho,
[27] J. Cao, M. Olvera-Cravioto, and Z.-J. Shen, ‘‘Last-mile shared delivery:
‘‘Multi-objective grey wolf optimizer: A novel algorithm for multi-
A discrete sequential packing approach,’’ Math. Oper. Res., vol. 45, no. 4,
criterion optimization,’’ Expert Syst. Appl., vol. 47, pp. 106–119,
pp. 1466–1497, Nov. 2020.
Apr. 2016.
[28] J. Chen and Z. Chen, ‘‘Spatial path–energy optimization for UAV oper-
[5] S. Shadravan, H. R. Naji, and V. K. Bardsiri, ‘‘The sailfish optimizer:
ation in arial–ground networking,’’ J. Comput. Civil Eng., vol. 34, no. 3,
A novel nature-inspired metaheuristic algorithm for solving constrained
May 2020, Art. no. 04020008.
engineering optimization problems,’’ Eng. Appl. Artif. Intell., vol. 80,
[29] K. Dabiri, M. Malekmohammadi, A. Sheikholeslami, and H. Tamura,
pp. 20–34, Apr. 2019.
‘‘Replica exchange MCMC hardware with automatic temperature selec-
[6] J. Xue and B. Shen, ‘‘A novel swarm intelligence optimization approach:
tion and parallel trial,’’ IEEE Trans. Parallel Distrib. Syst., vol. 31, no. 7,
Sparrow search algorithm,’’ Syst. Sci. Control Eng., vol. 8, no. 1,
pp. 1681–1692, Jul. 2020.
pp. 22–34, Jan. 2020.
[30] N. Rokbani, R. Kumar, A. Abraham, A. M. Alimi, H. V. Long,
[7] Y. Shi, ‘‘Brain storm optimization algorithm,’’ in Proc. Int. Conf. Swarm I. Priyadarshini, and L. H. Son, ‘‘Bi-heuristic ant colony optimization-
Intell., 2011, pp. 303–309. based approaches for traveling salesman problem,’’ Soft Comput., vol. 25,
[8] S. Mirjalili and A. Lewis, ‘‘The whale optimization algorithm,’’ Adv. Eng. no. 5, pp. 3775–3794, Mar. 2021.
Softw., vol. 95, pp. 51–67, May 2016. [31] Y. Saji and M. Barkatou, ‘‘A discrete bat algorithm based on Lévy flights
[9] L. M. Abualigah, A. T. Khader, and E. S. Hanandeh, ‘‘Hybrid clustering for Euclidean traveling salesman problem,’’ Expert Syst. Appl., vol. 172,
analysis using improved krill herd algorithm,’’ Appl. Intell., vol. 48, pp. 114639–114650, Jun. 2021.
no. 11, pp. 4047–4071, 2018. [32] X. Li, L. Gong, X. Liu, F. Jiang, W. Shi, L. Fan, H. Gao, R. Li, and J. Xu,
[10] X. Zhou, X. Zhao, and Y. Liu, ‘‘A multiobjective discrete bat algorithm for ‘‘Solving the last mile problem in logistics: A mobile edge computing
community detection in dynamic networks,’’ Appl. Intell., vol. 48, no. 9, and blockchain-based unmanned aerial vehicle delivery system,’’ Con-
pp. 3081–3093, Sep. 2018. currency Comput., Pract. Exper., Nov. 2020, Art. no. e6068.
[11] S. Kumaravel and V. Ponnusamy, ‘‘An efficient hybrid technique [33] G. Laporte, ‘‘The traveling salesman problem: An overview of exact and
for power flow management in smart grid with renewable energy approximate algorithms,’’ Eur. J. Oper. Res., vol. 59, no. 2, pp. 231–247,
resources,’’ Energy Sour. A, Recovery, Utilization, Environ. Effects, Jun. 1992.
pp. 1–21, Dec. 2020. [34] H. Zhou, M. Song, and W. Pedrycz, ‘‘A comparative study of improved
[12] B. Liu and D. Rodriguez, ‘‘Renewable energy systems optimization by GA and PSO in solving multiple traveling salesmen problem,’’ Appl. Soft
a new multi-objective optimization technique: A residential building,’’ Comput., vol. 64, pp. 564–580, Mar. 2018.
J. Building Eng., vol. 35, Mar. 2021, Art. no. 102094. [35] S. Yuan, B. Skinner, S. Huang, and D. Liu, ‘‘A new crossover approach
[13] G. Liu, C. Shu, Z. Liang, B. Peng, and L. Cheng, ‘‘A modified sparrow for solving the multiple travelling salesmen problem using genetic algo-
search algorithm with application in 3D route planning for UAV,’’ Sen- rithms,’’ Eur. J. Oper. Res., vol. 228, no. 1, pp. 72–82, Jul. 2013.
sors, vol. 21, no. 4, p. 1224, Feb. 2021. [36] Y. Lin, Z. Bian, and X. Liu, ‘‘Developing a dynamic neighborhood struc-
[14] T. Liu, Z. Yuan, L. Wu, and B. Badami, ‘‘Optimal brain tumor ture for an adaptive hybrid simulated annealing–tabu search algorithm to
diagnosis based on deep learning and balanced sparrow search algo- solve the symmetrical traveling salesman problem,’’ Appl. Soft Comput.,
rithm,’’ Int. J. Imag. Syst. Technol., vol. 31, no. 4, pp. 1921–1935, vol. 49, pp. 937–952, Dec. 2016.
Dec. 2021. [37] M. A. Lones, ‘‘Mitigating metaphors: A comprehensible guide to recent
[15] C. Peng, Z. Xu, and M. Mei, ‘‘Applying aspiration in local search for nature-inspired algorithms,’’ Social Netw. Comput. Sci., vol. 1, no. 1,
satisfiability,’’ PLoS ONE, vol. 15, no. 4, Apr. 2020, Art. no. e0231702. pp. 1–12, Jan. 2020.
[16] J. Yuan, Z. Zhao, Y. Liu, B. He, L. Wang, B. Xie, and Y. Gao, ‘‘DMPPT [38] K. Sörensen, ‘‘Metaheuristics—The metaphor exposed,’’ Int. Trans. Oper.
control of photovoltaic microgrid based on improved sparrow search Res., vol. 22, no. 1, pp. 3–18, Jan. 2015.
algorithm,’’ IEEE Access, vol. 9, pp. 16623–16629, 2021. [39] I. Khan and M. K. Maiti, ‘‘A swap sequence based artificial bee
[17] J. Zhang, K. Xia, Z. He, Z. Yin, and S. Wang, ‘‘Semi-supervised ensemble colony algorithm for traveling salesman problem,’’ Swarm Evol. Comput.,
classifier with improved sparrow search algorithm and its application in vol. 44, pp. 428–438, Feb. 2019.
pulmonary nodule detection,’’ Math. Problems Eng., vol. 2021, pp. 1–18, [40] S. Kim and I. Moon, ‘‘Traveling salesman problem with a drone sta-
Feb. 2021. tion,’’ IEEE Trans. Syst., Man, Cybern., Syst., vol. 49, no. 1, pp. 42–52,
[18] J. Zhou and S. Wang, ‘‘A carbon price prediction model based on the May 2019.
secondary decomposition algorithm and influencing factors,’’ Energies, [41] G. N. Zhukova, M. V. Ul’yanov, and M. I. Fomichev, ‘‘A hybrid exact
vol. 14, no. 5, p. 1328, Mar. 2021. algorithm for the asymmetric traveling salesman problem: Construction
[19] K. K. Bhattacharjee and S. P. Sarmah, ‘‘Modified swarm intelligence and a statistical study of computational efficiency,’’ Autom. Remote Con-
based techniques for the knapsack problem,’’ Int. J. Speech Technol., trol, vol. 80, no. 11, pp. 2054–2067, Nov. 2019.
vol. 46, no. 1, pp. 158–179, Jan. 2017. [42] Y. Zhou, R. Wang, C. Zhao, Q. Luo, and M. A. Metwally, ‘‘Dis-
[20] R. Dewangan, A. Shukla, and W. Godfrey, ‘‘Three dimensional path crete greedy flower pollination algorithm for spherical traveling sales-
planning using grey wolf optimizer for UAVs,’’ Appl. Intell., vol. 49, no. 6, man problem,’’ Neural Comput. Appl., vol. 31, no. 7, pp. 2155–2170,
pp. 2201–2217, 2019. Jul. 2019.
[21] G. Serpen and C. Dou, ‘‘Automated robotic parking systems: Real-time, [43] Y. Zhong, L. Wang, M. Lin, and H. Zhang, ‘‘Discrete pigeon-inspired
concurrent and multi-robot path planning in dynamic environments,’’ Int. optimization algorithm with metropolis acceptance criterion for large-
J. Speech Technol., vol. 42, no. 2, pp. 231–251, Mar. 2015. scale traveling salesman problem,’’ Swarm Evol. Comput., vol. 48,
[22] A. Al-Adwan, A. Sharieh, and B. A. Mahafzah, ‘‘Parallel heuristic local pp. 134–144, Aug. 2019.
search algorithm on OTIS hyper hexa-cell and OTIS mesh of trees [44] L. Zhao, S. Wang, Y. Hao, and Y. Wang, ‘‘Energy-dependent mission
optoelectronic architectures,’’ Int. J. Speech Technol., vol. 49, no. 2, planning for agile earth observation satellite,’’ J. Aerosp. Eng., vol. 32,
pp. 661–688, Feb. 2019. no. 1, Jan. 2019, Art. no. 04018118.
[23] T. Kobayashi, ‘‘Student-t policy in reinforcement learning to acquire [45] K. Yang, X. You, S. Liu, and H. Pan, ‘‘A novel ant colony optimization
global optimum of robot control,’’ Int. J. Speech Technol., vol. 49, no. 12, based on game for traveling salesman problem,’’ Int. J. Speech Technol.,
pp. 4335–4347, Dec. 2019. vol. 50, no. 12, pp. 4529–4542, Dec. 2020.

VOLUME 9, 2021 153469

 C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

[46] I. Khan, M. K. Maiti, and K. Basuli, ‘‘Multi-objective traveling salesman [69] A. Antoniadis, K. Fleszar, R. Hoeksma, and K. Schewior, ‘‘A PTAS
problem: An ABC approach,’’ Int. J. Speech Technol., vol. 50, no. 11, for Euclidean TSP with hyperplane neighborhoods,’’ ACM Trans. Algo-
pp. 3942–3960, Nov. 2020. rithms, vol. 16, no. 3, pp. 1–16, Jun. 2020.
[47] S. Verma, J. J. Jena, S. C. Satapathy, and M. Rout, ‘‘Solving travelling [70] M. Mavrovouniotis, F. M. Müller, and S. Yang, ‘‘Ant colony optimization
salesman problem using discreet social group optimization,’’ J. Sci. Ind. with local search for dynamic traveling salesman problems,’’ IEEE Trans.
Res., vol. 79, no. 10, pp. 928–930, 2020. Cybern., vol. 47, no. 7, pp. 1743–1756, Jul. 2017.
[48] T.-S. Pham, P. Leyman, and P. De Causmaecker, ‘‘The intermittent [71] M. Drexl and M. Schneider, ‘‘A survey of variants and extensions
travelling salesman problem,’’ Int. Trans. Oper. Res., vol. 27, no. 1, of the location-routing problem,’’ Eur. J. Oper. Res., vol. 241, no. 2,
pp. 525–548, Jan. 2020. pp. 283–308, 2015.
[49] A. F. Tuani, E. Keedwell, and M. Collett, ‘‘Heterogenous adaptive ant [72] M. J. Santos, P. Amorim, A. Marques, A. Carvalho, and A. Póvoa,
colony optimization with 3-opt local search for the travelling salesman ‘‘The vehicle routing problem with backhauls towards a sustainability
problem,’’ Appl. Soft Comput., vol. 97, Dec. 2020, Art. no. 106720. perspective: A review,’’ Top, vol. 28, no. 2, pp. 358–401, Jul. 2020.
[50] D.-H. Tran, T. X. Vu, S. Chatzinotas, S. ShahbazPanahi, and B. Ottersten, [73] D. R. Viloria, E. L. Solano-Charris, A. Muñoz-Villamizar, and
‘‘Coarse trajectory design for energy minimization in UAV-enabled,’’ J. R. Montoya-Torres, ‘‘Unmanned aerial vehicles/drones in vehicle rout-
IEEE Trans. Veh. Technol., vol. 69, no. 9, pp. 9483–9496, Sep. 2020. ing problems: A literature review,’’ Int. Trans. Oper. Res., vol. 28, no. 4,
[51] B. C. H. Silva, I. F. C. Fernandes, M. C. Goldbarg, and E. F. G. Goldbarg, pp. 1626–1657, Jul. 2021.
‘‘Quota travelling salesman problem with passengers, incomplete ride and [74] H. Liu, P. Zhang, B. Hu, and P. Moore, ‘‘A novel approach to task
collection time optimization by ant-based algorithms,’’ Comput. Oper. assignment in a cooperative multi-agent design system,’’ Appl. Intell.,
Res., vol. 120, Aug. 2020, Art. no. 104950. vol. 43, no. 1, pp. 162–175, 2015.
[52] K. Saito, M. Aono, and S. Kasai, ‘‘Amoeba-inspired analog electronic [75] M. C. de Oliveira, M. R. Delgado, and A. Britto, ‘‘A hybrid greedy
computing system integrating resistance crossbar for solving the travel- indicator-and Pareto-based many-objective evolutionary algorithm,’’
ling salesman problem,’’ Sci. Rep., vol. 10, no. 1, pp. 1–9, Dec. 2020. Appl. Intell., pp. 1–23, Jan. 2021.
[53] D. C. Popescu and P. Kilby, ‘‘Approximation of the Shapley value for the [76] T. Zhang, Q. Zeng, and X. Zhao, ‘‘Optimal local dimming based on an
Euclidean travelling salesman game,’’ Ann. Oper. Res., vol. 289, no. 2, improved greedy algorithm,’’ Int. J. Speech Technol., vol. 50, no. 12,
pp. 341–362, Jun. 2020. pp. 4162–4175, Dec. 2020.
[54] M. Cavaleri and A. Donno, ‘‘Distance-balanced graphs and travelling [77] B. Cazaux and E. Rivals, ‘‘The power of greedy algorithms for approxi-
salesman problems,’’ Ars Mathematica Contemporanea, vol. 19, no. 2, mating max-ATSP, cyclic cover, and superstrings,’’ Discrete Appl. Math.,
pp. 311–324, Nov. 2020. vol. 212, pp. 48–60, Oct. 2016.
[55] F. Carrabs, C. Cerrone, R. Cerulli, and B. Golden, ‘‘An adaptive [78] B. Cazaux and E. Rivals, ‘‘Relationship between superstring and com-
heuristic approach to compute upper and lower bounds for the close- pression measures: New insights on the greedy conjecture,’’ Discrete
enough traveling salesman problem,’’ Informs J. Comput., vol. 32, no. 4, Appl. Math., vol. 245, pp. 59–64, Aug. 2018.
pp. 1030–1048, Sep. 2020. [79] Y. H. Sun, ‘‘Brain storm optimization using a slight relaxation selec-
[56] V. Cacchiani, C. Contreras-Bolton, and P. Toth, ‘‘Models and algorithms tion and multi-population based creating ideas ensemble,’’ Appl. Intell.,
for the traveling salesman problem with time-dependent service times,’’ vol. 50, pp. 3137–3161, May 2020.
Eur. J. Oper. Res., vol. 283, no. 3, pp. 825–843, Jun. 2020. [80] A. K. Das and D. K. Pratihar, ‘‘A directional crossover (DX) operator
[57] C. Wu and X. Fu, ‘‘An agglomerative greedy brain storm optimization for real parameter optimization using genetic algorithm,’’ Int. J. Speech
algorithm for solving the TSP,’’ IEEE Access, vol. 8, pp. 201606–201621, Technol., vol. 49, no. 5, pp. 1841–1865, May 2019.
2020. [81] X. Dong and Y. Cai, ‘‘A novel genetic algorithm for large scale colored
[58] M. Abbasi, M. Rafiee, M. R. Khosravi, A. Jolfaei, V. G. Menon, and balanced traveling salesman problem,’’ Future Gener. Comput. Syst.,
J. M. Koushyar, ‘‘An efficient parallel genetic algorithm solution for vol. 95, pp. 727–742, Jun. 2019.
vehicle routing problem in cloud implementation of the intelligent trans- [82] A. Khanra, T. Pal, M. K. Maiti, and M. Maiti, ‘‘Multi-objective four
portation systems,’’ J. Cloud Comput., vol. 9, no. 1, pp. 1–14, Dec. 2020. dimensional imprecise TSP solved with a hybrid multi-objective ant
[59] J. Zhang, L. Hong, and Q. Liu, ‘‘An improved whale optimization algo- colony optimization-genetic algorithm with diversity,’’ J. Intell. Fuzzy
rithm for the traveling salesman problem,’’ Symmetry, vol. 13, no. 1, p. 48, Syst., vol. 36, no. 1, pp. 47–65, Feb. 2019.
Dec. 2021. [83] I. Chaouch, O. B. Driss, and K. Ghedira, ‘‘A novel dynamic assignment
[60] I. Zelinka and S. Das, ‘‘Gamesourcing: An unconventional tool to assist rule for the distributed job shop scheduling problem using a hybrid ant-
the solution of the traveling salesman problem,’’ Natural Comput., no. 2, based algorithm,’’ Int. J. Speech Technol., vol. 49, no. 5, pp. 1903–1924,
pp. 1–11, Nov. 2020. May 2019.
[61] M. Yousefikhoshbakht, ‘‘Solving the traveling salesman problem: [84] J. Liu, H. Peng, Z. Wu, J. Chen, and C. Deng, ‘‘Multi-strategy brain
A modified metaheuristic algorithm,’’ Complexity, vol. 2021, pp. 1–13, storm optimization algorithm with dynamic parameters adjustment,’’ Int.
Feb. 2021. J. Speech Technol., vol. 50, no. 4, pp. 1289–1315, Apr. 2020.
[62] Z. Wu, J. Wu, M. Zhao, L. Feng, and K. Liu, ‘‘Two-layered ant colony [85] S. Gupta and K. Deep, ‘‘A novel hybrid sine cosine algorithm for global
system to improve engraving robot’s efficiency based on a large-scale optimization and its application to train multilayer perceptrons,’’ Int. J.
TSP model,’’ Neural Comput. Appl., vol. 2020, pp. 1–11, Jun. 2021. Speech Technol., vol. 50, no. 4, pp. 993–1026, Apr. 2020.
[63] S. A. Vasquez, G. Angulo, and M. A. Klapp, ‘‘An exact solution method [86] A. C. Cinar, S. Korkmaz, and M. S. Kiran, ‘‘A discrete tree-seed algorithm
for the TSP with Drone based on decomposition,’’ Comput. Oper. Res., for solving symmetric traveling salesman problem,’’ Eng. Sci. Technol.,
vol. 127, pp. 105–127, Mar. 2021. Int. J., vol. 23, no. 4, pp. 879–890, Aug. 2020.
[64] Y. Sun, A. Ernst, X. Li, and J. Weiner, ‘‘Generalization of machine [87] M. S. Kıran, H. İşcan, and M. Gündüz, ‘‘The analysis of discrete artificial
learning for problem reduction: A case study on travelling salesman bee colony algorithm with neighborhood operator on traveling salesman
problems,’’ OR Spectr., pp. 1–27, Sep. 2021. problem,’’ Neural Comput. Appl., vol. 23, no. 1, pp. 9–21, 2013.
[65] A. Stieber and A. Fügenschuh, ‘‘Dealing with time in the multiple trav- [88] M. Neshat, G. Sepidnam, M. Sargolzaei, and A. N. Toosi, ‘‘Artificial
eling salespersons problem with moving targets,’’ Central Eur. J. Oper. fish swarm algorithm: A survey of the state-of-the-art, hybridization,
Res., no. 3, pp. 1–27, Oct. 2020. combinatorial and indicative applications,’’ Artif. Intell. Rev., vol. 42,
[66] J. C. N. Silva, I. M. Coelho, U. S. Souza, L. S. Ochi, and V. N. Coelho, no. 4, pp. 965–997, 2014.
‘‘Finding the maximum multi improvement on neighborhood explo- [89] E. Osaba, X.-S. Yang, F. Diaz, P. Lopez-Garcia, and R. Carballedo,
ration,’’ Optim. Lett., no. 1, pp. 1–19, Feb. 2020. ‘‘An improved discrete bat algorithm for symmetric and asymmetric
[67] Y. V. Salii and A. S. Sheka, ‘‘Improving dynamic programming for traveling Salesman problems,’’ Eng. Appl. Artif. Intel., vol. 48, pp. 59–71,
travelling salesman with precedence constraints: Parallel Morin–Marsten Feb. 2016.
bounding,’’ Optim. Methods Softw., pp. 1–27, Sep. 2020. [90] E. Osaba, J. Del Ser, A. Sadollah, M. N. Bilbao, and D. Camacho, ‘‘A dis-
[68] M. A. H. Akhand, S. I. Ayon, S. A. Shahriyar, N. Siddique, and H. Adeli, crete water cycle algorithm for solving the symmetric and asymmetric
‘‘Discrete spider monkey optimization for travelling salesman problem,’’ traveling salesman problem,’’ Appl. Soft Comput., vol. 71, pp. 277–290,
Appl. Soft Comput., vol. 86, Jan. 2020, Art. no. 105887. Oct. 2018.

153470 VOLUME 9, 2021

C. Wu et al.: Novel Sparrow Search Algorithm for Traveling Salesman Problem

[91] H. Zhu, X. You, and S. Liu, ‘‘Multiple ant colony optimization based on XISONG FU received the B.E. degree in informa-
Pearson correlation coefficient,’’ IEEE Access, vol. 7, pp. 61628–61638, tion management and information system from the
2019. Institute of Technology, East China Jiaotong Uni-
[92] T.-T. Nguyen, Y. Qiao, J.-S. Pan, S.-C. Chu, K.-C. Chang, X. Xue, versity, China, in 2019. He is currently pursuing
and T.-K. Dao, ‘‘A hybridized parallel bats algorithm for combinatorial the master’s degree with the Engineering College,
problem of traveling salesman,’’ J. Intell. Fuzzy Syst., vol. 38, no. 5, Shandong Institute of Business and Technology.
pp. 5811–5820, May 2020. His current research interest includes swarm
[93] Y. Saji and M. Barkatou, ‘‘A discrete bat algorithm based on Lévy flights
intelligence optimization algorithm and its
for Euclidean traveling salesman problem,’’ Expert Syst. Appl., vol. 172,
application.
Jun. 2021, Art. no. 114639.
[94] P. Wang, Y. Zhang, and H. Yang, ‘‘Research on economic optimization
of microgrid cluster based on chaos sparrow search algorithm,’’ Comput.
Intell. Neurosci., vol. 2021, no. 3, pp. 1–18, Mar. 2021.
[95] Y. Wu, Z. Zhang, R. Xiao, P. Jiang, Z. Dong, and J. Deng, ‘‘Operation
state identification method for converter transformers based on vibration
detection technology and deep belief network optimization algorithm,’’
Actuators, vol. 10, no. 3, p. 56, Mar. 2021.
[96] D. Pengzhen, T. Zhenmin, and S. Yan, ‘‘An object-oriented multi-role ant
colony optimization algorithm for solving TSP problem,’’ Control Decis.,
vol. 29, no. 10, pp. 1729–1736, 2014. JUNKE PEI received the B.E. degree in infor-
[97] U. Ashraf, J. Liang, and A. Akhtar, ‘‘Meta-heuristic hybrid algorithmic mation management and information system from
approach for solving combinatorial optimization problem,’’ in Proc. Int. Liaocheng University, China, in 2020. She is cur-
Conf. Bio-Inspired Comput., Appl., 2020, pp. 645–656. rently pursuing the master’s degree with the Engi-
[98] D. Qiao, P. Jie, L. Hao, and W. Xiaoyu, ‘‘Research on improving ant neering College, Shandong Institute of Business
colony algorithm to solve TSP problem,’’ Machinery Design Manuf.,
and Technology.
no. 10, pp. 144–149, 2019.
[99] G. Şaban, M. Mostafa, B. O. Kaan, and H. Kodaz, ‘‘A parallel cooperative
Her current research interest includes intelligent
hybrid method based on ant colony optimization and 3-Opt algorithm optimization algorithm and its application.
for solving traveling salesman problem,’’ Soft Comput., vol. 22, no. 5,
pp. 1669–1685, Mar. 2018.
[100] Y. Xu, Y. Wu, Y. Fu, X. Wang, and A. Lu, ‘‘Discrete brain storm optimiza-
tion algorithm based on prior knowledge for traveling salesman prob-
lems,’’ in Proc. 13rd IEEE Conf. Ind. Electron. Appl. (ICIEA), May 2018,
pp. 2740–2745.
[101] Y. Wu, X. Wang, J. Qi, and L. Huang, ‘‘An adaptive brain storm optimiza-
tion algorithm based on heuristic operators for TSP,’’ in Proc. Int. Conf.
Bio-Inspired Comput., Appl. Singapore: Springer, 2019, pp. 662–672.
ZHIGUI DONG received the B.E. and M.E.
CHANGYOU WU received the B.E. and M.E. degrees in industrial engineering and manage-
degrees in industrial engineering and manage- ment science engineering and the Ph.D. degree in
ment science engineering and the Ph.D. degree in agricultural system engineering and management
mechanized engineering from Northeast Agricul- engineering from Northeast Agricultural Univer-
tural University, China, in 2004, 2006, and 2009, sity, China, in 2005, 2008, and 2018, respectively.
respectively. He is currently an Associate Professor with
He is currently an Associate Professor with the the Liaoning Institute of Science and Technology.
Engineering College, Shandong Institute of Busi- He published over 20 articles in domestic and
ness and Technology. He published over 30 articles international academic journals and conference
in domestic and international academic journals proceedings. His current research interests include neural network theory
and conference proceedings. His current research interests include genetic and its application, optimization algorithm theory and its application, big
algorithm theory and its application, neural network theory and its applica- data technology and application, and agricultural big data.
tion, parallel computing, and water resources management.

VOLUME 9, 2021 153471