Array 10 (2021) 100062

Contents lists available at ScienceDirect

Array
journal homepage: www.elsevier.com/journals/array/2590-0056/open-access-journal

A multi-objective parameter-tuned soft computing-based algorithm to

optimize competitive congested location-pricing problem within
multi-type service
Parviz Fattahi a, Vahid Hajipour b, *, Sara Hajiloo c
a
Department of Industrial Engineering, Faculty of Engineering, Alzahra University, Tehran, Iran
b
Department of Industrial Engineering, Islamic Azad University, West Tehran Branch, Tehran, Iran
c
Industrial Engineering Department, Faculty of Engineering, Bu-Ali Sina University, Hamedan, Iran

A R T I C L E I N F O A B S T R A C T

Keywords: One of the issues that has attracted many researchers in the last decade is the problem of locating facilities i.e.
Soft computing-based algorithms hospitals, shops, banks and ATMs. One of the basic needs of the people of the community is easy access to the
Competitive facility location facilities, so that by spending little time can reach to the facility and with spending low cost to receive their
Pricing
facilities. In this paper, the location-pricing problem of the congested facilities by considering competition be-
Queuing theor\
tween available facilities and new facilities were investigated and bi-objective non-linear mathematical model
that follow from M/M/m/k queuing system, was presented. In the first goal, maximize the profit of system by
minimizing the total cost of establishing the facilities, shipping costumers and expectation time of the costumers
in the queue and in the second goal the share of facility market minimized. The proposed model is in the category
of non-linear integer programming problems, that, due to the complexity of the problem in the large scales and in
order to solve the model, different approaches such as multi-objective meta-heuristic algorithms including non-
dominated sorting genetic algorithm (NSGA-II) and multi-objective particle swarm optimization (MOPSO) has
been presented. At the end, by applying the Taguchi method, the efficiency performance of NSGA-II algorithm
perform better than MOPSO.

1. Introduction the first time, Berman and Larson [1] set the optimal location for service
providers in a queuing network with the M/G/1 structure. Wang et al. [2]
Nowadays, the combination of facility location problems with other presented several models for equipment with capacity constraints
approaches, such as structuring, pricing, competition, etc., has become studying the application of locating service providers in communication
more prominent and attracted the attention of researchers. The estab- networks and ATMs. In fact, these models have been expanded for situ-
lishment of facilities among the previous facilities has always been along ations where fixed service locations, constraint service capacities, and
with competition, so each facility compete to gain more of the market random demand exist. Berman et al. [3] presented the M/M/1/k queue
share. In the case of this research, customer demand depends on price model for location problems. Assuming that the number of equipment
and distance parameters, so paying attention to pricing policies is selected has a queuing length constraint to k and a percentage of
important. Adopting an appropriate pricing policy control the arrival rate customer demand due to model constraints may be lost. The paper ana-
and, consequently, lead to controlling the queue length, which take into lyzes the experiments using nine innovative algorithms. Fernandez and
account the above factors, and increase the level of customer satisfaction. Hendrix [4] combine the concept of location, pricing, and queue with the
In problems with one of the decision-pricing variables, the objective single-facilitation location problems to maximize the benefits of the fa-
function is maximizing the profit of the system, while in queuing prob- cility, and concluded that customer demand rates depend on price, dis-
lems objectives are usually devoted to minimizing costs. tance, and waiting time depends on the facility and provides an
The various applications of combining facility location problem with algorithm for solving the presented model. Zarrinpoor and Seifbarghy [5]
the queuing structure in the real world have become very important. For developed a competitive model with a queuing structure M/M/m/k with

* Corresponding author.
E-mail addresses: [email protected] (P. Fattahi), [email protected], [email protected] (V. Hajipour), [email protected] (S. Hajiloo).

https://doi.org/10.1016/j.array.2021.100062
Received 25 September 2019; Received in revised form 15 October 2020; Accepted 21 March 2021
Available online 26 March 2021
2590-0056/© 2021 Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
P. Fattahi et al. Array 10 (2021) 100062

the aim of gaining a certain percentage of market share and minimizing model with an M/M/m/k queue system was proposed in which the
total costs; to solve the proposed model, two meta-heuristics genetic al- number of service providers was fixed and the demand was random.
gorithm and greedy search have been used. Since the problem of locating facilities is one of the hard ones, a
Location problems can be categorized in different ways and based on multi-objective optimization algorithm based on the theory of vibrations
various features. Some of these categories are similar to the traditional is utilized to solve the model. In this bi-objective model the profit
categories in locational problems. The first study on competitive location generated by the simulation system was maximized and the total waiting
was made by Hotelling in Ref. [6]. To gain more market share, companies time for customers in the queue was minimized. Gorji et al. [13] provided
need to focus on the needs and expectations of customers, which is one of a model for a competitive leader-follower facility location problem the in
the most important features in the customer’s point of view. During their a discrete space. The purpose of this study is to review the optimal lo-
research, researchers have described different competitive qualities as cations for new facilities establishment and, in addition, seek to increase
important factors for customers that have a significant impact on supply their market share. Table 1 shows an overview of the most important
chain competitiveness. Types of studies show that the price that retailers researches in the field of facility location problem in recent years. In this
consider for the product can affect the demand of each competitor. Dis- paper, the problem of competitive location-pricing is investigated in
tance is also considered as an important competitive feature. In recent which the facilities provides multiple service in congestion. The problem
studies the price is also regarded as an important factor. However, in is a bi-objective mathematical model in order to minimize total costs
several studies by Fernandez et al. [7]; distance is considered as an along with maximizing service levels to customers. The existence of
important factor as price. Service level is another important feature of contradictions in the objectives and business methods makes it difficult
competition. Although the description of the service level varies in to achieve these objectives. An organization can only hope to be suc-
different studies. cessful in its survival in competitive markets, hence the pricing policies
Berman et al. [8] introduced the location-pricing problem to maxi- can be considered as one of the most important decisions in a competitive
mize total system profits, then expanded their study considering several environments. To do so, new facilities compete for gaining market share
facilities. They studied the location of m facilities on a network with n with previously existing facilities.
demand points considering that customers can make their decision In Section 2, the problem statement and the mathematical formula-
before logging in by viewing the length of the queue. The offer price is tion are described. For validation of the proposed model, numerical ex-
considered the same for all the facilities and the queue system is amples are presented in Section 4. Finally, the conclusions and
considered to be M/M/1. suggestions are presented in Section 5.
One of the most important researches in this field is carried out by
Hajipour et al. [9,10]; who proposed a multi-objective multi-layer 2. Problem formulation
location-allocation model with M/M/1 queuing system. To solve their
model, they proposed two multi-objective meta-heuristic algorithms, one The problem of facilities locating-pricing is one of the important
based on the multi-objective optimization vibration damping, and the location problems in which finding a place to establish a number of fa-
other based on the multi-objective harmony search algorithm. Magh- cilities among potential points is studied; the establishment of facilities
soudlou et al. [11] introduced a bi-objective model for optimizing the among former facilities has always been a long-standing competition,
multidirectional 3D multi-server supply chain that follows the M/M/m therefore each facility will compete to gain more market share. In this
queue system and solved it using the MOPSO algorithm. Fernandez et al. problem, we buy the facility; each of these facilities offer different ser-
[12] studied the problem of competitive location in binary and vices, customers will go to each facility according to their needs. Cus-
semi-binary and formulated the choice of a number of new locations for tomers randomly request the facilities they need in a variety of options,
the establishment of facilities among potential points that were previ- and traveling a distance, they face the considered facility. Since the
ously established companies and solved this They attributed the issue to arrival of customers is random, they are often congested and therefore
the behavior of customers, and concluded that increasing market share must be queued for getting services. The queuing system considered is
would only depend on the distance between facilities and customers. influenced by several factors depending on the facility. Fig. 1 shows the
They presented two heuristic algorithms to solve the model and used it to scheme of the problem network structure.
evaluate the classical genetic algorithm and a number of real examples. The assumptions and objectives considered for the problem include:
Hajipour et al. [9,10] presented a two-objective location-pricing

Table 1
An overview of the most important researches on location-pricing problems.
Year Authors Queuing Main Scope Competitive Pricing Solution methods

[14] Aboolian et al. M/G/1 The problem of locating a congested facility ✓ Heuristic
[15] Drezner et al. M/M/k Multiple Serving Location Problem SA, TS, & DESCENT
[5] Zarrinpoor., Seifbarghy M/M/m/k The problem Location-competitive ✓ TS, GA
[16] Pasandideh, Niaki M[x]/M/1 The problem of locating a multi-purpose facility GA & SA
[17] Chambari, Rahmaty M/M/1 Location-allocation issue NSGA-II, & NRGA
[18] Fernandez et al. - Competitive- pricing Location problem ✓ ✓ B&B & Weiszfeld
[19] Vidyarthi, Jayaswal M/G/1 Location-allocation problem CG
[20] Renando et al. - The problem of locating and designing competitive facilities ✓ MOEAD, SPEA2, & NSGA-II
[21] Drezner et al. - Competitive Location problem B&B & TS
[9,10] Hajipour et al. M/M/m/k pricing Location problem ✓ NSGA-II
[9,10] Hajipour et al. M/M/1 Location-allocation problem ✓ NSGA-II & MOSA
[22] Zhou et al. - Competitive Location problem ✓ GA
[23] Tavakkoli-Moghaddam et al. M/M/m/k pricing Location problem ✓ NSGA-II
[24] Nasiri et al. Competitive location ✓ GA & PSO
Problem
[25] Khodemani et al. M/M/c Location problem – queuing framework NSGA-II & HSA
2019 Zamberano-Ret et al. [26] Competitive- pricing Location problem ✓ ✓ PSO
[27] Ahmadi, Ghezavati Competitive location ✓ Benders’ decomposition method
Problem
This Research M/M/m/k Competitive- pricing Location problem ✓ ✓ NSGA-II & MOPSO

2
P. Fattahi et al. Array 10 (2021) 100062

Fig. 1. Facility location problems behaving as M/M/m/k queues.

XX  XX XXX

In order to receive the service, demand points move towards the fa- Maximize ¼ pcj  cscj τcj  fj ycj  τcj π j tcij
cility. This is the type of immobile server. j2N c2R j2N c2R j2N c2R i2M
XX

Each facility has a different service cost.  θj wj τcj (1)

Each facility does not provide all services. There are different service j2N c2R

type in a facility.
P PP

Customers historically choose facilities appropriate to their needs, λcij ðp; dÞxcij
i2M c2Rj2N
and their rate of entry follows the Poisson’s distribution. Maximize ¼ P P P (2)

Each customer receives services only from one facility λcij ðp; dÞxcij
i2M c2Rj2N[N 0

The service time of each facility follows the exponential distribution

The number of server in each facility can be more than one. 0

Queuing system capacity is limited. λcij ðp; dÞ ¼ gci  αci pcj  dij βci 8c 2 R; 8i 2 M; 8j 2 N [ N (3)

X 0
The goal is to establish a facility with the lowest possible cost in a way τcj ¼ λcij ðp; dÞxcij 8c 2 R; 8j 2 N [ N (4)
that the quality of services does not reduce. In addition, the maximum i2M

number of people can have access to the facility so customers can spend a   0
little time to get the desired facilities and get their service at low cost and τcj 1  π kj  mj kj 8j 2 N [ N ; 8c 2 R (5)
low waiting times, and the facilities will maximize their profit and will
  h
compete with other facilities to maximize their market share by applying  k m i
0
X
N[N XR
π 0:j τcj mj rj kj mj þ1  
right policies. Table 2 summarizes the values of the variables, parame- wj ¼  2 1  rj  1  rj kj  mj þ 1 rj j j
mj ! μj 1  rj
ters, and decision variables of the proposed model. j¼1 c¼1

(6)

3
P. Fattahi et al. Array 10 (2021) 100062

Table 2
Summarizes the parameters and decision variables.
X
N
ycj  V 8c 2 R (14)
Parameters j¼1

M set of customer nodes

0
N set of locations for potential facilities ycj 2 f0; 1g 8j 2 N; ycj ¼ 1 8j 2 N ; 8c 2 R
0 0
N Set of locations of the competitor xcij 2 f0; 1g 8i 2 M; 8j 2 N [ N ; 8c 2 R
V (V  N )maximum number of servers which can be on-duty; 0
λcij  0 8i 2 M; 8j 2 N [ N ; 8c 2 R
R Type of services that provide facilities 0

I (i ¼ 1, …,M) customer node index;

pcj  0 8i 2 M; 8j 2 N [ N (15)
0
j 0
(j ¼ 1, …, N [ N ); facility location index wcj  0 8i 2 M; 8j 2 N [ N
0
c (c ¼ 1, …,R) Type of service provided; τcj  0 8i 2 M; 8j 2 N [ N
dij The traveling distance from costumer i to facility node j kj  0; Integer
fj The cost of establishing or purchasing facilities j mj  0; Integer
μcj Rate of service at the node j of service type c
gci Number of users in the customer node i of service type c; (i ¼ 1, …,M & c ¼ 1, The first objective function maximizes the system’s total profit. The
…,R) second objective function minimizes the market share of the previously
cscj The cost of servicing as a kind of service c in facilities j; (i ¼ 1, …,M & c ¼ 1, existing facilities. The explanations related to the constraints of the two
…,R)
objective functions are as follows:
αci The price sensitivity coefficient in the customer node i of service type c; (i ¼ 1,
…,M & c ¼ 1, …,R) The first constraint in Eq. (3) expresses the demand rate of customers.
βci Distance sensitivity coefficient in customer node i of service type c; (i ¼ 1, …,M Constraints (4) show the rate of entry for customers in each facility,
& c ¼ 1, …,R) which is a function of price and distance. Constraints (5) shows the
tcij Travel time between customer nodes i to facilities j of services type c; (i ¼ 1, service capacity in each facility. Constraints (6) shows the waiting time
…,M & c ¼ 1, …,R,& j ¼ 1, …,N))
πj Travel cost to node j per time unit (j ¼ 1, …,N)
for customers in facilities. Constraints (7–10) are equations of the M/M/
θj Waiting cost at node j per time unit (j ¼ 1, …,N) m/k queuing system. The number of customers in the system is a birth-
0
pcj the cost of servicing the open competitor j of services type c; (j ¼ 1, …,N & c ¼ 1, death process for a stable distribution with appropriate rates that can
…,R) be deduced from the main functionality of the queue by a Marcov chain
Decision Variables:
with a corresponding transmission matrix and Little laws. Constraints
xcij if customer i of services j is assigned to facility j 1, otherwise 0; (i ¼ 1, …,M & c
¼ 1, …,R,& j ¼ 1, …,N))
(11) denotes that all demands are met. Constraints (12) shows that all j
ycj if facility j the provision of services c open 1, otherwise 0; (c ¼ 1, …,R,& j ¼ 1, facilities that provide c services can provide R (service type) at most.
…,N)) Constraints (13) indicates that all customers are only receiving services
λcij number of customer i of services c to facility j; (i ¼ 1, …,M & c ¼ 1, …,R,& j ¼ 1, from open facilities. Constraints (14) shows the maximum possible fa-
…,N))
cilities that can be established. Constraints (15) shows the range of de-
τcj the demand rate at open facility node j of services c; (c ¼ 1, …,R,& j ¼ 1, …,N))
pcj the cost of servicing in potential facility j of services c; (c ¼ 1, …,R,& j ¼ 1, …,N) cision variables.
wj The expected waiting time of customer batches assigned to facility node j; (j ¼
1, …,N 3. Problem optimization
mj The number of servers at opened facility j; (j ¼ 1, …,N)
kj Queuing capacity at opened facility j; (j ¼ 1, …,N)
In order to solve the problem at hand, we applied multi-objective
particle swarm optimization (MOPSO) and compare it with the one of
X τcj 0 best-developed Pareto-based algorithm called NSGA-II.
rj ¼ 8j 2 N [ N (7)
c2R
m j μj

3.1. MOPSO
1 0
π O;j ¼ "  s  mj # 8j 2 N [ N (8)
mP
j 1 P
R P
R P
kj Due to the complexity of the proposed model and the spatial config-
τcj 1 τcj 1
μj s!
þ μj mj !
rj smj uration structure of the PSO algorithm (introduced by Ref. [28], which is
s¼0 c¼1 c¼1 s¼mj
capable of solving the problem, it is used. The PSO algorithm is based on
Xτcj  1 1 0
swarm intelligence and maintains the balance between exploration and
C kj ¼ 8j 2 N [ N (9) exploitation. Also, we found PSO gives very fast and accurate solution.
c2R
μj m j ! m
kj mj
j Added to this, PSO requires less iteration to obtain global optimum
compare to other algorithms (See Section 3). We have compared the
0
π kj ¼ CKJ π 0;J 8j 2 N [ N (10) proposed MOPSO with the best-developed multi-objective evolutionary
algorithm named NSGA-II.
X
xcij ¼ 1 8c 2 R; 8i 2 M (11)
0
j2N[N 3.2. NSGA-II

X
R
0
As outlined in the introduction, the multi-objective competitive
ycj  R 8j 2 N [ N (12) location-pricing problem is a NP-hard problem and has two objective
c¼1
functions. Therefore, the exact solution methods cannot be used to solve
0
large-size problems. Since multi-objective meta-heuristic genetic algo-
xcij  ycj 8c 2 R; 8i 2 M; 8j 2 N [ N (13) rithm has proven to be acceptable in solving several problems with non-

Table 3
The first part uses the answer vector.
Chromosome Structure 1st Candidate Point 2nd Candidate point 3rd Candidate point 4th Candidate point 5th Candidate point

1st Service 203.18 563.09 942.55 342.83 603.54

2nd Service 661.04 729.34 145.24 735.03 524.26

4
P. Fattahi et al. Array 10 (2021) 100062

Table 4
The second part uses the answer vector.
Chromosome Structure 1st Candidate 2nd Candidate 3rd Candidate 4th Candidate 5th Candidate 1st 2nd 3rd
Point point point point point Facilitate Facilitate Facilitate

1st 1st 0.81 0.15 0.65 0.70 0.43 0.27 0.75 0.84
Service customer
2nd 0.90 0.97 0.03 0.03 0.38 0.67 0.25 0.25
customer
3rd 0.12 0.95 0.84 0.27 0.76 0.65 0.50 0.81
customer
4th 0.91 0.48 0.93 0.04 0.18 0.16 0.69 0.24
customer
5th 0.63 0.80 0.67 0.09 0.48 0.11 0.89 0.92
customer
2nd 1st 0.09 0.14 0.75 0.82 0.44 0.49 0.35 0.95
Service customer
2nd 0.27 0.42 0.74 0.69 0.64 0.95 0.54 0.19
customer
3rd 0.54 0.91 0.39 0.31 0.79 0.34 0.13 0.25
customer
4th 0.95 0.79 0.65 0.95 0.70 0.58 0.14 0.61
customer
5th 0.96 0.95 0.17 0.03 0.75 0.22 0.25 0.47
customer

Fig. 2. The sample solution representation.

dominated sorting, it can be an appropriate option for solving this the proposed problems in Hajipour et al. [9,10] are used. The problems
problem. NSGA-II is one the best-developed Pareto-based algorithms are made using random quantities, except that depending on the nature
introduced by Deb et al. [29]. In this paper, due to the high number of of the model some changes were made in defining some of the parame-
decision variables, a five-part chromosome has been used to determine ters, as well as the possibility distributions of new parameters are
the numerical values of each of these variables. The first to second sec- introduced. The distribution of each of the parameters are:
tions of the solution vector used in this problem are presented in
Tables 3–4, and third to fifth sections of the solution vector used in this
The establishment or purchase cost of the jth new facility, fj follows the
problem are presented in Fig. 2, respectively. continuous uniform distribution in the range of [10,000, 15,000].
The NSGA-II and MOPSO algorithms are used the illustrated solution
representation in order to solve the proposed model.

4. Results analysis Table 6

The levels of parameters in the MOPSO algorithm.
To analyze the proposed model, 30 experimental problems based on Parameters Proposed parameters values

Small Size Medium Size Large Size

Table 5 nPop 75 50 75
nRep 50 50 50
The levels of parameters in the NSGA-II algorithm.
W 0.5 0.5 0.5
Parameters Proposed parameters values C1 0.1 0.1 0.1
C2 0.1 0.1 0.1
Small Size Medium Size Large Size
nGrid 5 7 7
nPop 75 100 100 Alpha 0.2 0.2 0.3
pCrossover 0.8 0.8 0.8 Beta 5 5 5
pMutation 0.2 0.2 0.2 Gamma 3 5 5

5
P. Fattahi et al. Array 10 (2021) 100062

The service cost for each unit of service C, CS j follows the continuous
The waiting cost in jth node for each unit of time θ j follows the
uniform distribution in the range of [100, 500]. continuous uniform distribution in the range of [1, 5].

Price sensitivity coefficient αcj and the distance Bcj in the nodes of
The travel time from node i to facility j, tij and also the travel time to j
customers follows the uniform distribution in range of [1, 10] for all node for time unit of πj both follow the continuous uniform distri-
different services. bution in the range of [0.1, 1].

The distance between ith customer and jth facility follows the contin-
The number of service providers in rival facility mj & j 2 N0 follow the
uous uniform distribution in the range of [100, 500]. discrete uniform distribution in the range of [0, 20].

The number of service users in all gci nodes follow the discrete uni-
The queue capacity of the rival facility k j & j 2 N0 follow the discrete
form distribution in the range of [5000, 10,000]. uniform distribution in the range of [0, 20].

The service rate in jth node follows the continuous uniform distribu-
The service price of the rival facility Pcj & j 2 N0 follow the continuous
tion in the range of [100, 1000]. uniform distribution in the range of [0, 700].

Fig. 3. Main effects plot for means in NSGA-II.

Fig. 4. Main Effect Plot for SN ratio in NSGA-II.

6
P. Fattahi et al. Array 10 (2021) 100062

Fig. 5. Main effects plot for means in MOPSO

Fig. 6. Main Effects for SN ratio in MOPSO.

The results of parameter setting of NSGA-II and MOPSO algorithms the basis of these indices. The results of calculating and comparing the
using Taguchi method in numerical examples are presented in indices are presented in Table 7. The highlighted values actually repre-
Tables 5–6. sent the best results obtained in each problem in a particular index by
Figs. 3–4 show the main effects of the average and the signal-to-noise each algorithm. It is worth noting that these indices are calculated after
ratio in the tests performed by the NSGA-II algorithm, respectively. the normalization of the values of the objective function.
Figs. 5–6 show the main effects of the average and the signal-to-noise As can be seen from the results of Table 7, the NSGA-II algorithm is
ratio in the tests performed by the MOPSO algorithm respectively. almost absolutely better than the MOPSO algorithm in both indices of the
In the next section, we try to check both NSGA-II and MOPSO algo- spacing and mean ideal distance. However, in the most expanding index,
rithms in order to more accurately calculate the five metrics of the mean both algorithms have almost the same performance, and none of the al-
ideal distance, the diversity, the distancing, the number of Pareto solu- gorithms can be considered better on this criterion. On the other hand,
tions, and the computational times; the comparisons are performed on MOPSO algorithm has been able to generate more Pareto solutions than

7
P. Fattahi et al. Array 10 (2021) 100062

Table 7
The data obtained from the comparison of the two algorithms NSGA_II and MOPSO.
Problem No. Pareto Solutions Spacing Metric Diversity Metric Mean Ideal Distance Metric Time

NSGA-II MOPSO NSGA-II MOPSO NSGA-II MOPSO NSGA-II MOPSO NSGA-II MOPSO

1 41 50 0.02206 0.041021 1.243813 1.270273 0.869709 0.854032 12.12 10.76

2 26 36 0.027715 0.025338 1.300875 0.948774 0.904247 0.929177 14.43 12.98
3 23 10 0.081073 0.191144 1.34565 1.008046 0.791547 0.788962 17.84 14.73
4 26 13 0.042331 0.05596 1.386474 1.145598 0.977889 0.989028 22.74 21.90
5 41 44 0.030691 0.022311 1.414214 0.680442 0.907984 0.978112 27.98 22.72
6 39 28 0.023697 0.052038 1.119701 1.351996 0.842278 0.940493 33.54 28.10
7 32 35 0.018909 0.055081 1.110723 1.208698 0.781524 0.901249 54.89 32.71
8 37 28 0.028888 0.073084 1.110462 1.261739 0.816526 0.904248 78.87 48.91
9 41 49 0.025993 0.033386 1.414214 1.2756 0.936693 1.005937 96.91 89.98
10 29 19 0.063152 0.047909 1.287854 1.22129 0.82383 0.916433 101.71 100.81
11 39 49 0.012084 0.03086 0.78861 1.355823 0.699008 0.916091 143.73 143.11
12 39 41 0.038493 0.064383 1.289289 1.281035 0.841345 0.926292 149.83 148.91
13 38 47 0.032497 0.055551 1.130055 1.269737 0.814869 0.923599 178.98 176.91
14 37 50 0.02388 0.058344 1.1651 1.220976 0.853448 0.834216 189.71 181.91
15 42 42 0.025721 0.042783 1.218934 1.346002 0.823548 0.926535 210.73 201.71
16 43 49 0.020743 0.042376 1.214563 1.313864 0.910263 0.941509 258.73 230.81
18 41 25 0.023581 0.090761 1.11302 1.160005 0.739806 0.836011 289.98 271.12
18 43 37 0.037876 0.066321 1.025607 1.284104 0.81145 0.98707 349.82 343.78
19 40 37 0.031104 0.075574 1.414214 1.120817 0.869746 0.958915 388.87 381.60
20 42 43 0.02749 0.061072 1.241615 1.227841 0.884033 0.944967 482.01 476.18
21 38 43 0.088408 0.034546 1.414214 0.838685 0.776355 0.759314 650.81 620.22
22 36 43 0.024583 0.071699 1.181043 1.196871 0.836235 0.966735 750.81 780.81
23 43 41 0.01756 0.061973 1.116798 1.306701 0.805183 0.915266 890.71 910.74
24 38 47 0.033787 0.055369 1.409315 1.396902 0.928097 0.990399 1098.87 1299.77
25 42 50 0.050667 0.069241 1.414214 0.909582 0.86483 0.865166 1161.63 1309.12
26 40 40 0.023898 0.061537 1.354785 1.032085 0.854951 0.901408 1389.12 1509.10
27 45 37 0.026758 0.084886 1.159992 1.210302 0.785044 0.946534 1598.30 1765.39
28 41 36 0.015954 0.052061 1.076731 1.342298 0.778736 0.942386 1868.87 2087.93
29 39 50 0.024725 0.049191 1.299702 1.235855 0.927856 0.961024 2018.05 2280.19
30 43 43 0.017803 0.045328 1.370415 1.360936 0.941497 0.976807 2405.19 2789.81

Fig. 7. Comparison of two algorithms according to Pareto solutions.

the other algorithm, but it should be noted that the great number of the indices of the diversity, the mean ideal distance in both algorithms
solutions generated by this algorithm are all focused on specific points have not been altered. However, in the number of Pareto solutions ob-
and have not been able to distribute in the solution space. In the terms of tained in the MOPSO algorithm in Fig. 7 lots of changes can be seen in a
computational time, MOPSO is better in the small size of the problems; way that in these numerical examples, the number of Pareto solutions
while NSGA-II perform better in large scale problems. Therefore, it can be varies from 10 to 50, but gradually increases with the dimensions of the
concluded that in general NSGA-II algorithm has been more successful. problem and the MOPSO algorithm is also stable in this index.
Accordingly, the combination of discrete and continuous crossover and
mutation operators are far more compatible than the continuous opera- 5. Conclusions and future suggestions
tors of the MOPSO algorithm in considered problem. The results of these
comparisons can be found in Figs. 7–10. In this research, a mathematical model was presented in the field of
As can be seen, with the increase in the dimensions of the problem, the competitive location-pricing problem for congested facilities with

8
P. Fattahi et al. Array 10 (2021) 100062

Fig. 8. Comparison of two algorithms according to the diversity criterion.

important decision variables in this problem, where the demand function

is considered as a function of the customer’s behavior and is very
important.
On the other hand, new facilities compete for gaining market share
with previously existing facilities. One of the most important factors in
these problems can be the little waiting time, the existence of a different
service strategy, low cost, and the distance. To solve the mathematical
model, various approaches such as multi-objective multi-objective meta-
heuristic algorithms including the Non-dominated sorting genetic algo-
rithm (NSGA-II) and the multi-objective particle swarm optimization
algorithm (MOPSO) have been used. By comparing the results of the
proposed methods for solving the model, the NSGA-II algorithm has been
able to achieve a relatively better Pareto front than the MOPSO algo-
rithm, although in some cases the MOPSO algorithm has been able to
Fig. 9. Compare two algorithms according to the spacing criterion. overcome the other algorithm. Nevertheless, generally the NSGA-II al-

Fig. 10. Comparison of two algorithms according to mean ideal distance criteria.

multiple service. The main objective of this problem was to reduce costs gorithm was able to overcome the other algorithm and produce a better
along with increasing service levels to customers. On the other hand, the Pareto front. Also, as is clear from the Pareto fronts, the Pareto front
existence of confliction in the objectives and business methods makes it produced by the NSGA-II algorithm is more regular and gains more so-
difficult to achieve these objectives. What is clear is that an organization lutions with relatively the same distance from each other and based on
can only hope to be successful in its survival in competitive markets, the four indicators that were introduced in the previous sections, the
hence the price setting parameter is considered as one of the most NSGA- II is better in two indices of distancing and distances than the

9
P. Fattahi et al. Array 10 (2021) 100062

MOPSO algorithm. However, in the most exponential index, both algo- [13] Gorji M, Makui A, Ramezanian R. A robust for a leader- follower competitive facility
location problem in a discrete space. Appl Math Model 2013;37:62–71.
rithms behaved the same. Therefore, it can be concluded that in general
[14] Aboolian R, Berman O, Krass D. Competitive facility location and design problem.
NSGA-II algorithm has been more successful. All the parameters of the Eur J Oper Res 2007;182:40–62.
algorithms were tuned by Taguchi experiment design to better solve the [15] Drezner T, Drezner Z. Multiple facilities location in the plane using the gravity
model. For the future research, one can develop the problem of this model. Geogr Anal 2011;38:391–406.
[16] Pasandideh SHR, Niaki STA. Genetic algorithm in a facility location problem with
research by considering different service levels. random demand within queuing framework”. J Intell Manuf 2012;23:651–9.
[17] Chambari A, Rahmaty SH, Hajipour V, Karimi A. A bi-objective model for location-
Declaration of competing interest allocation problem within queuing framework”. World Academy of Science,
Engineering and Technology 2011;78:138–45.
[18] Fernandez J, Salhi S, G-Toth B. Location equilibria for a continuous competitive
No potential conflict of interest was reported by the authors. facility location problem under delivered pricing. Comput Oper Res 2014;41:
185–95.
[19] Vidyarthi N, Jayaswal S. Efficient solution of a class of location-allocation problem
References with stochastic demand and congestion. Comput Oper Res 2014;48:20–38.
[20] Renando J, Fernandez J, Hervars JD, Arrondo AG, Ortigosa P. Approximating the
[1] Berman O, Larson RC. Optimal 2-facility network districting in the presence of pareto-front of a planar bi-objective competitive facility location and design
queuing. Transport Sci 1985;19(3):261–77. problem. Comput Oper Res 2015;62:337–49.
[2] Wang Q, Batta R, Rump CM. Facility location models for immobile servers with [21] Drezner T, Drezner Z, Kalczynski P. A leader-follower model for discrete
stochastic demand. Nav Res Logist 2004;51:137–52. competitive facility location. Comput Oper Res 2015;64:51–9.
[3] Berman O, Krass D, Wang J. Locating service facilities to reduce lost demand. IIE [22] He Zhou, Cheng TCE, Dong J, Wang S. Evolutionary location and pricing strategies
Trans 2006;38:933–46. for service merchants in competitive O2O markets. Eur J Oper Res 2016;254:
[4] Fernandez J, Hendrix E. Recent insights in Huff-like competitive facility location 595–609.
and design. Eur J Oper Res 2013;227:581–5. [23] Tavakkoli-Moghaddam R, Noshafagh SV, Taleizadeh AA, Hajipour V, Mahmoudi A.
[5] Zarrinpoor N, Seifbarghy M. A competitive location model to obtain a specific Pricing and location decisions in multi-objective facility location problem with M/
market share while ranking facilities by shorter travel time. Int J Adv Manuf M/m/k queuing systems. Eng Optim 2017;49(1):136–60.
Technol 2011;55(5):807–16. [24] Nasiri M.M., Mahmoodian V., Rahbari A., Farahmand S., A modified genetic
[6] Hotelling H. Stability in competition. Econ J 1929:41–57. algorithm for the capacitated competitive facility location problem with the partial
[7] Fernandez P, Pelegrin B, Perez MDG, Peeters PH. A discrete long term location price demand satisfaction, Comput Ind Eng 124(1):435-448.
problem under the assumption of discriminatory pricing: formulations and [25] Khodemani-Yazdi, Tavakkoli-Moghaddam R, Bashiri M, Rahimi Y. Solving a new bi-
parametric analysis. Eur J Oper Res 2008;179:1050–62. objective hierarchical hub location problem with an M∕M∕c queuing framework.
[8] Berman O, Tong D, Krass D. Pricing, location and capacity planning with Eng Appl Artif Intell 2019;78:53–70.
equilibrium driven demand and congestion. University of Toronto; 2010 [Working [26] Zamberano-Ret G, Lopez-Ospina H, Perez J. Retail store location and pricing within
paper]. a competitive environment using constrained multinomial logit. Applied
[9] Hajipour V, Farahani RZ, Fattahi P. “Bi-objective vibration damping optimization mathematical modeling 2019;75:521–34.
for congested location–pricing problem. Comput Oper Res 2016a;70:87–100. [27] Ahmadi Z, Ghezavati V. Developing a new model for a competitive facility location
[10] Hajipour V, Fattahi P, Tavana M, Di Caprio D. Multi-objective multi-layer congested problem considering sustainability using Markov chains. J Clean Prod 2020;273:
facility location-allocation problem optimization with Pareto-based meta-heuristics. 122971.
Appl Math Model 2016b;40:4948–69. [28] Kennedy J, Eberhart R. Particle swarm optimization. In: Proceedings of the 1995
[11] Maghsoudlou H, Rashidi M, Akhavan niaki T, Pourvaziri H. Bi- objective IEEE international conference on neural network, vol. IV; 1995. p. 1942–8. 4.
optimization of a three-echelon multi-server supply-chain problem in congested [29] Deb K, Pratap A, Agarwal S, Meyarivan T. A fast and elitist multiobjectivegenetic
system: modeling and solution. Comput Ind Eng 2016;99:41–62. algorithm: NSGA-II. IEEE Trans Evol Comput 2002;6:182–97.
[12] Fernandez P, Plegrin B, Lancinskas A, Ziliniskas J. New heuristic algorithms for
discrete competitive location problems with binary and partially binary customer
behavior. Comput Oper Res 2016;79:12–8.

10