INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION

https://doi.org/10.1080/23248378.2020.1719372

Optimal railway station locations for high-speed trains based

on partial coverage and passenger cost savings
Sunarin Chanta and Ornurai Sangsawang
Department of Industrial Management, King Mongkut’s University of Technology North Bangkok,
Prachinburi, Thailand

ABSTRACT ARTICLE HISTORY

Advances in high-speed railway technology directly affect industrial Received 31 December 2018
competitiveness and regional economies. In this study, the location Revised 11 October 2019
and allocation problem of high-speed train stations is investigated Accepted 10 January 2020
using a two-stage optimization model. In the first stage, we select KEYWORDS
candidate station locations by considering partial coverage to max- Railway station; high-speed
imize the demand under the limitation of the number of stations. In train; facility location;
the second stage, we determine the optimal number of station optimization; coverage
locations to maximize the passenger transportation cost savings.
The optimal solutions in the first stage are used as candidate
stations in the second stage. Computational experiments are per-
formed to test the proposed two-stage optimization model using
a case study of the northern railway line in Thailand. The results
provide decision-makers with useful information on the design and
construction of high-speed railways.

1. Introduction
There is an increase in the demand for public transportation owing to population growth.
A public transportation system has important economic benefits and social advantages
[1–3]. High-Speed Rail (HSR) is an efficient mass public transportation that reduces
travel time by connecting major cities in a country and linking countries across borders.
HSR facilitates rapid transportation, with typical maximum speeds of approximately
250 km/h [4]. The first HSR system began operation in Japan in 1964, which was known
as the bullet train [4]. In Europe, the Trans-European HSR network was developed,
which connected several countries across borders, including Spain, Italy, Germany,
Austria, Sweden, Belgium, the Netherlands, Russia, and the United Kingdom. In Asia,
many countries, such as China and South Korea, have succeeded in implementing HSR
projects similar to those implemented by Japan. China has the world’s longest
HSR network and operates more than 60% of the world’s total [5]. Presently, HSR is
part of the government’s plan in several countries in Southeast Asia, including Thailand.
In Thailand, it is expected that HSR will be established on four major route lines:
northern, north-eastern, eastern, and southern. In this study, we consider the north
route line, from Bangkok to Chiang Mai, as a case study.

CONTACT Sunarin Chanta sunarin.c@fitm.kmutnb.ac.th Department of Industrial Management, King Mongkut’s

University of Technology North Bangkok, Thailand
© 2020 Informa UK Limited, trading as Taylor & Francis Group
2 S. CHANTA AND O. SANGSAWANG

In the planning stage of establishing an HSR system, one of the most important factors to
be considered is the location of railway stations, given that this parameter directly affects the
volume of the railway passengers for the entire line. To be successful and competitive, high-
speed rail stations should be optimally positioned. Optimization of station locations can
increase the number of passengers, which increases the profitability of railway lines. In
contrast, poor choices in terms of station location may lead to a low number of passengers,
underutilized resources, and failure. Users may expect HSR stations to be as close as possible
to their locations, while the government must consider overall system efficiency. Increasing
the number of stations may result in an increase in the number of passengers; however, every
stop increases cost and decreases the average speed of the train. Moreover, the construction
cost of HSR systems is extremely high, as such, the choice of the design of station locations, the
number of stops, the distance between stops, etc., should be carefully considered.
To account for these factors, we propose an approach to determine the optimal locations
of stations of HSRs based on the consideration of passenger concerns. We focus on two
criteria: demand coverage and total passenger transportation cost savings. Coverage is
usually defined for facility location problems to ensure that customers can access or receive
a service from a facility that is located within a fixed standard distance or coverage distance.
The concept is applied to several facility location problems including emergency medical
service units, fire department stations, vehicle charging locations, wireless sensor networks,
etc. [6–9]. In this case, we apply coverage to the railway station. If the railway station is
located sufficiently close to a population, they may be willing to travel by train; otherwise,
they may choose other modes of transportation. However, individuals may be willing to
travel by HSR at a different threshold. For example, an individual may decide to travel by
HSR if the closest station is located within 30 min of travel from his home, while others may
be willing to travel farther. Therefore, to address these different passenger requirements, we
apply partial coverage, which allows a fraction of coverage instead of the traditional 0–1
coverage. Once the railway stations are addressed, we allocate demand by considering the
total transportation cost savings of all the passengers, given that cost is a major factor that
influences customer choice of the transportation modes [10]. Individuals may compare
their transport mode choices by cost, convenience, travel time, safety, etc.
The problem is formulated and solved using the proposed two-stage optimization
method. The objective of the first stage is to determine the appropriate station locations
that maximize the expected demand that can be covered. The objective of the second stage
is to assign the expected HSR passengers to the railway stations while maximizing the total
passenger transportation cost savings. We calculate passenger savings based on the reduc-
tion in cost based on the assumption of travelling via HSR instead of personal vehicles. The
proposed model is applied to a case study of the north route railway line, from Bangkok to
Chiang Mai, Thailand. We used the number of passengers that travelled by train on this
route line in 2014, for the estimate of HSR demand. The origin-destination trips of all
passengers are also considered in the calculation of the total transportation cost savings, in
addition to the station capacity, the number of stops, and the distance between stations.
The main contributions of this paper are:

(1) We present a two-stage optimization model, which capture two important factors
that represent passenger behaviour: partial coverage of demand and total trans-
portation cost savings. Partial coverage in the first stage helps us correctly quantify
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 3

the number of passengers who are willing to travel to a railway station with
a different level of coverage. The total transportation cost savings of passengers
in the second stage assist in assigning demand to the optimal railway station
location according to the passenger’s choice.
(2) We formulate a mathematical model to determine the optimal location for
a railway station; the partial coverage is introduced in the coverage model for
locating railway stations. The passenger flow with original-destination (O/D)
pairs, the transportation cost of travelling using a personal vehicle and by railway
line is included in the assignment model.
(3) For solving the problem in two stages, we are able to reduce the running time.
The second stage is considered as an assignment location problem, which is
known as NP-hard, and needs to be solved in polynomial time. To reduce the
size of the problem, we select potential candidate stations in the first stage. The
optimal solutions of the first stage are the inputted as candidate stations in
the second stage so that the size of the problem if effectively reduced.

The proposed model can determine the optimal solution in a few seconds. The results are
of potential benefit to railway passengers in terms of transportation accessibility and the
reduction of travelling cost, and are also of benefit to the management of railway
operation in terms of profit and efficient resource utilization.
The rest of the paper is organized as follows: in Section 2 we briefly describe the related
literature review on train station location problems. Section 3 presents the details of
a two-stage optimization model. The covering location model is presented in the first
stage, and then the assigning location model is presented in the second stage. A case study
of HSR in Thailand is presented in Section 4. In Section 5, we conduct computational
experiments to show the performance of the two-stage optimization model and perform
a sensitivity analysis for different choices of discount factors. Finally, the conclusion and
future research are summarized in Section 6.

2. Literature review
The problem of determining railway station location for long-term planning can be divided
into two parts. The first part involves finding alternative solutions, in which several factors
must be analysed. These factors include the origin-destination (O/D) flow, city structure,
type of area, important buildings, transportation link, etc. Once alternative solutions are
determined, the second part, evaluation, is performed. There are many different approaches
for performing evaluation such as technical feasibility, construction cost, population cover-
age, utilization, impact on land use, trade, traffic, environment safety, etc.
There are two categories of research related to the railway station location. In the first
type, the attempt is to determine the optimal station location on a plane with geometry
constraints, in which case the result represents the optimal coordinates (x, y) or (x, y, z).
In the second case, the aim is to find the optimal station location on a network with a set
of existing candidate station locations, and the result represents the optimal selected
locations. For studies that involve the determination of railway station location and
alignment on a plane, there are previous studies that incorporate optimization tools.
Samanta and Jha [11] used a genetic algorithm (GA) that works in combination with
4 S. CHANTA AND O. SANGSAWANG

a geographical information system (GIS) to optimize station locations on a plane along

a rail transit line. A two-stage analytical model that considers ridership and public
perception was developed. The potential station sites were identified in the first stage,
and the optimization using the GA was performed in the second stage by minimizing the
total cost associated with the location of the stations. Lai [12] optimized station locations
and track alignments on plane with geometric requirements for an urban rail transit line,
which considered the trade-off between a system with construction and operation costs,
and a system that provides convenient service to the public. The GA algorithm was
developed to generate feasible alignments and was integrated into a supporting GIS to
find optimal solutions. Wang et al. [13] studied the train unit routing problems to
determine the number of train units required to perform trips, for which maintenance
resource constraints for both time and travel distance were considered.
Several studies have been conducted to determine the location of railway stations on
a network, which include the determination of the number of stops, optimal routes, optimal
schedules, etc. In the literature, primary works on station location focus on population
coverage. Laporte, Mesa, and Ortega [14] addressed the problem of locating a prefixed
number of stations such that the weighted coverage was maximized. To evaluate the
coverage of each station, an estimation of the street network metric and triangulation of
the census tracts was used. Blanco et al. [15] proposed a model for transportation network
expansion, which incorporated the requirements of population coverage, budget con-
straints, intermediate goals, and origin-destination flows. The model was applied to the
Spanish HSR network. Chapleau et al. [16] used the number of individuals that reside
within a certain distance from the line to measure coverage. More precisely, Dufourd,
Gendreau, and Laporte [17] measured the coverage around a station using contour lines,
with decreasing weights. The coverage distance was set as walking distance units to the
station, and the coverage was the weighted population of several concentric diamonds
around the station. Other measures are considered in Laporte, Mesa, and Ortega [18]. To
determine railway station location based on population coverage, we focus on the volume at
a station but neglect the volume on a transportation route.
There are also previous works that address the issue of cost and benefit related to the
transportation route. Lovett et al. [19] determined the most profitable high-speed rail
route based on a set of stations on the route and the system data. They selected eight
alternative lines based on the total profit, then performed a sensitivity analysis with
involved environment impact, project ridership, farebox, returns, land development
revenue, distance, travel time, etc. Repolho, Antunes, and Church [20] developed a mix-
integer optimization model to determine the optimal location of stations along a railway
line with the objective of maximizing travel cost savings, in which travel cost savings are
associated with travelling by high-speed train instead of existing transportation. The
model was applied to a case study of a high-speed rail line that is expected to be built in
Portugal. Repolho, Church, and Antunes [21] extended the optimization model devel-
oped in [20] to make decisions regarding station location and fleet composition, which
considered line planning, train schedules, and fleet assignment issues. The objective is to
maximize net public benefits, in which these benefits are measured as travel cost savings,
minus the investment cost for the construction of stations and the acquisition of trains.
Huang and Shuai [22] developed a loop optimization by applying the Frank–Wolfe
algorithm to assign passenger flow, to obtain a new railway train stop plan. The
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 5

algorithms were investigated using the case study of the Beijing–Shanghai HSR. The
number of transfer passengers was reduced while the number of direct passengers
increased.
Since many factors involve in determining the location of a station for a railway line,
many works attempt to address these related factors, which can increase the complexity of
the problem. Many studies have been performed with the objective of integrating these
factors into a model. Hamacher et al. [23] proposed two models that considered the effect of
new train stops in an existing railway network. The first model involved accessibility
(assuming that residing close to a train stop increases the viability of the railway network)
and the second model considered the travel time induced by introducing a new stop.
Mateus, Ferreira, and Carreira [24] investigated the location of HSR stations that will be
built in the city of Porto, Portugal. They compared a set of location alternatives using multi-
criteria decision analysis (MCDA). These alternatives were evaluated based on a range of
technical, economic, social, and environmental criteria. Hong et al. [25] developed a two-
phase train-set routing algorithm for the management of train timetables. Initially, they
obtained the minimum cost routes without maintenance requirements by solving the
polynomial relaxation. Maintenance-feasible routes were then generated from the cross-
overs of the minimum cost routes. The results were applied to the Korea high-speed
railway. Wang et al. [26] presented a two-layer optimization model, in which the model
focused on finding the optimal stop-schedule set with service frequencies on the top layer,
while minimizing the total operation cost and unserved passenger volume. On the bottom
layer, the model focused on the assignment of the weighted passenger flow, while max-
imizing the passenger volume and minimizing the total travel time for all passengers.
Recently, Fu et al. [27] developed the line planning problem, which consisted of a bi-
level programming model and heuristics, to determine a portion of the routes of the HSR
network in China. The upper-level objective was to minimize the total passenger travel
times and the lower objective was to maximize the passenger demand. By using
a timetabling tool, the line plans can generate better travel times and improve customer
satisfaction. Martin [28] proposed a mathematical approach to determine the appro-
priate arrival path, parking track, and departure path for individual trains under different
constraints. They divided the constraints into two groups; hard constraints involved
track access and platform length, whereas soft constraints involved reallocation, crossing,
and passenger flow. The goal was to find the optimal arrival path, parking track, and
departure path under hard constraints while minimizing the number of broken soft
constraints. Qi et al. [29] formulated a two-objective mixed-integer programming model
to determine the train operation zone and stop planning with the objectives of minimiz-
ing the total running distance of unoccupied seats and the number of stops. Bababeik,
Khademi, and Chen [30] investigated the optimal location and allocation problem of
relief trains to increase the resilience level of the rail network by considering the priority
of the demand on a link. The problem was formulated as bi-objective programming with
the objectives of maximizing coverage of link exposure or the stations and minimizing
total travel time of the entire links. The complex models can capture many related factors;
however, it has a limitation on problem size.
Based on a literature review, we have outlined some of the most important key factors
that have been considered in previous studies related to the determination of railway
station locations and stop plans. They include coverage or accessibility of passengers,
6 S. CHANTA AND O. SANGSAWANG

total cost or travel time/distance on the networks, cost savings/profits, and ridership. The
previous works are classified based on these factors as shown in Table 1. Based on this
review, we considered the factors of coverage and total transportation cost savings as
primary objectives in our model. The total travel time and passenger ridership were also
used to calculate the total cost saving in the model.
In this paper, we focus on the optimization of the locations of railway stations when
candidate station locations are given. Most previous studies ignored the behaviour of
individual passengers. In this study, passenger demand and choice are considered as
a different level of coverage in the first stage, while passenger travel trips including
original-destination (O/D) pairs and their satisfaction with respect to saving are con-
sidered in the second stage. Alternative station locations were selected by maximizing the
expected demand in the first stage, then subsequently determining the optimal station
locations. A railway station location is then assigned to each passenger by maximizing the
total transportation passenger saving in the second stage. The proposed model was
improved based on the previous work in [20], which also considered maximization of
the total transportation savings but ignored passenger coverage. Our model also has
fewer variables compared to [20], requires less processing time, and can be utilized in
large problems.

3. Two-stage optimization model

3.1. The first stage
In this stage, we attempt to identify candidate station locations by developing
a facility location optimization model based on the maximal covering location
problem (MCLP) model. The objective is to maximize the demand that can be
satisfied by open stations. The MCLP was introduced by Church and ReVelle [31],
wherein coverage is defined for a fixed standard distance. The demand point is
covered if there is an open station within the standard distance; otherwise, it is
considered to be uncovered. Subsequently, Daskin [32] presented the maximal
expected covering location problem (MEXCLP) model, in which the probability
that a vehicle is busy is taken into account. As such, the coverage is calculated as
an expected value based on the probability that a vehicle is available. These two
classical models had a significant impact on the field of facility location. Several
extensions were then developed. Drezner, Wesolowsky, and Drezner [33] addressed
the gradual covering problem, wherein they defined a certain distance for ‘fully

Table 1. Summary of previous works on station location and stop planning design for railway.
Goal
Max coverage/ Max passenger Min cost/total travel Max cost savings/
Publications accessibility ridership time profits
[14–17,28] •
[19–21] •
[22,24,29] •
[23] • •
[25–27] • •
[30] • •
Our work • •
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 7

covered,’ and another specified distance for ‘not covered.’ Coverage is linear
between these two distances from the demand point to the facility. Karasakal and
Karasakal [34] developed the concept of partial coverage, wherein coverage is
allowed to change from ‘covered’ to ‘not covered’ within a range instead of
a single critical distance. This intermediate coverage level is referred to as partial
coverage. Other partial coverage measures are considered in Wang et al. [35].
Unlike other facilities, the coverage of demand in the railway station case may be
defined differently. In this paper, we define the coverage based on the maximum distance
that individuals are willing to travel to the nearest railway station to travel by train. It
should be noted that people may travel to the railway station using a personal vehicle or
via other transportation modes. To identify the candidate stations that maximize covered
demand, we developed a maximal location model based on probabilistic coverage for
railway stations. The proposed model is based on the MCLP of Church and ReVelle [31],
using the step function coverage proposed by Dufourd, Gendreau, and Laporte [17],
which was also introduced as partial coverage by Karasakal and Karasakal [34]. It should
be noted that the proposed model was modified to capture the characteristics of railway
transportation, which is not similar to those of mobile facilities. We incorporated
passenger concern to define coverage. At a particular distance away from a railway
station, passengers may or may not want to travel to this destination. This probability
is considered and defined as partial coverage.
Figure 1 shows the difference between 0–1 traditional coverage and the partial
coverage, where c = full covered area, u = uncovered area, c1, and c2 = partial
covered area. Suppose that the station is located in the middle of the circle, and then
Figure 1(a) represents the 0–1 coverage. Demand that is close to the centre or in
this case the demand within the solid circle will be covered (c), and the demand far
away from the circle or outside the circle will be uncovered (u). For the partial
coverage in Figure 1(b,c), further demand (in the areas c1 or c2) will be covered with
some probability. It should be noted that the probability of demand to be covered in
c1 is greater than the probability of demand in c2 because c2 is further from the
station than c1. The bold shade represents a high probability of being covered. The
light shade represents the area farther from the train station with a low probability
of being covered. Given that people in this area may have a lower chance of going
to the train station, they might travel via another mode of transportation that is
more convenient.

a: 0-1 coverage b: partial coverage c: partial coverage

with two levels with three levels

Figure 1. Types of coverage.

8 S. CHANTA AND O. SANGSAWANG

The proposed model is described as the following:

Index and notations
i = index of demand point, where i = 1, 2, . . ., I
j = index of station location, where j = 1, 2, . . ., J
p = maximum number of open railway stations
Parameters
dij = distance between demand point i and station location j
smin = the lower standard distance (the maximum distance that people are willing to
travel to a railway station with full coverage)
smax = the upper standard distance (the maximum distance that people are willing to
travel to a railway station with partial coverage)
aij = probability that people at location i are willing to travel to railway station j
with a given standard distance sl, where l = min, max, smin<smax
=1 if dij≤smin
= P(people at location i will travel by rail at station j) if smin<dij ≤smax
=0 otherwise
hi = the number of passengers at point i
lmin = the minimum distance allowed between two open stations
Decision variables
xj = 1 if railway station at j is open,
0 otherwise
yi = the coverage level (full/partial covered or uncovered) of demand point i by a
station
= 1 (demand point i is full covered)
= fraction (demand point i is partial covered)
= 0 (demand point i is uncovered)
Maximize
X
I
h i yi (1)
i¼1

Subject to

X
J
aij xj  yi ; "i (2)
j¼1

X
J
xj ¼ p (3)
j¼1

xi ¼ 1; for i ¼ 1; J (4)

xj þ xjþ1  1; for j : dj;jþ1 < lmin (5)

 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 9

xj 2 f0; 1g; "j (6)

yi  0; "i (7)
Objective (1) attempts to maximize the number of expected passengers that can be
covered by open stations. It should be noted that each demand point can be covered
(yi>0) with different levels; full coverage (yi = 1) and partial coverage (0< yi<1). The value
of yi depends on the values of aij and xij, where aij can be pre-calculated. Constraint (2)
ensures that the demand point i can be covered by station j if this station is open, where
aij is defined depending on whether a demand point i associated with the distance away
from the station j is full covered, partial covered or uncovered. Constraint (3) is the
limitation of the number of total open stations. Constraint (4) is forced to open stations
at the starting and ending points. Constraint (5) ensures that two inline adjacent stations
will not be opened within the minimum distance allowed. Constraints (6)–(7) define the
domain of the decision variables.

3.2. The second stage

The candidate station locations are selected based on the maximum coverage in the first
stage. Appropriate station locations are then selected by considering the total transporta-
tion cost savings for all passengers in the system, if they travel via the HSR train. In this
case, the total cost savings are associated with the total distance, which is calculated based
on the total amount that passengers will save if they travel by train instead of via
a personal car. We applied the concept of the hub location problem to our case. HSR
station operates as a hub to transfer passengers from the beginning station to the ending
station. O’Kelly [36] first introduced the single allocation p-hub median problem, using
a quadratic integer programming formulation. In this approach, the objective is to
minimize the total transportation cost to serve a given set of flows for a limited number
of hubs (p). Hub location can be utilized to solve transportation network problems
related to air passengers, cargo service, postal delivery, etc. However, the single allocation
p-hub median problem is NP-hard, even when the locations of the hubs are given [37].
Campbell [38] developed the first integer programming formulation for the single
allocation p-hub median problem with n4+n2+n variables. Skorin-
Kapov et al. [39] then proposed a mixed-integer formulation with n4+n2 variables.
Ernst and Krishnamoorthy [40] improved the mixed-integer formulation, such that
a lower number of variables, n3+n2 are required. For additional details on the hub
location problem, the review paper by Alumur and Kara [41] is useful. We developed
an assignment location model for a railway station. The objective of the model is to
maximize transportation cost savings, which is similar to the objective of Repolho,
Antunes, and Church [20]. However, their formulations require n5+n variables, which
lead to a long processing time for large problems. Therefore, we improved the model by
applying the formulation of Ernst and Krishnamoorthy [40], which has fewer variables.
The presented model is described as the following:
Index and notations:
i = index of origin location, where i = 1, 2, . . ., I
j = index of destination location, where j = 1, 2, . . ., J
10 S. CHANTA AND O. SANGSAWANG

m = index of beginning station, where i = 1, 2, . . ., M

n = index of ending station, where j = 1, 2, . . ., N
χ = the discount factor for collection (non-hub to hub)
δ = the discount factor for distribution (hub to non-hub)
α = the discount factor for transshipment between hub links (hub to hub).
Tm = the maximum capacity of origin station m
Sn = the maximum capacity of destination station n
Parameters
cij = transportation cost from node i to node j
wij = the flow of passengers travelling from origin i to destination j P J
oi = the total amount of expected demand originating at node i = PI wij
dj = the total amount of expected demand destined to node j = j¼1 wij
i¼1
Decision variables
xim = 1 if there exists a trip made from node i to the railway stations at m,
0 otherwise
xmm = 1 if station at m is open,
0 otherwise
yimn = the proportion of the expected demand travel from node i via railway
station at
m and n.
Maximize
X
I X
J XI X
M X
I X
M X
N
cij wij  ½ cim xim ðχoi þ δdi Þ þ αcmn ymn
i
 (8)
i¼1 j¼1 i¼1 m¼1 i¼1 m¼1 n¼1

Subject to
X
M
xim ¼ 1; "i (9)
m

xim  xmm ; "i; m (10)

X
N X
N X
N
i
ymn  i
yim ¼ oi xim  wij xjm ; "i; m (11)
n¼1 n¼1 j¼1

X
I
oi xim  Tm xmm ; "m (12)
i¼1

X
J
oi xnj  Sn xnn ; "n (13)
j¼1

X
M
xmm ¼ p; (14)
m¼1
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 11

i
ymn  0; "i; m; n (15)

xim 2 f0; 1g; "i; m (16)

Objective (8) maximizes the total cost savings of all passengers travelling via train instead
PI PJ
of their personal cars. The term cij wij represents the total cost of all passengers
i¼1 j¼1
directly travelling by their personal cars from origin i to destination j. The term
I P
P M I P
P M P
N
½ cim xim ðχoi þ δdi Þ þ αcmn ymn
i
 represents the total cost of all passengers
i¼1 m¼1 i¼1 m¼1 n¼1
travelling via train, where χ denoted as the weight for inbound trip from origin to a train
station, δ denoted as the weight for outbound trip from a train station to the destination,
and α denoted as the weight for travel between train stations. Constraint (9) ensures that
all passengers will travel once at most. Constraint (10) ensures that passenger flow is
assigned to a station only if the station is open (xmm= 1). Constraint (11) is the flow
balance for passengers originated from i at station m where the inbound and outbound
flows at the station are determined by the allocation variable xim. If the variable xim= 1, so
the constraint is accounted for the demand ordinated from i (Oi), it should be noted that
there are trips from i to j via stations m and n if these stations are selected to be open.
Constraints (12)–(13) ensure that the total number of expected passengers assigned to
stations m and n do not exceed the capacity of the origin and destination stations,
respectively. Constraint (14) limits the number of total open stations. Constraints
(15)–(16) define a domain of decision variables.
The details of the proposed two-stage optimization approach as presented in Figure 2.
The input data that include the candidate stations, distance matrix, demand at each
candidate station, and the minimum distance between the stations are used to select the
maximum coverage station locations in the first stage. The optimal solutions obtained for
the first stage are then used as candidate stations in the second stage with another set of
input data that include the transportation cost matrix, flow of passengers from origin to
destination, discount factors, and the maximum capacity at each station. In the second
stage, the optimal location station is determined and demand is assigned to optimal
solutions with the maximum transportation cost savings.

4. Case study
The case study is based on the HSR in Thailand, which is expected to be built by the
government. Based on a recent plan, four major route lines will be established; north,
northeast, east, and south. In this study, we selected the north route railway line for our case
study, which is routed from Bangkok to Chiang Mai, Thailand, with a total distance of
745 km and a maximum speed of 250 km/h [42]. Given that the north route line of the HSR
is planned to be built along the existing railway line, we used the locations of the current
north railway line as candidate HSR stations, with a total of 120 stations. The number of
passengers who travelled by train on the north route line in 2014 was used to estimate the
expected HSR demand and the flow of passengers. It should be noted that we assumed that
these two parameters demand at each station and flow of passengers are known and fixed.
12 S. CHANTA AND O. SANGSAWANG

Data Input: candidate stations, distance matrix,

demand at each candidate station, minimum
distance between stations

First Stage Solve the coverage location model

Output: Optimal stations to open with

maximum coverage demand

Data Input: optimal stations from first stage,

transportation cost matrix, flow of passengers
from origin to destination, discount factors,
maximum capacity at each stations

Second Stage Solve the assignment location model

Output: Optimal assignment of passengers to

station with maximum transportation cost
savings

Figure 2. Two-stage optimization approach flow chart.

The total travelled cost savings are associated with the distance travelled, which is calculated
based on the difference in cost between travelling via a personal car from origin to
destination, and travelling via an HSR train. The distance from an origin to a destination
is calculated based on the shortest geographical distance obtained from Google Maps.

5. Computational experiments
The two-stage optimization was applied to high-speed rail in Thailand to determine the
optimal location of the stations. The mathematical models in Section 3 were implemented in
OPL (Optimization Programming Language) 12.7 and solved using an Intel Core i5-2410M
CPU 2.3 GHz with 6 GB of RAM. In the first stage, we considered appropriate stations based
on partial coverage. Therefore, at the end of this stage, we obtained the baseline stations of
the north route line. Passengers were then assigned to railway stations at the second stage by
considering the total transportation cost savings. The candidate stations are on the existing
railway of the north route line, from Bangkok to Chiang Mai, Thailand, with a total of 104
stations (j = 1,2, . . .,104). Stations with very few passengers were neglected. It should be
noted that station No.1 refers to Bangkok and station No.104 refers to Chiang Mai. The
demand point is assumed to be at stations (j = 1,2, . . .,104). The number of open stations (p)
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 13

was set from 2 to 15 and the number of passengers at each demand point (hi) is estimated
based on the number of passengers that travel by train on the north route line in 2014. For
the partial coverage parameters, we conducted a survey on the characteristics of expected
high-speed train demand on the north route line based on 600 passengers who use public
transportation. Based on our survey, we set smin = 50, smax = 100, where P(individuals who
are willing to travel by rail if the station is located within 50 km) = 1, P(individuals who are
willing to travel by rail if the station location exceeds 100 km) = 0, and P(individuals who are
willing to travel by rail if the station is located between 51 km and 100 km) = 0.25.

5.1. Results of the first stage

Three experiments were conducted to determine the difference between using full
coverage (0–1 coverage) and partial coverage. The first two cases are the full coverage
with standard distance (s) at 50 km and 100 km, respectively. The third case is the
partial coverage with the minimum standard distance (smin) and maximum standard
distance (smax) at 50 km and 100 km, respectively. It should be noted that the origin and
destination stations are fixed to be always open. The results are shown in Tables 2–3,
which summarize the objective or the coverage of expected passengers associated with
the open stations that are suggested by the first stage of the optimization model. To be
able to compare the differences of these three cases, we present the coverage in Table 2
and the open stations in Table 3. For a smaller number of stations, the number of open
stations significantly affects the coverage. Different standard distances lead to different
sets of open stations; therefore, partial coverage is useful in reflecting real demand.
Using the standard of 50 km, the number of passengers was underestimated, whereas
the standard of 100 km resulted in an overestimation. In Figure 3, we represent the
location of open stations as a railway line. Figure 4 shows the open stations of the
partial coverage case on a geographical map. Figure 5 shows a comparison between the
coverage for each station of the three cases represented in the graph. The running times
are in the range of 3 to 5 s as detailed in Figure 6. The average running time of the

Table 2. The results of the first stage with maximum coverage.

Full coverage Partial coverage
Case1: smin = smax = 50 Case2: smin = smax = 100 Case3: smin = 50, smax = 100
Objective Objective Objective
p (106 persons) Coverage (%) (106 persons) Coverage (%) (106 persons) Coverage (%)
2 14.015 60 14.310 61 14.088 60
3 17.115 73 19.587 84 17.608 75
4 19.992 85 21.715 93 20.160 86
5 21.549 92 22.960 98 21.875 93
6 22.382 96 23.396 100 22.636 97
7 22.934 98 23.396 100 23.161 99
8 23.188 100 23.392 100 23.288 60
9 23.396 100 23.396 100 23.396 100
10 23.396 100 23.396 100 23.396 100
11 23.396 100 23.396 100 23.396 100
12 23.396 100 23.396 100 23.396 100
13 23.396 100 23.396 100 23.396 100
14 23.396 100 23.396 100 23.396 100
15 23.396 100 23.396 100 23.396 100
14 S. CHANTA AND O. SANGSAWANG

Table 3. The results of the first stage with optimal stations.

Full coverage Partial coverage
p Case1: smin = smax = 50 Case2: smin = smax = 100 Case3: smin = 50, smax = 100
2 {1,104} {1,104} {1,104}
3 {1 18 104} {1 24 104} {1 18 104}
4 {1 16 28 104} {1 24 64 104} {1 16 28 104}
5 {1 16 28 56 104} {1 24 46 66 104} {1 16 28 56 104}
6 {1 16 28 39 56 104} {1 7 28 43 62 104} {1 16 28 40 56 104}
7 {1 16 28 39 52 65 104} {1 15 39 55 74 88 104} {1 16 28 40 52 65 104}
8 {1 7 16 28 39 52 65 104} {1 23 24 43 60 80 92 104} {1 12 16 28 40 52 65 104}
9 {1 7 16 28 39 48 58 66 104} {1 2 3 24 28 48 67 68 104} {1 12 16 28 39 48 57 67 104}
10 {1 12 22 28 39 47 56 63 70 104} {1 2 3 24 28 48 67 68 69 104} {1 10 20 28 39 48 53 63 71 104}
11 {1 12 22 28 39 48 55 59 68 70 104} {1 2 7 28 40 48 65 67 68 69 104} {1 12 21 28 40 48 55 65 74 102 104}
12 {1 12 22 28 39 47 56 66 68 69 70 104} {1 2 7 28 40 48 64 65 67 68 69 104} {1 10 20 28 39 41 50 53 65 66 71 104}
13 {1 10 20 31 42 50 60 69 77 84 89 97 {1 10 15 28 32 33 39 45 46 48 58 67 {1 10 15 28 32 33 39 45 46 48 58 67
104} 104} 104}
14 {1 10 20 28 39 48 58 66 67 68 69 70 {1 16 40 41 59 64 65 66 67 68 69 74 {1 2 4 15 28 30 32 44 48 55 65 76 82
77 104} 88 104} 104}
15 {1 10 20 21 33 42 50 60 70 75 76 84 {1 2 3 18 41 48 65 66 67 68 69 73 {1 12 15 28 30 34 45 50 60 68 75 77
89 97 104} 74 88 104} 84 94 104}

1 3 12 15 20 28 33 41 45 50 65 67 69 73 75 77 88 94 104
Station locations
Smin=Smax=50 Smim=Smax=100 Smin=50,Smax=100
Figure 3. Station locations selected to open in the railway line by maximizing coverage.

partial coverage (4.25 s) is slightly higher than the full coverage cases (case 1 = 3.54 s,
case 2 = 3.60 s). The set of 15 open stations is used as a baseline in the second stage to
assign passengers to stations.

5.2. Results of the second stage

In the second stage, the optimal locations of the stations were determined based on the
total travel cost savings of all passengers if they travelled via HSR. The appropriate
locations of the stations on the first stage were used as candidate stations in this stage.
The candidate stations were 1, 12, 15, 28, 30, 34, 45, 50, 60, 68, 75, 77, 84, 94, and 104, for
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 15

Figure 4. Map of open stations with maximum partial coverage.

a total of 15 stations (j= 1,2, . . .,15). From the first stage opening, all of the 15 stations
cover 100% of the expected demand of the north route railway line, including 104
demand points (i= 1,2, . . .,104). We varied the maximum number of open stations on
the route from 2 to 15 stations. The discount factors (χ,α,δ) were set differently to
determine the effect on the solutions, which consisted of 3 cases (1,1,1), (1,0.2,1), and
(2,0.2,2). The first case is the base case in which there is no discount for travel via a hub or
rail station. The discount factor for the second case is represented by the time saved when
travelling via HSR. The last case considers the inconvenience of passengers given that
they have to travel from their origin and destination to the rail stations. The results are
shown in Tables 4–5, which report the total travel transportation cost savings associated
with the open stations suggested by the second stage of the optimization model, in
16 S. CHANTA AND O. SANGSAWANG

24

22
Coverage (106 persons)
20

18

16

14

12
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Number of stations (p)
s=50 s=100 smin=50,smax=100

Figure 5. Comparison of full and partial coverage at a number of different stations.

5
Run time (sec.)

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Number of stations (p)
s=50 s=100 smin=50,smax=100

Figure 6. Run times of the first stage at a number of different stations.

addition to the average distance and the maximum distance to the closest rail station.
Based on the experiments performed for these three cases, the set of open stations
selected was different, and opening more stations may not always result in more travel
cost savings. The maximum total transportation cost savings are highlighted in bold text
and are also displayed graphically in Figure 7. It should be noted that in this case, we only
accounted for the personal cost of travelling, which is associated with distance. The
running times differ between 29 and 1767 s. The average run times for case 1, 2, and 3 are
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 17

Table 4. The results of the second stage with maximum total transportation cost savings.
Case1 Case2 Case3
Discount factor (1,1,1) Discount factor (1,0.2,1) Discount factor (2,0.2,2)
Max dis- Max dis- Max dis-
Avg. dis- tance Avg. dis- tance Avg. dis- tance
tance to closest tance to closest tance to closest
Objective to closest station Objective to closest station Objective to closest station
p (units) station (km.) (km.) (units) station (km.) (km.) (units) station (km.) (km.)
2 −55.87 102.9 826.3 753.04 174.0 423.6 −1048.64 174.0 423.6
3 403.05 96.1 823.6 1381.16 136.8 423.6 245.37 136.8 423.6
4 574.70 44.5 423.6 1819.80 44.5 423.6 1090.15 44.5 423.6
5 663.64 36.4 423.6 2010.93 36.4 423.6 1476.37 36.4 423.6
6 681.35 35.2 423.6 2111.55 22.3 423.6 1697.55 22.3 423.6
7 696.99 24.2 423.6 2131.41 22.3 366.7 1743.42 22.3 366.7
8 705.53 17.1 423.6 2147.18 19.9 366.7 1780.59 19.9 366.7
9 713.86 17.1 272.4 2156.49 18.7 366.7 1798.55 15.6 366.7
10 721.81 16.6 272.4 2165.32 14.4 366.7 1816.04 14.4 366.7
11 725.11 14.0 272.4 2171.76 14.0 366.7 1825.09 14.0 366.7
12 725.72 14.0 272.4 2171.73 14.0 360.5 1825.06 14.0 272.4
13 725.51 14.0 118.8 2171.61 14.0 272.4 1834.97 14.0 118.8
14 709.16 14.0 118.8 2171.57 14.0 118.8 1836.75 14.0 74.1
15 464.40 14.0 74.1 2138.45 14.0 74.1 1836.89 14.0 74.1
Note: Bold values indicate solutions with the maximum total transportation cost savings.

Table 5. The results of the second stage with optimal stations.

Case1 Case2 Case3
p Discount factor (1,1,1) Discount factor (1,0.2,1) Discount factor (2,0.2,2)
2 {1 28} {1 68} {1 68}
3 {1 15 28} {1 15 68} {1 15 68}
4 {1 15 28 68} {1 15 28 68} {1 15 28 68}
5 {1 15 28 34 68} {1 15 28 34 68} {1 15 28 34 68}
6 {1 12 15 28 34 68} {1 15 28 34 50 68} {1 15 28 34 50 68}
7 {1 12 15 28 34 60 68} {1 15 28 34 50 68 75} {1 15 28 34 50 68 75}
8 {1 12 15 28 34 45 60 68} {1 15 28 34 45 50 68 75} {1 15 28 34 45 50 68 75}
9 {1 12 15 28 34 45 60 68 84} {1 12 15 28 34 45 50 68 75} {1 15 28 34 45 50 60 68 75}
10 {1 12 15 28 30 34 45 60 68 84} {1 12 15 28 34 45 50 60 68 75} {1 12 15 28 34 45 50 60 68 75}
11 {1 12 15 28 30 34 45 50 60 68 84} {1 12 15 28 30 34 45 50 60 68 75} {1 12 15 28 30 34 45 50 60 68 75}
12 {1 12 15 28 30 34 45 50 60 68 77 {1 12 15 28 30 34 45 50 60 68 75 77} {1 12 15 28 30 34 45 50 60 68 75 84}
84}
13 {1 12 15 28 30 34 45 50 60 68 77 84 {1 12 15 28 30 34 45 50 60 68 75 77 {1 12 15 28 30 34 45 50 60 68 75 94
94} 84} 104}
14 {1 12 15 28 30 34 45 50 60 68 75 77 {1 12 15 28 30 34 45 50 60 68 75 77 {1 12 15 28 30 34 45 50 60 68 75 84 94
84 94} 84 94} 104}
15 {1 12 15 28 30 34 45 50 60 68 75 77 {1 12 15 28 30 34 45 50 60 68 75 77 {1 12 15 28 30 34 45 50 60 68 75 77
84 94 104} 84 94 104} 84 94 104}
Note: Bold values indicate solutions with the maximum total transportation cost savings.

35.83, 623.56, and 178.80 s, respectively. The results will be useful for planning the
optimal number of stops on a route, for a given system.

5.3. Sensitivity analysis

In this section, we study the effect of discount factors (χ,α,δ). For the first case, the hub
discount factor (α) is varied as 0.2, 0.4, 0.6, 1, while collection (χ) and the distribution (δ)
factors are fixed at 1 to determine the effect on the total transportation cost saving for
different number of open stations. The result is presented graphically in Figure 8. For a low
18 S. CHANTA AND O. SANGSAWANG

1 12 15 28 30 34 45 50 60 68 75 77 84 94 104

Station locations
(1,1,1) (1,0.2,1) (2,0.2,2)
Figure 7. Station locations selected to open in the railway line by maximizing transportation cost
savings.

2500
Total transportation cost savings (units)

2000

1500

1000

500

0
2 3 4 5 6 7 8 9 10 11 12 13 14 15
-500
Number of stations (p)
(1,1,1) (1,0.6,1) (1,0.4,1) (1,0.2,1)

Figure 8. Total transportation savings of passengers at different hub discount factors.

number of open stations (p < 5), the hub discount factor had less effect on the total
transportation cost savings when compared to the case of a large number of open stations
(p ≥ 5). However, for the 15 open stations, the total transportation cost savings for each case is
slightly reduced.
For the second case, the hub discount factor was fixed at 0.2, while the others were varied.
The results were compared to the base case when there was no discount on travel. The results
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 19

2500

2000
Total transportation cost savings (units)
1500

1000

500

0
2 3 4 5 6 7 8 9 10 11 12 13 14 15
-500

-1000

-1500
Number of stations (p)
(1,1,1) (2,0.2,2) (1,0.2,1) (1,0.2,0.6)
Figure 9. Total transportation savings of passengers at different collection and distribution discount
factors.

2000
1800
1600
1400
Run time (sec.)

1200
1000
800
600
400
200
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Number of stations (p)
(1,1,1) (1,0.6,1) (1,0.4,1)
(1,0.2,1) (2,0.2,2) (1,0.2,0.6)
Figure 10. Run times of the second stage at a different number of stations discount factors are varied.

are shown in Figure 9. For a low number of open stations (p < 5), the collection and
distribution discount factors had a different effect on the total transportation cost savings,
20 S. CHANTA AND O. SANGSAWANG

while for a large number of open stations (p ≥ 5), the effect was more stable. The total
transportation cost savings for hub discount factors were higher than the base case, regardless
of the change in the collection and distribution factors.
The running times of all the experiments for the second stage are (1,1,1), (1,0.6,1), (1,0.4,1),
(1,0.2,1), (2,0.2,2), and (1,0.2,0.6) are reported graphically in Figure 10. The running times
vary between 29 and 1767 s as reported in Figure 10. It is evident for the graph that in the case
of no discount, the difference for the hub discount and other discount is small, whereas the
run time is high, with an average of 623, 200, 417 s for cases (1,1,1), (1,0.6,1), (1,0.2,0.6),
respectively. In contrast, for the cases where the difference of the hub discount and other
discounts was greater, the runtime was fast, with an average of 74, 35, 178 s for cases (1,0.4,1),
(1,0.2,1), (2,0.2,2), respectively. It should be noted that without the station selection process
on the first station, based on the primary experiments with 104 demand points and 104
alternative stations, it is not possible to obtain a feasible solution after 1 h. Therefore, dividing
the problem into two stages allows us to solve the problem efficiently.

6. Conclusion and discussion

In this study, a two-stage optimization model was proposed. The objective was to determine
the optimal station locations for HSR based on passenger satisfaction, coverage, and cost
savings. We selected appropriate stations based on partial coverage (in the first stage) and
total transportation cost savings (in the second stage). Alternative stations were selected as
a baseline in the first stage. In the second stage, the expected demand points were assigned
to the selected stations on the baseline. A case study of HSR in Thailand was presented
using actual number of passengers to estimate the expected HSR demand.
The optimal solutions for each stage can be found within a reasonable time of a few
seconds on average in the first stage, and a few minutes on average in the second stage. Given
that the model used in the second stage is complex with a large number of variables, a long
time is required to obtain a solution. By dividing the problem into two stages, we reduced the
number of variables in the second stage, which resulted in a shorter processing time and the
solution was more easily obtained. Moreover, partial coverage facilitates the accurate repre-
sentation of actual demand without overestimation or underestimation. Considering the total
transportation cost savings of passengers also allows us to focus more on users. The optimal
location of HSR suggested by the proposed model can be used as a guideline to provide more
information for decision-makers, which leads to effective policy planning.
For future research, we are interested in analysing the problem when the size of
a station is varied and the location of stations is not fixed. In this study, we assumed
that there are fixed existing station locations. The maximum capacity at each station was
set to a high number of passengers. It would be interesting to observe the effect of relaxing
these two assumptions. We could then potentially determine the optimal size of a station
on a line and allow any location with a high demand to be the optimal station location.
Moreover, the ability to predict a trend for expected demand will be more advantageous
for planning the location of HSR stations. We used the annual number of passengers for
the current railway to estimate the HSR demand in this study. However, there are other
ways to estimate future HSR demand. One possibility is to use a population with a growth
rate or to consider related factors such as competitiveness, commercial zones, a link to
other modes of transportation, etc., to predict the expected HSR demand.
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 21

Disclosure statement
No potential conflict of interest was reported by the authors.

Funding
This work was supported by King Mongkut’s University of Technology North Bangkok
[KMUTNB-GOV-58-49].

References
[1] Givoni M. Development and impact of the modern high-speed train: a review. Transp Rev.
2006;26(5):593–611.
[2] Kim H, Sultana S. The impacts of high-speed rail extensions on accessibility and spatial
equity changes in South Korea from 2004–2018. J Transp Geogr. 2015;45:48–61.
[3] Button K. High-speed railways: do they produce economic growth? Mercatus Research,
Mercatus Center at George Mason University; Arlington, VA; 2017 Mar.
[4] UIC. What is high speed rail? 2018a [cited 2018 May 13]. Available from: https://uic.org/
highspeed
[5] World Economic Forum. These are the world’s fastest trains; 2019 [cited 2019 Mar 29].
Available from: https://www.weforum.org
[6] Enayati S, Mayorga ME, Rajagopalan HK, et al. Real-time ambulance redeployment
approach to improve service coverage with fair and restricted workload for EMS
providers. Omega (Westport). 2018;79:67–80.
[7] Yao J, Zhang X, Murray AT. Location optimization of urban fire stations: access and service
coverage. Comput Environ Urban Syst. 2019;73:184–190.
[8] Dong G, Ma J, Wei R, et al. Electric vehicle charging point placement optimisation by
exploiting spatial statistics and maximal coverage location models. Transp Res D Transp
Environ. 2019;67:77–88.
[9] Wang Z, Chen H, Cao Q, et al. Achieving location error tolerant barrier coverage for
wireless sensor networks. Comput Netw. 2017;112:314–328.
[10] Chang C, Thai, VV. Shippers’ choice behaviour in choosing transport mode: the case of
South East Asia (SEA) region. The Asian Journal of Shipping and Logistics. 2017;33(4):199–
210.
[11] Samanta S, Jha MK. Identifying feasible locations for rail transit stations: two-stage analy-
tical model. Transp Res Rec. 2008;2063:81–88.
[12] Lai X. Optimization of station locations and track alignments for rail transit lines [Master
Thesis]. College Park: Department of Civil & Environmental Engineering, University of
Maryland; 2012.
[13] Wang Y, Gao Y, Yu X, et al. Optimization models for high- speed train unit routing
problems. Comput Ind Eng. 2018 [cited 2018 Mar 6]. Available from: https://doi.org/10.
1016/j.cie.2018.02.039
[14] Laporte G, Mesa JA, Ortega FA. Locating stations on rapid transit lines. Comput Oper Res.
2002;29(6):741–759.
[15] Blanco V, Pureto J, Ramos AB. Expanding the Spanish high-speed railway network. Omega
(Westport). 2011;39(2):138–150.
[16] Chapleau R, Lavigueur P, Baass K. A posteriori impact analysis of a subway extension in
Montreal. Montreal: Publication 503, Centre for Research on Transportation; 1986.
[17] Dufourd H, Gendreau M, Laporte G. Locating a transit line using tabu search. Location Sci.
1996;4:1–19.
[18] Laporte G, Mesa JA, Ortega FA. Optimization methods for the planning of rapid transit
systems. Eur J Oper Res. 2000;122:1–10.
22 S. CHANTA AND O. SANGSAWANG

[19] Lovett A, Munden G, Saat MR, et al. High-speed rail network design and station location.
Transp Res Rec. 2013;2374:1–8.
[20] Repolho HM, Antunes AP, Church RL. Optimal location of railway stations: the Lisbon-
Porto high-speed rail line. Transportation Science. 2013;47(3):330–343.
[21] Repolho HM, Church RL, Antunes AP. Optimizing station location and fleet composition
for a high-speed rail line. Transp Res Part E. 2016;93:437–452.
[22] Huang W, Shuai B. Approach and application on high-speed train stop plan for better
passenger transfer efficiency: the China case. Int J Rail Trans. 2019;7(1):55–78.
[23] Hamacher H, Liebers A, Schobel A, et al. Locating new stops in a railway network. Electron
Notes Theor Comput Sci. 2001;50(1):13–23.
[24] Mateus R, Ferreira JA, Carreira J. Multicriteria decision analysis (MCDA): central Porto
high-speed railway station. Eur J Oper Res. 2008;187:1–18.
[25] Hong S, Kim KM, Lee K, et al. A pragmatic algorithm for the train-set routing: the case of
Korea high-speed railway. Omega (Westport). 2009;37:637–645.
[26] Wang L, Jia L, Qin Y, et al. A two-layer optimization model for high-speed railway line
planning. J Zhejiang Univ- Sci A. 2011;12(12): 902–912. The Lisbon-Porto high-speed rail
line. Transp Sci. 2013;47(3): 330–343.
[27] Fu H, Nie L, Meng L, et al. A hierarchical line planning approach for a large-scale high speed
rail network: the China case. Transp ResA Policy Pract. 2015;75:61–83.
[28] Martin B. Optimization of train positioning in railway stations [Master Thesis in
Optimization and Systems Theory]. Sweden: Degree Program in Engineering Physics,
Royal Institute of Technology; 2016.
[29] Qi J, Yang L, Di Z, et al. Integrated optimization for train operation zone and stop plan with
passenger distributions. Transp Res E Logist Transp Rev. 2018;109:151–173.
[30] Bababeik M, Khademi N, Chen A. Increasing the resilience level of a vulnerable rail
network: the strategy of location and allocation of emergency relief trains. Transp Res
E Logist Transp Rev. 2018;119:110–128.
[31] Church RL, ReVelle C. The maximal covering location problem. Pap Reg Sci Assoc.
1974;32:101–118.
[32] Daskin MS. A maximal expected covering location model: formulation, properties, and
heuristic solution. Transp Sci. 1983;17:48–70.
[33] Drezner Z, Wesolowsky GO, Drezner T. The gradual covering problem. Nav Res Logistics.
2004;51:841–855.
[34] Karasakal O, Karasakal EK. A maximal covering location model in the presence of partial
coverage. Comput Oper Res. 2004;31:1515–1526.
[35] Wang J, Liu H, An S, et al. A new partial coverage locating model for cooperative fire
services. Inf Sci. 2016;373:527–538.
[36] O’Kelly ME. A quadratic integer program for the location of interacting hub facilities. Eur
J Oper Res. 1987;32:393–404.
[37] Love RF, Morris JG, Wesolowski GO. Facilities location: models and methods. New York:
Elsevier Science Publishing Co.; 1988.
[38] Campbell JF. Integer programming formulations of discrete hub location problems. Eur
J Oper Res. 1994;72:387–405.
[39] Skorin-Kapov D, Skorin-Kapov J, O’Kelly M. Tight linear programming relaxations of
uncapacitated p-hub median problems. Eur J Oper Res. 1996;94:582–593.
[40] Ernst AT, Krishnamoorthy M. Efficient algorithms for the uncapacitated single allocation
p-hub median problem. Location Sci. 1996;4(3):139–154.
[41] Alumur S, Kara BY. Network hub location problems: the state of the art. Eur J Oper Res.
2008;190(1):1–21.
[42] UIC. High speed line in the world; 2018b [cited 2018 May 13]. Available from: https://uic.
org/IMG/pdf/20180420_high_speed_lines_in_the_world.pdf