See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/344387545

Urban transit network optimization under variable demand with single and
multi-objective approaches using metaheuristics: The case of Daejeon, Korea

Article  in  International Journal of Sustainable Transportation · September 2020

DOI: 10.1080/15568318.2020.1821414

CITATIONS READS

0 21

5 authors, including:

Mohammad Hadi Almasi Yoonseok Oh

University of Malaya Korea University
7 PUBLICATIONS   47 CITATIONS    6 PUBLICATIONS   5 CITATIONS   

SEE PROFILE SEE PROFILE

Ali Sadollah Young-Ji Byon

University of Science and Culture Khalifa University
98 PUBLICATIONS   2,087 CITATIONS    42 PUBLICATIONS   542 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Time Domain Nonlinear Structural Identification View project

Neural Network Algorithm View project

All content following this page was uploaded by Ali Sadollah on 26 September 2020.

The user has requested enhancement of the downloaded file.

 International Journal of Sustainable Transportation

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/ujst20

Urban transit network optimization under

variable demand with single and multi-objective
approaches using metaheuristics: The case of
Daejeon, Korea

Mohammad Hadi Almasi , Yoonseok Oh , Ali Sadollah , Young-Ji Byon &

Seungmo Kang

To cite this article: Mohammad Hadi Almasi , Yoonseok Oh , Ali Sadollah , Young-Ji Byon &
Seungmo Kang (2020): Urban transit network optimization under variable demand with single and
multi-objective approaches using metaheuristics: The case of Daejeon, Korea, International Journal
of Sustainable Transportation, DOI: 10.1080/15568318.2020.1821414

To link to this article: https://doi.org/10.1080/15568318.2020.1821414

Published online: 23 Sep 2020.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=ujst20
INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION
https://doi.org/10.1080/15568318.2020.1821414

Urban transit network optimization under variable demand with single and
multi-objective approaches using metaheuristics: The case of Daejeon, Korea
Mohammad Hadi Almasia, Yoonseok Ohb, Ali Sadollahc , Young-Ji Byond , and Seungmo Kangb
a
Department of Engineering, Gorgan Branch, Islamic Azad University, Gorgan, Iran; bSchool of Civil, Environmental and Architectural
Engineering, Korea University, Seoul, Republic of Korea; cDepartment of Mechanical Engineering, University of Science and Culture, Tehran,
Iran; dDepartment of Civil Infrastructure and Environmental Engineering, Khalifa University of Science and Technology, Abu Dhabi, UAE

ABSTRACT ARTICLE HISTORY

Internationally, there are heightened demands for efficient public transportation systems due to Received 29 June 2019
high population growth rates in urban areas and their associated increased trip demands within Revised 4 September 2020
and across city boundaries. An ideal and sustainable public transportation system should satisfy its Accepted 6 September 2020
passengers while minimizing operation costs that are often associated with energy consumptions.
KEYWORDS
One such cost-effective approach is establishing an integrated public transit system. A transit sys- Integrated transit;
tem generally includes a set of bus routes and rail lines connected by transfer stations. The main metaheuristics; multi-
objective of this research is to propose a sustainable and integrated transit establishment model objective optimization;
to design an optimal bus transit system in combination with an existing railway system dealing network design;
with both fixed and variable demands while satisfying multiple objectives. Moreover, this paper sustainable transit
finds an optimum set of transit routes that corresponds to chosen tradeoffs between user cost,
operator cost and, notably, unsatisfied demand cost. Optimal transit networks have been achieved
using single and multi-objective approaches via metaheuristic optimization algorithms including
the genetic algorithm and the non-dominated sorting genetic algorithm II (NSGA-II). The study
area is chosen as Daejeon City, South Korea for its strategic location. Compared with existing tran-
sit networks, the proposed approach shows significant improvements in terms of costs. In add-
ition, the proposed approach can provide an efficient methodology for finding alternative
alignments of existing transit systems for decision makers.

1. Introduction tested in the network of Daejeon City, South Korea. Such a

multi-objective approach on the transit network design
Sustainable public transportation systems of modern cities
problem (TNDP) with variable demand, considering a con-
strongly rely on efficient operations of urban mass rapid
transit systems. Because of environmental issues, congestion nected railway system, especially as an application to a large
mitigation, and vulnerable road users, surface mass public real-size metropolitan area, has been rarely shown in the
transport systems should be improved and promoted existing literature.
(Oguchi et al., 2017). It is essential to design an efficient bus The structure of this paper is arranged as follows: Section
network to accommodate the demand of the mass rapid 2 is about the literature review on the TNDP and solution
transit systems. Such public transit routes evolve over time approaches. Section 3 presents the description and definition
due to changes in demand caused by variations including a of the problem and model formulation. In Section 4, the
household income from land-use planning. Additionally, solution methodology used in this research is discussed. The
there are demand/supply interactions due to the implemen- computational optimization results obtained by applied opti-
tation of new transit technologies or changes in the mode of mizers (i.e., single and multi-objective approaches) and the
service (Thilakaratne & Wirasinghe, 2016). An ideal bus net- corresponding discussion are given in Section 5. Finally, the
work would provide an integrated transit system with an concluding remarks and future research are represented in
optimal accessibility for a larger variety of area-coverages Section 6.
with their services.
Based on such issues, this paper proposes a transit net-
work design framework to find an optimal bus transit sys- 2. Literature review
tem in combination with an existing railway system dealing 2.1. Literature on multi-objective approaches
with both fixed and variable demands while satisfying mul-
tiple objectives including user, operator and unsatisfied Transit Planning is classified according to the aimed design
demand costs. The suggested model has been applied to and period and its strategic, tactical and operational stages.

CONTACT Seungmo Kang [email protected] School of Civil, Environmental and Architectural Engineering, Korea University, 145, Anam-ro, Seongbuk-gu,
Seoul, Republic of Korea.
ß 2020 Taylor & Francis Group, LLC
2 M. H. ALMASI ET AL.

Route design is part of the strategic stage, frequency settings dominated Sorting Genetic Algorithm II (NSGA II) (Deb
are undertaken at the tactical stage, the design is finalized et al., 2002) and multi-objective evolutionary algorithms,
and the remaining activities are handled at the operational and local search procedures to solve the problem
stage (Desaulniers & Hickman, 2007). If a route design is formulation.
combined with a frequency design, the problem becomes Ibarra-Rojas et al. (2014) proposed a multi-objective
TNDFSP. An extensive review on design, frequency setting, approach to solve timetabling and vehicle scheduling prob-
timetabling of transit lines, and their combinations can be lems in transit systems. They represented the tradeoff
found in the paper by Guihaire and Hao (2008), who classi- between the level of service and the operator costs for scen-
fied 69 approaches dealing with the problem using various arios of up to 50 lines and five depots in their case study
objectives and solution methods. Also, Kepaptsoglou and located in Monterrey, Mexico.
Karlaftis (2009) presented a comprehensive and systematic Arbex and da Cunha (2015) examined the multi-objective
review of research on TNDP, based on design objectives, approach and suggested a GA heuristic for both network
operating environment parameters, and solution methods. It design and frequency setting minimizing user and operator
is a common argument that bus route network design and costs, however, the demand was not variable, while it could
frequency setting should be dealt simultaneously (Guihaire be changed by a competition with other modes. (See Lee
& Hao, 2008). and Vuchic (2005) for the importance of constraint on the
Most of the main objectives in the literature are about variable demand that requires a modal split procedure.) Fan
optimizing the problem to achieve minimum user and oper- and Machemehl (2006) proposed a multi-objective mathem-
ator costs. Cost estimates can be required at early project atical model including the variable demand, but they only
stages, before completion of a detailed design for several tested it with a number of different weight combinations. A
purposes, including budgeting and feasibility decisions. This rare example of dealing with the variable demand in the
estimation is usually determined by a parametric modeling multi-objective platform is the study of Guti
errez-Jarpa et al.
technique (Sonmez & Ontepeli, 2009). Xiong et al. (2013) (2017), which found optimum routes with variable demands
developed a solution for the optimal routing design problem for a rapid transit network. However, the test problem size
with the objective of minimizing the total cost, including was relatively small, yielding less than 6 lines in a medium-
user and supplier costs, considering passenger traffic level network.
demand and budget constraints. Kuah and Perl (1989) opti- Transfer coordination is also a major goal in many stud-
mized routing structures and operating frequency to minim- ies. The global network schedule should examine each trans-
ize the total bus operator costs. User costs include the bus fer area and its associated routes in order to allow efficient
riding time, waiting time, and rail costs. Golub et al. (2009) transfers among transit lines in regards to distance and
showed in their research that improvements of buses and time. Transferring between the lines can be achieved in
upgrading the train system are key components to achieve a accordance with various criteria including the number of
reduction in travel time for most origin-destination (OD) travelers. Wirasinghe et al. (1977) designed a multimodal
pairs and overall safety improvements. transit system that served peak travel time periods of an
The efficiency of a transport system depends on several urban area and a central business district. They developed a
elements such as available technology, governmental policies, model with three related variables (i.e., inter station spacing,
the planning process, and control strategies. Indeed, the feeder bus zone boundary, and train headways) to minimize
interaction between these elements is quite complex, leading the total operator and user costs. Moreover, bus-network
to intractable decision-making problems (Ibarra-Rojas et al., designs with schedule coordination have been studied by
2015). In order to overcome these problems, multi-objective many researchers (Shrivastava & Dhingra, 2006; Shrivastava
approaches will be useful. Mauttone and Urquhart (2009) & O’Mahony, 2007; Verma & Dhingra, 2006). They opti-
proposed a multi-objective optimization approach for the mized the coordination scheduling, while minimizing wait-
TNDP minimizing users’ costs (based on in-vehicle time, ing and transfer times for the rail stations.
waiting time, and transfer time) and the fleet size.
Moreover, numerical results on instances presented by
2.2. Literature on solution methods
Mandl (1980), show that the proposed heuristic is capable of
obtaining more Pareto optimal solutions that use a weighted In terms of the solution approaches, Wirasinghe (1980)
objective function. applied a mathematical technique for analyzing a case of
Chew et al. (2013) formulated a bi-objective version of feeder bus access to a rail line on a rectangular street net-
the TNDP that minimizes the user and operator costs sub- work. Buses along parallel routes fed passengers to the
ject to constraints such as a limited number of lines, limited rail lines.
number of line nodes, and a limited percentage of passen- Moreover, an analytic model was introduced by Kuah
gers using several transfers. To solve the proposed approach, and Perl (1988) for designing an optimum bus network for
a GA was designed and tested on a benchmark data set. granting accesses to an existing railway. In order to reduce
Kwan and Chang (2008) presented a bi-objective formula- costs, they utilized their mathematical technique and
tion for the TNDP minimizing the cost of the number of avoided the synchronized combination of the decision varia-
transfers and minimizing the cost caused by deviations from bles (i.e., stop spacing, route spacing, and operat-
an initial time-tabling. The authors implement the Non- ing headway).
 INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 3

Shrivastav and Dhingra (2001) developed a heuristic algo- to these problems, as used in Kwan and Chang (2008),
rithm to integrate suburban train and bus services. They Chew et al. (2013), and Arbex and da Cunha (2015).
optimized bus schedules in coordination with those of sub- Therefore, in this paper, a multi-objective optimization
urban trains. In addition, Chien (2005) proposed a unique model using a non-dominated sorting GA (NSGA-II) has
feeder bus network for efficient public bus services by con- been considered and applied for finding optimal solutions.
necting a recreational center and a main transportation facil- This paper models the TNDP with appropriate metaheuristic
ity. In their study, an integrated methodology was proposed algorithms in single and 3-dimensional multi-objective
for the improvement and optimization of key deci- approaches to overcome the above difficulties.
sion variables.
Ul Abedin et al. (2018) compared models and methodol-
ogies by creating a quality-evaluation platform that enables 3. Problem statement and objective function
comparisons of qualities of their solutions. The developed
evaluation framework consists of an integration component In metropolitan cities with high demands, an integrated
transit system plays an essential role of providing a sustain-
between criteria and objectives of passengers, agencies, and
able public transportation. Accordingly, designing appropri-
authorities. They selected a set of performance indicators
ate transit services that provide acceptable accessibility along
from the literature to build an evaluation rating system. The
the entire network that coordinate with the schedules are
evaluation results reveal that the quality of solutions declines
crucial for reliable transit operations. The main objective of
with an increase in the network size.
this study is to design an integrated transit system to
As Baaj and Mahmassani (1991) and Fan and Machemehl
increase the efficiency with the aim of minimizing costs and
(2006) noted, because of this exponential complexity and
achieving an optimal balance among the operators’, users’
NP-hard nature of TNDP, metaheuristics have been found
and unsatisfied costs.
to be suitable for large size practical problems. Kuan et al.
The optimal transit-route-network design includes deter-
(2004, 2006) utilized metaheuristic approaches including
mining a network configuration with a set of transit routes
ant-colony optimization (ACO), genetic algorithms (GAs),
and associated service frequencies that achieve a desired
tabu search (TS), and simulated annealing (SA) to determine objective via stated constraints. Indeed, a fixed rail network
the best solution for the feeder network design problem. and its associated scheduling are considered in the proposed
Several random tests have been carried out to evaluate the transit model and optimization model. Therefore, in the pre-
performance of the solution methods in terms of effective- sented model, the optimum solution is obtained base on the
ness and accuracy. Shrivastava and O’Mahony (2009) pro- whole transit service (bus and rail). The objective function
posed a model combining GAs and heuristic approach to implemented by various researchers vary widely because of
improve public bus services in a suburban area. Szeto and the operator’s diverse requirements, including minimizing
Wu (2011) utilized the GAs to change a route structure by operator cost, maximizing transfers saved, maximizing
representing each route-node-sequence as a substring. profit, etc. (Ceder & Wilson, 1986). An ideal objective func-
Therefore, a new operator became required to modify the tion should take care of the operator as well as the user per-
obtained routes (Chakroborty, 2003; Szeto & Wu, 2011). spectives in transit operations (Ceder & Israeli, 1998).
Cipriani et al. (2012) described a procedure and its applica- In this study, the transit demand matrix is applied for
tion for designing transit network and frequency setting in fixed and variable demands. The aim is to find an optimum
the city of Rome, considering complex road network and set of routes that correspond to chosen tradeoffs among
multimodal public transport system using GAs. The proced- user cost, operator cost and unsatisfied demand cost.
ure includes generating a route-set and finding a sub-opti- Operator costs can be decreased by an overall cooperation
mal set of routes with associated frequencies. among different public transportation modes. The satisfac-
Also, a swarm intelligence (SI) is considered as a rela- tion of passengers is achieved by serving a broader coverage
tively new metaheuristics approach for the transit-route net- area, decreased access and waiting costs, shorter travel times,
work design problem. The SI is inspired from social and smallest delay. The travel time is a significant variable
behaviors of insects in selecting their routes between their that has been identified as one of the key components for
nest and food. Tracking the behavior of ants in finding determining how the public perceives the quality of public
paths according to the density of their pheromone is found transportation (Dell’Olio et al., 2011).
in the literature (Yu & Yang, 2011). The bee-behavior is also The quality of travel itself also time plays an important
studied by Nikolic and Teodorovic (2013). role in increasing the likability of public transport among
Yao et al. (2014) presented a tabu search-based transit the public and estimating travel times for a public transpor-
network optimization method that considers a travel time tation system is a complex task. The unsatisfied demand
reliability. The model seeks to maximize the efficiency of cost is the penalty set for those passengers who cannot be
passenger trips in a transit network. The results show that served by the proposed transit route network. As a result,
the proposed method can effectively improve the reliability solution to the problem requires designing a system of bus
of a transit network and reduce travel times of passengers. routes and selection of a set of bus dispatching frequencies
In complex multi-objective problems, it is important to on each route. The total network cost is considered to be
find a large number of good solutions in the pareto front, the ‘objective function’ in this study. The total cost of the
and it makes GA, a population-based approach, well suited transit model is formulated as in the following:
4 M. H. ALMASI ET AL.

CT ¼ Cu þ Co þ Cud, (1) User

zfflfflfflfflfflfflffl}|fflfflfflfflfflfflffl{
where, CT, Cu, Co and Cud represent total cost, user cost, Minimize ðCw þ Cui Þ (11)
operator cost, and unsatisfied demand cost, respectively. To
Operating
present the mode more comprehensively and realistically, zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
this research tried to consider more cost terms when formu- Minimize ðCf þ Coi þ Cm þ Cp Þ (12)
lating user costs and operator costs. The user cost is related
Unsatisfied
to travelers and is formulated as the product of passengers’
z}|{
travel times and user’s value of travel time (i.e., value of Minimize Cud (13)
time for passenger’s waiting and in-vehicle costs). In this
study, the operation cost of bus or railway system was classi- Subject to: (4)(10)
fied into four parts: in-vehicle cost, maintenance cost, per- As shown in Table 3, decision variables can include dwell
sonnel cost and fixed costs. The cost is formulated as the time for the bus and train, travel waiting time along direct
product of the number of buses and trains per round trip and transfer routes, in-vehicle travel time along direct and
per each unit of time. Table 1 shows all cost terms in the transfer routes, bus transit travel demand along direct and
proposed model. All parameters and variables of the formu- transfer routes, bus frequency for each route, demand of
lated intermodal transit model are described in Table 2. each route, and length of each route, that are stated in
The objective is to minimize the sum of user cost, oper- Equation (3) and described in the following.
ator cost and unsatisfied demand cost for the studied bus The first term seen in Equation (3) refers to the user-wait-
transit network. The objective function is as follows: ing costs, which contain passengers’ waiting times for the
User Operating Unsatisfied buses (wtdij) and trains (wttrij). Then they are combined with
zfflfflfflfflfflfflffl}|fflfflfflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{ z}|{ the value of users’ waiting time, average wait time, and passen-
Minimize CT ¼ ðCw þ Cui Þ þ ðCf þ Coi þ Cm þ Cp Þ þ Cud ger demand for the buses (qdij) and trains (qtrij). In the pre-
(2) sented model, one half of determined headway is assumed to
Therefore, it can be formulated after substitution of all be the average waiting times at the stops and stations.
cost terms as given below: User in-vehicle costs contain in-vehicle time, passenger
demand, and the value of user in-vehicle time. This cost,

2 3
I h
I X
X    i
6 lw wtijd :qdij þ wtijtr :qtrij 7
þ lI tijd :qdij þ tijtr :qtrij þ pt :ntrij :qtrij þ
6 7
6 i¼1 j¼1 7
6X 7
6 K    7
6
Minimize CT ¼ 6 kf þ kp :ð2Fk :Tk þ Qk :td þ 2Fk :Sk Þ þ ðkl þ km Þ:ð2Fk :Lk Þ þ kI :Qk :td þ 7
7
6 k¼1 7
6 i 7
6 XK  XX h 
I I 7
4 þkIT ðQk  tdT Þ þ ðFT  TT Þ þ lud pud qij  qdij  qtrij 5
k¼1 i¼1 j¼1

(3) subject to: Cui, is formulated based on the average trip time and is
lmin  L  lmax k ¼ 1, :::, K (4) determined in two main parts: the user dwell time and the
K
user running time. User running costs for all travelers (Crui,)
Qmax depend on the travel time on buses and rail system links.
lfmax  k
 lfmin k ¼ 1, :::, K (5) The dwell costs are a function of demands and dwell times
Fk :C
at the stations and stops.
fmin  Fk  fmax k ¼ 1, :::, K (6) The time durations spent for boarding and alighting, are
Qk dependent on the number of passengers and the dwell times at
 Fk k ¼ 1, :::, K (7) each stops and stations, and hence different. The dwell time has
LF:C
an important role in determining user in-vehicle cost. Because of
X
K
  the variation in the time spent, a geometric series equation was
ð2Fk :Tk Þ þ ðQk :td Þ þ ð2Fk :Sk Þ  N (8)
adopted by Almasi et al. (2016, 2015) to present a new model to
k¼1
determine user dwelling costs along the bus routes and rail links.
K  Kmax (9) The operator costs are formulated as the sum of Coi, Cm,
tmin  tij  tmax i ¼ 1, :::, I j ¼ 1, :::, J (10) Cp, and Cf. The buses’ fixed and personnel costs are deter-
mined according to the fleet size; and hourly fixed and per-
The multi-objective form of the problem divides the sonnel cost for the modes are specified for transit services.
objective function into three parts. To improve accuracy, the dwell time and bus-mode slack
 INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 5

Table 1. Illustration of the total cost with all terms in the proposed improved model.
Total cost (CT)
User cost (Cu) Operator cost (Co)
Unsatisfied cost (Cud)
User Fixed
Waiting cost in-vehicle cost cost Operating in-vehicle cost Maintenance cost Personnel cost Unsatisfied cost
Cw Cui Cf Coi Cm Cp Cud
Cu¼CwþCui Co¼CfþCoiþCmþCp Cud

Table2. Description of parameters used in the proposed transit model.

Parameter Description Unit
CT Total system cost ($/h)
Cu User cost ($/h)
Co Operation cost ($/h)
Cud Unsatisfied demand cost ($/h)
Cw Waiting cost ($/h)
Cp Personnel cost ($/h)
Cui User in-vehicle cost ($/h)
Coi Operating in-vehicle cost ($/h)
Cf Fixed costs ($/h)
Cm Maintenance cost ($/h)
lw Passenger waiting cost value ($/passenger-h)
lI Passenger riding cost value ($/passenger-h)
lud Unsatisfied demand cost value ($/passenger-h)
kf Fixed cost of bus ($/veh-h)
kl Vehicle operator cost of bus ($/veh-km)
kI Vehicle operator cost of bus ($/veh-h)
km Maintenance cost of bus ($/veh-km)
kp Personnel cost of bus ($/veh-h)
klT Unit vehicle operator cost of train ($/veh-h)
Pud Time penalty for unsatisfied demand (h/passenger)
pt Time penalty associated to a transfer (h/passenger)
ntrij Number of transfer on travel between node i and j –
Vk Operating bus speed of route kth (km/h)
Sk Slack time route kth (min)
tij Total travel time between node i and j (h)
Tk Bus running time of kth bus route (h)
tmin Minimum travel time between each origin and destination (h)
tmax Maximum travel time between each origin and destination (h)
Tmin Minimum one way travel time of one each bus route (h)
Tmax Maximum one way travel time of one each bus route (h)
TT Train link travel time (h)
FT Frequency of train (veh/h)
fmin Minimum frequency (veh/h)
fmax Maximum frequency (veh/h)
N Total fleet size of bus (veh)
C Capacity of bus (passenger/veh)
LF Load factor of bus (passenger/seat)
lfmin Minimum load factor of one route (passenger/seat)
lfmax Maximum load factor of one route (passenger/seat)
lmin Minimum length of one route (km)
lmax Maximum length of one route (km)
qij Bus transit travel demand between node i and node j (passenger/h)
Qmaxk Maximum flow demand occurring on the route kth (passenger/h)
I Number of bus stops –
K Number of routes –
Kmax Maximum number of routes –

time are used in this study. The buses’ maintenance cost (Cm) the transit model, one cost value for rail operation system is
depends on the round-trip distance and fleet size given in assumed. All the components of operator cost containing in-
Equation (3). The buses’ operating in-vehicle cost is derived vehicle, fixed, maintenance, and personnel costs, are repre-
based on the dwell cost (Cdoi) and the running cost (Croi). The sented by kIT. The operator cost for a rail system is deter-
stop delay time incurred at feeder stops, and the running cost mined by multiplying kIT by the fleet size. More
for the feeders are defined according to the round-trip link information on operator cost for a rail can be found in
time. The derivation of operating in-vehicle cost for buses is detail in the literature (Almasi et al., 2015).
represented in the literature in detail (Almasi et al., 2016). The unsatisfied demand cost is the penalty assumed for
The operator cost for a rail service depends on passenger those passengers who cannot be served by the proposed
demand, fleet size of the rail network, and route station dis- transit route network. The last term seen in Equation (3) is
tance. Since a fixed rail way service is assumed, to simplify unsatisfied demand costs, which is resulted by the value of
6 M. H. ALMASI ET AL.

Table3. Description of decision variables used in the proposed transit model.

Decision variable Description Unit
td Dwell time for boarding and alighting from the bus (h/passenger)
tdT Dwell time for boarding and alighting to the train (h/passenger)
wtijd Waiting time on travel between node i and j along direct route (h)
wtijtr Waiting time on travel between node i and j along transfer path (h)
tijd Total in vehicle travel time between node i and j along direct route (h)
tijtr Total in vehicle travel time between node i and j along transfer path (h)
Fk Frequency of bus on route kth (veh/h)
Lk Length of route kth for the bus (km)
qdij Bus transit travel demand between node i and node j along direct route (passenger/h)
qtrij Bus transit travel demand between node i and node j along transfer path (passenger/h)
Qk Demand of route kth (passenger/h)

unsatisfied demand cost (lud), time penalty for unsatisfied feasible implementation of the RGP. Parallel implementa-
demand (Pud), and non-assigned demand. tions, including performance analysis of two prominent
There are six constraints that were considered and graph algorithms (i.e., Floyd-Warshall and Dijkstra) were
included for the bus transit network design problem in this utilized by Pradhan and Mahinthakumar (2013) in a large-
research, namely route length, load factor, frequency, fleet scale transportation network. The structure of the RGP is
size, maximum numbers of routes, and trip length. These shown in Figure 2.
constraints are related to user and operator and the subse- The knowledge of the transit planners has a significant
quent table presents all terms and descriptions. The method- impact on building the initial route-set skeletons, that is, dif-
ology and solving strategy are given in detail in the ferent user requirements result in a different route-set-solu-
following section. tion space.

4. Solution methodology
4.2. Network analysis procedure
The proposed solution methodology for transit network
NAP is an analytical tool that can evaluate and analyze the
problem consists of following components:
input bus transit network and possesses the ability to deter-
mine the route frequencies. Input data for the NAP contain
a. Initial Candidate Route Set Generation Procedure
the transit network information, demand data, and design
(RGP) generates the initial set of transit routes from the
given geography and network of the target area. parameters. The NAP in this research employs a multi-path
b. Network Analysis Procedure (NAP), that assigns the transit-trip assignment model that explicitly considers the
transit trips, determines the service frequencies on each transfer among routes. Moreover, the NAP can explicitly
route and computes other performance measures. consider the transit trip assignment model for the TNDP
c. Optimization algorithms using metaheuristic search under two distinct scenarios, i.e., fixed transit demand and
guides the search techniques with respect to the single variable transit demand. An iterative procedure is employed
and multi-objective optimization approaches. to achieve an internal consistency of route frequencies.
Furthermore, the iterative procedure in the NAP contains
The solution framework employed in this research is pre- two major components, namely, the transit trip assignment
sented in the Figure 1. model and the frequency setting procedure.
The improvement steps are iterated using the optimiza- Once a specific set of routes is proposed in the overall
tion algorithms until the termination criterion is met. The candidate-solution route-set generated by the RGP, the NAP
termination criteria are either reaching stable answers or is called to evaluate the alternative network structure and
completing a predetermined number of iterations (number determine route frequencies. Furthermore, the iterative pro-
of function evaluations). In this study, the maximum num- cedure in the NAP contains two major components, namely,
ber of iterations is considered to be the stopping condition. the transit trip assignment model and the frequency setting
Detailed information about the solution process is presented procedure. Once a specific set of routes is proposed in the
in the following subsections. overall candidate-solution route-set generated by the RGP,
the NAP is called to evaluate the alternative network struc-
ture and determine route frequencies.
4.1. The initial candidate route set The proposed optimal transit route network configur-
generation procedure
ation, the route frequencies and associated transit demand,
The initial candidate route set generation procedure is a along with a wide variety of performance measures, are then
design algorithm that configures all candidate routes for the obtained in the output data from the NAP, which can be
current transit network. Generating and determining the ini- categorized as network description data, demand data, cost
tial routes by applying local search algorithms, with a rea- terms, and level of service. Figure 3 presents the flow chart
sonable computation time, is essential for a successful and for the NAP.
 INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 7

Figure 1. Framework of transit network analysis and optimization procedure.

The applied trip assignment model had been presented  

1
=Crkij
by Horowitz (1987) initially, and followed by Baaj and dij r ¼ PK   dij (14)
k 1=
Mahmassani (1991), Shih et al. (1997) and Fan and k¼1 Crkij

Machemehl (2006), with some modifications made to

where Cr kij denotes the total travel cost of transit route rkij
accommodate more complex considerations for real-world
and dij r is the demand assigned to rkij :
applications. In this research, the transfer and travel time k
Note that the transit demands can be assigned to com-
minimizations are considered as the primary criterion for
bined paths of bus and rail line, or rail-only paths if the
transit user route choice. It is assumed that trip-makers
above assignment criteria (number of transfer and travel
always choose the path that has the lowest number of trans-
time) are met. All the bus stops that are transferable to rail
fers to get to the destination. If no paths are connecting OD
within the maximum number of transfers (for example, two stations are predefined. Any generated bus lines that include
transfers), the corresponding OD demand is categorized as those transferable stops may constitute the combined paths
“no service available” or “unsatisfied,” and not considered of bus and rail. Transfer time between bus and rail line is
afterward in the transit assignment process. considered based on the distance from the bus stop to the
To consider the unquantified factors in the route choice rail platform, and frequency and headway of the rail service.
process, the demand is split between multiple transit paths, Timetable synchronization between bus and rail is not con-
if they exist, with the lowest number of transfer and whose sidered because the headway of rail is comparably short (less
travel time is within a particular range from the minimum than 6 min). Demand share is calculated based on their
travel time, instead of assigning all demands to a single min- travel time using the same methods as the bus services.
imum cost path. Fan and Machemehl (2006) suggested the The actual interactions between the transit network pat-
inversely proportional model for the split of transit terns and the variable demands steer the process toward the
demands. optimal solution of TNDP with fixed transit demand, which
8 M. H. ALMASI ET AL.

Figure 2. Skeleton of the initial candidate route set generation procedure.

may not be realistic in real-world. Therefore, we propose wherecm ij is the generalized cost and b is the parameter for
that, when considering the TNDP, one needs to take the the calibration. The NAP is used as a transit network evalu-
variable demand nature into account. Figure 4 shows a cyc- ation tool for the TNDP with a variable transit demand with
lic relationship regarding the TNDP with the variable tran- the ability to decide the transit trip demand between each
sit demand. node pair, assign the transit trips to each route on the pro-
A binary logit model is applied in this research, where posed solution network and determine associated route fre-
the travel choice between two modes is made. The traveler quencies. To accomplish these tasks for the TNDP with a
will associate a certain value for the utility of each mode. variable transit demand, NAP employs an iterative proced-
The disutility here is the travel cost. This can be represented ure that seeks to achieve an internal consistency of the tran-
as in Equation (15). sit trip demand and the route frequencies. Furthermore, the
iterative procedure in the NAP contains three major compo-
if ¼ a1 atif þ a2 wtif þ a3 vtij þ a4 fij þ a5 /j þ d
m
cm m m m m
(15)
nents, namely, a transit demand equilibrium procedure, a
where vtijm is the in-vehicle travel time between i and j, atijm transit trip assignment procedure and a frequency setting
is the accessing time to and from stops, wtijm is the waiting procedure. Figure 5 shows the flow chart for the transit
time at stops, ftijm is the fare charged to travel between i and assignment procedure with a variable demand.
j,øm
j is the parking cost, and d is a parameter representing The iterative frequency setting procedure is applied to
comfort and convenience. cm ij is the cost of traveling from yield internally consistent service frequencies when it is
zone i to zone j using the mode m. And the proportion of coupled with the transit trip assignment model. The route-
trips by mode m from zone i to zone j is pm ij : The relation- based transit demands are obtained by assigning the trips in
ship is expressed by a logit curve and therefore the propor- a transit demand matrix using the trip assignment model.
tion of trips by mode m is given by Equation (16) as given Based on these transit demands, the total transit trips using
follows: route K and its correspondingQmax k , the maximum link flow
of route k are computed. As a result, assuming that the tran-
ebcij
m
Tijm
pm
ij ¼ P M m
¼ P M bcm
m ¼ 1,    M (16) sit demand is symmetric, Fk, the route frequency for route k
m¼1 Tij m¼1 e
ij
can always be computed as in the following:
 INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 9

Figure 3. Flowchart for the network analysis procedure.

Figure 4. Cyclic relationships regarding TNDP with variable transit demand.

Qmax 4.3. Application of optimization algorithms

Fk ¼ k
(17)
LF:C
In this section, the metaheuristics used to improve the
Obviously, different values for maximum load factors can TNDP is briefly explained. There are lots of optimizers
meet different operational considerations of transit operators. being used to solve transit network design problems. Based
10 M. H. ALMASI ET AL.

Figure 5. Flowchart for the transit assignment procedure with variable demand.

on the literature, there are pros and cons for all these opti- selection seen in nature, which indeed, results in better solu-
mization methods. The GA, and NSGAII algorithms, labeled tions in real-life over generations of different species.
for single and multi-objective optimizers, have shown great The conceptual simplicity may be considered as the main
potentials for solving optimization problems as they employ advantage and efficiency of GAs for solving complicated
a global stochastic search (Deb et al., 2002; Holland, 1975; optimization problems (Golberg, 1989). From the perspec-
Mateescu, 2006). tive of applications, many engineering optimization prob-
The main reason for applying and considering these algo- lems have been solved for finding the optimum solutions
rithms in this study is that the reported optimizers are the using the GAs due to their strong search ability and efficient
state-of-art optimization methods for solving wide range of search strategy.
optimization problems. In fact, GAs are the most well- Regarding the search operators used in GAs, crossover,
studied and well-used metaheuristic optimization methods mutation, and selection strategy are the most important fac-
during the past two decades (Nayeem et al., 2014). However, tors. Similar to other population-based optimization meth-
it cannot be concluded that the reported metaheuristic ods, GAs start with an initial population of individuals
methods applied in the proposed model are the best known as a population of chromosomes (Mateescu, 2006).
approaches, as other existing methods may be implemented As the most other optimizers do, GAs also start their opti-
in the future for the assessment of their performances. mization processes by randomly generating initial popula-
tions. A population size (number of chromosomes in the
population) is a user parameter value and can vary.
4.3.1. Genetic algorithms The next step is a function evaluation process to see how
Genetic algorithms (GAs) are categorized as an evolutionary fit/good the solutions are based on the cost/fitness function
optimization algorithm and is considered as one of the defined by a given problem (i.e., minimization or maximiza-
widely used optimizers in the literature. The basic idea of tion). In this study, the cost function is the sum of user
GAs was first proposed by Holland (Holland, 1975). The cost, operator cost and unsatisfied demand cost. After calcu-
GA is inspired from the process of evolution by natural lating the fitness/cost function, the selection strategy should
 INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 11

be defined for the selection process of parents for creating sorting based on the rank scores are applied, the solutions
offspring using the crossover search operator. Selection pro- are sorted based on their crowding distance values. In a
cess such as tournament selection approach tries to choose crowding distance mechanism, the extreme values for each
the best solutions among (between) other solutions. There objective are assigned infinite values, for keeping these val-
are different methods for the selection process including the ues as best solutions. Therefore, if the crowding distance of
roulette wheel selection, however the fundamental idea is a solution is higher, the solution is more isolated, giving us
the same. Selecting better individuals giving higher chance more diversity of new candidate solutions. In the tourna-
to survive for the next generations is the key. ment selection process, the selected solutions are compared
After selecting better chromosomes (from two to many), first with the Pareto rank, and then if the rank is same, the
the crossover operator will be applied by combining some crowding distance is compared for tie-breaker. Please refer
aspects of selected individuals. For instance, from two Deb et al. (2002) for the detailed function evaluation and
selected parents, two offspring will be generated by transfer- selection process of NSGA-II. The rest of search operators
ring their features (attributes), then, a population of offspring used in NSGA-II such as crossover and mutation are identi-
can be formed. However, for having more diversity in the cal to the standard GAs.
population of chromosomes, mutation search operator is
applied with a certain level of randomness into the popula-
tion features. A ratio of mutation rate is applied to the entire 5. Numerical results and discussions
population with a random selection, making a new popula- The results of extensive computational experiments are pre-
tion of offspring formed using this search operator. sented to verify and illustrate the above design procedures
Finally, by combining three populations including the for fixed and variable demands. The proposed method was
current population, populations formed using crossover and applied in real-large-scale networks in Daejeon, South
mutation, we have a new population having more than the Korea. This city has a complex street network (see Figure 6)
predefined population size. After sorting the new combined and includes 2,675 nodes, 7,473 links, and about 1.55 mil-
population (descending or ascending based on minimization lion populations in 2014. The city is located roughly in the
or maximization concept), only chromosomes equal to the center of the nation with equidistant to other major cities
size of population will be kept and the rest of chromosomes including Seoul, Busan, and Gwangjoo. In other words, if a
will be discarded from the optimization task. The aforemen- successful implementation and operation of an enhanced
tioned processes will continue until the stopping condition transit management system can be achieved in Daejeon,
(e.g., the maximum number of function evaluation) is met. there is also a great potential for a proposed methodology to
be applied on the expanded inter-city transit platform that
4.3.2. Non-dominated sorting genetic algorithm II would reach and connect the other major cities in
A non-dominated sorting strategy has been used for con- the nation.
verting GAs into an efficient multi-objective optimizer. The Existing traffic studies provide the transit origin-destin-
optimization strategy behind the non-dominated sorting ation (OD) demand matrix in 76 traffic zones. The existing
genetic algorithms II (NSGA-II) is based on the standard transit network is composed of more than 70 bus lines, with
GAs. In the previous section, the detailed steps of GAs have many overlapping routes with moderate frequencies, and a
been explained. The selection process in the NSGA-II is single subway rail line. Information about the number of
based on the binary tournament selection with replacement passengers was collected from the Korean Transportation
as for GAs. Regarding the exploration phase, a random Database (KTDB).
mutation operator is applied to a small portion of solution The fundamental network data provided in input files
to ensure further exploring unobserved regions. include all the information for zone, node, link and network
If the newly generated solution does not satisfy the data for the current studied network. The data structures are
applied constraints even after recombination and/or muta- developed to organize these data so that the MATLAB pro-
tion operators are applied, the solution is ignored and gram was used to design the optimal transit route network.
another new solution will be created using the aforemen- The information about design-related parameters used in
tioned operators until a feasible solution is obtained. The the models are represented in Table 4.
difference of NSGA-II from the conventional GA lies in the The parameters are based on the data collected from the
function evaluation and selection process. The fitness func- Korean Transportation Database (KTDB), and from other
tion evaluation process has two steps. In the first step, as related studies in the literature such as ridership and finan-
the name of non-dominating approach indicates, the solu- cial reports that have been published by Barton (2006) and
tions are ranked based on the Pareto dominance. Then, all Valley Metro (2012).
solutions in the Pareto set receive a rank 1. Afterwards, Metaheuristics (i.e., GA and NSGA-II) were employed to
these solutions are extracted from the Pareto set, and all optimize the transit model in the case study. These optimiz-
Pareto solutions in the remaining set receive rank 2 and so ers have illustrated their capability as efficient optimization
on. For example, solution A is defined as non-dominated, if tools with great potentials for solving problems (Deb et al.,
there is no other solution in the solution set that are better 2002; Mateescu, 2006). The transit model and the reported
or same in every objective (in this study, there are three optimizers were coded, and run in MATLAB programming
objectives: operators’, users’ and unsatisfied costs). After the software. The optimization procedure of TNDP was
12 M. H. ALMASI ET AL.

Figure 6. Network in city of Daejeon, South Korea.

performed using 30 independent runs for each solution evaluations was set to 15,000 for both single and multi-
algorithm. For the sake of reducing the computing time, the objective optimization problems.
parallel computing strategy embedded in MATLAB was uti- The following subsections present the comprehensive
lized by using 5 standard PCs with 8 GB RAM, and i7-grade numerical results the applied solution methods have pro-
CPU (2017 model). duced for both single objective and multi-objective
The proposed GAs use stochastic uniform as a selection approaches. Furthermore, performance comparisons and
function, and rank as a scaling function. Regarding the stop- characteristics underlying the TNDP with both fixed and
ping condition, the maximum number of function variable demand are also discussed.
 INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 13

Table 4. Selected values for parameters of used in the study area.

Parameter Description Unit Value
lw Passenger waiting cost value ($/passenger-h) 8
lI Passenger riding cost value ($/passenger-h) 4
lud Unsatisfied demand cost value ($/passenger-h) 8
kf Fixed cost of bus ($/veh-h) 14.37
kl Vehicle operator cost of bus ($/veh-km) 0.36
kI Vehicle operator cost of bus ($/veh-h) 8.94
km Maintenance cost of bus ($/veh-km) 0.75
kp Personnel cost of bus ($/veh-h) 10.2
Pud Time penalty for unsatisfied demand (hr/passenger) 1
pt Time penalty associated to a transfer (min/passenger) 15
td Dwell time for boarding and alighting from the bus (min/passenger) 0.096
Vk Operating bus speed of route kth (Running speed) (km/h) 30
Sk Slack time route kth (min) 15
C Capacity of bus (passenger/veh) 27
LF Load factor of bus (passenger/seat) 2
lfmin Minimum load factor of one route (passenger/seat) 0.5
lfmax Maximum load factor of one route (passenger/seat) 2
fmin Minimum frequency (veh/h) 2
fmax Maximum frequency (veh/h) 30
lmin Minimum length of one route (km) 4
lmax Maximum length of one route (km) 30
tmin Minimum travel time between each origin and destination (min) –
tmax Maximum travel time between each origin and destination (min) 120
Tmin Minimum one way travel time of one each bus route (min) 15
Tmax Maximum one way travel time of one each bus route (min) 120
Kmax Maximum number of routes – 75
ftijp Fare charged to travel between i and j with public transit mode ($/passenger trip) 1
ftija Fare charged to travel between i and j with private car ($/veh-km) 0.344
øaj Private car parking cost ($/veh) 8

Table 5. Parameters and factors levels for both GA and NSGA-II. Table 6. Parameter set and computational time using the Taguchi method for
Factor level sensitivity analysis.
Parameters GAs NSGA-II
GA & NSGA-II 1 2 3 4 5
Npop 10 30 50 75 100 Computational Computational
Pc 0.5 0.6 0.7 0.8 0.9 Run no. Npop Pc Pm time (min) Npop Pc Pm time (min)
Pm 0.1 0.3 0.4 0.7 0.9 1 10 0.5 0.1 1325.18 10 0.5 0.1 1450.70
2 10 0.6 0.3 1214.90 10 0.6 0.3 1594.73
3 10 0.7 0.4 1470.81 10 0.7 0.4 1619.52
5.1. Parameter sensitivity analysis 4 10 0.8 0.7 1483.44 10 0.8 0.7 1594.32
5 10 0.9 0.9 1347.26 10 0.9 0.9 1535.28
Since the quality of the solutions obtained by metaheuristic 6 30 0.5 0.3 1346.78 30 0.5 0.3 1564.10
7 30 0.6 0.4 1301.32 30 0.6 0.4 1488.90
algorithms depends on the values of their initial parameters, in 8 30 0.7 0.7 1470.02 30 0.7 0.7 1623.41
this paper, the Taguchi method of design of experiment 9 30 0.8 0.9 1310.77 30 0.8 0.9 1456.69
(Montgomery, 2005) is employed to evaluate the influence of 10 30 0.9 0.1 1233.36 30 0.9 0.1 1606.03
11 50 0.5 0.4 1434.08 50 0.5 0.4 1455.05
used parameters on the performance of GAs and NSGA-II. The 12 50 0.6 0.7 1316.92 50 0.6 0.7 1510.55
Taguchi method is a fractional factorial experiment method as 13 50 0.7 0.9 1272.51 50 0.7 0.9 1587.69
an efficient alternative for full factorial experiments. 14 50 0.8 0.1 1321.17 50 0.8 0.1 1634.07
15 50 0.9 0.3 1228.94 50 0.9 0.3 1424.34
The compared parameters of GAs and NSGA-II are the 16 75 0.5 0.7 1239.59 75 0.5 0.7 1678.82
population size (Npop), the cross-over probability (Pc), and 17 75 0.6 0.9 1482.62 75 0.6 0.9 1632.71
the mutation probability (Pm). The maximum number of 18 75 0.7 0.1 1486.84 75 0.7 0.1 1546.04
19 75 0.8 0.3 1372.56 75 0.8 0.3 1530.76
function evaluations (NFEs) of 15,000 was assumed as the 20 75 0.9 0.4 1217.93 75 0.9 0.4 1534.04
stopping condition for this sensitivity analysis. Table 5 21 100 0.5 0.9 1270.43 100 0.5 0.9 1491.90
22 100 0.6 0.1 1305.95 100 0.6 0.1 1552.55
shows the algorithms’ parameters, each at five levels. These 23 100 0.7 0.3 1446.36 100 0.7 0.3 1553.23
ranges are general in most of GA studies and accordance 24 100 0.8 0.4 1204.62 100 0.8 0.4 1645.29
with the survey result of Hassanat et al. (2019). Instead of 25 100 0.9 0.7 1212.91 100 0.9 0.7 1638.45
full 53 combinations, Taguchi’s L25 orthogonal array allows
the efficient analysis with specific combinations of only 25
experiments. Refer Montgomery (2005) for detailed sam- interval of the suggested level of system re-design is usually
pling process of Taguchi method. longer than a few years, the computational time shown in
Table 6 tabulates 25 experiments on GAs and NSGA-II this study does not critically curtail the feasibility of the pro-
along with their computational time in minutes. Regarding posed algorithm. Nevertheless, more efficient algorithms or
the computational efficiency, it is noted that this problem is processing methods could help develop more or better
related to a large city-wide transit system design and thus options within a given time in the real- world case studies
does not require near real-time solutions. Given that the in the future.
14 M. H. ALMASI ET AL.

Figure 7. Objective value plots of each level of the factors for (a) GAs, (b) NSGA-II.

Table 7. Comparison of existing and proposed services with fixed demand.

Proposed service (B) Proposed service (C) Proposed service (D)
Existing service (A) (B/A-1) (C/A-1) (D/A-1)
Total cost ($/h) 161,700 143,920 –11% 144,520 –11% 147,720 –9%
Operator cost ($/h) 45,064 41,084 –9% 44,526 –1% 46,304 3%
User Cost ($/h) 116,640 102,280 –12% 99,940 –14% 101,420 –13%
Assigned demand 52,033 52,022 0% 52,032 0% 52,033 0%
Number of route 70 74 6% 64 –9% 72 3%
Fleet size 936 919 –2% 951 2% 984 5%
Direct trip demand (%) 55.1% 55.6% 1% 54.3% –2% 57.1% 4%
Total trip time (min) Max. 66 66.1 0% 68 3% 68.2 3%
Mean 30 27.1 –10% 27.3 –9% 27.9 –7%
Route length (km) Max. 28.2 20.5 –27% 27.9 –1% 27.9 –1%
Mean 16.2 10.2 –37% 12.6 –22% 13.1 –19%
Total routes length (km) 1,148.7 768.3 –33% 943.2 –18% 984.3 –14%
Total operating length(km/h) 6,696.3 5,073.4 –24% 6,270.6 –6% 6,712.3 0%
Headway Min. 4 2.4 –41% 2.5 –39% 2.1 –48%
Max. 35 30 –14% 30 –14% 30 –14%
Mean 10.6 9.2 –13% 8.8 –17% 9.8 –7%

Based on the results of independent tests, as shown in additional alternatives that may also be practically and polit-
Figure 7, the best values of the initial parameters are ically feasible as well. This also shows the multi-objective
Npop¼100, Pc ¼0.7, Pm ¼0.3 for both the GAs and NSGA-II. nature of TNDP. In this section, the existing service is com-
pared with the first, second and third best solutions in terms
of the total cost (fixed demand case) and the total cost per
5.2. Single objective approach demand (variable demand case) among the final generation
population of a single run of GA.
The transit network design in this section includes two main
As it can be seen in Table 7, the best result in terms of
purposes, namely, simulation of existing network, and pro-
the objective function value is associated with the proposed
posing new solutions. The numerical results were compared
Service (B) with a total cost of $1,43,920. The total cost
according to performance measures such as demand satisfac- value of proposed service (B) and proposed service (C) (the
tion levels, user travel costs, operation costs, etc. Table 7 second solution in the objective function) were 11% lower
summarizes the numerical results obtained by applying the than the existing services. Proposed service (C) gave solu-
TNDP with a fixed demand. This table shows the values of tions with user’s cost of 14% lower than the existing serv-
the different terms of the objective function and comparison ices. On the other hand, proposed service (B) shows the
among all solutions and existing network. operator cost with the lowest value of $41,084 which was
Note that because GA generates a number of solutions as 29% of the total cost. The total demand satisfied by all net-
much as the population size at each generation, it is easy to works was almost the same, thus fulfilling an important
identify first, second and third solutions. It would be useful requirement of a transit demand allocation.
to compare those solutions with the existing service and the A direct demand allocation is a very important network
optimal solution for a decision maker to consider the evaluation parameter, which takes into a consideration the
 INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 15

Table 8. Comparison of existing service and proposed services with variable demand.
Redesigned Existing Service
(H) Proposed Service (E) Proposed Service (F) Proposed Service (G)
Existing Service (A) (H/A-1) (E/A-1) (F/A-1) (G/A-1)
Total cost ($/h) 161,700 155,601 –4% 147,872 –9% 147,263 –9% 146,423 –9%
Total cost per demand 3.108 2.824 –9% 2.617 –16% 2.637 –15% 2.651 –15%
Operator cost ($/h) 45,064 50,301 12% 45,272 0% 42,763 –5% 44,993 0%
User cost ($/h) 116,640 105,300 –10% 102,600 –12% 104,500 –10% 101,430 –13%
Assigned demand 52,033 55,097 6% 56,514 9% 55,842 7% 55,239 6%
Transit mode share 25.9% 27.4% 6% 28.1% 9% 27.8% 7% 27.5% 6%
Number of route 70 70 0% 73 4% 74 6% 65 –7%
Fleet size 936 1,048 12% 961 3% 957 2% 956 2%
Direct trip demand (%) 55.1% 61.2% 11% 62.2% 13% 55.9% 1% 60.2% 9%
Total trip time (min) Max. 66 524.1 12% 69.5 5% 59.9 –9% 75.2 14%
Mean 30 35.3 0% 28.6 –5% 27.2 –10% 28.9 –4%
Route length (km) Max. 28.2 28.2 0% 27.9 –1% 20.5 –27% 27.9 –1%
Mean 16.2 16.2 0% 13.1 –19% 10.2 –37% 12.5 –23%
Total routes length (km) 1,148.7 1,148.7 0% 984.3 –14% 767.8 –33% 934.4 –19%
Total operating length (km/h) 6,696.3 7,795.2 16% 6,478.4 –3% 5,371.4 –20% 6,415.7 –4%
Headway Min. 4 2.4 –41% 2.5 –38% 2.3 –42% 2.1 –48%
Max. 35 35 0% 30 –14% 30 –14% 30 –14%
Mean 10.6 9.7 –8% 10.2 –3% 9 –15% 9.1 –14%

reluctance of the users to transfer. The proposed service (D) had the lowest user cost and the lowest number of bus route
(the third solution in the objective function) satisfied more with the best service headway.
direct demands than all other cases with 57.1% of coverage Similar to the fixed demand scenario, the numerical
level. In addition, the proposed service (C) had a low per- results were compared according to performance measures
formance with the direct demand allocation of 54.3%. such as demand satisfaction levels, user travel costs, oper-
Moreover, it is important to highlight that, regardless of the ation costs, etc. Table 8 summarizes the results obtained by
number of routes of the transit network, the procedure pro- applying the TNDP with a variable demand. Table 8 shows
vides comparable and satisfactory results for all the gener- the values of the different terms of the objective function
ated solutions. The proposed service (B) gave the lowest and comparisons among all the proposed solutions against
fleet size of 919 buses and lowest operated bus kilometers the existing service (i.e., Service (A)) and redesigned existing
with a value of 5,073 kilometers per peak hour. The reason service (i.e., Service (H)). The results of the redesigned exist-
for having the lowest fleet size and bus kilometers, in spite ing service are obtained by applying the existing network
of the greater number of routes, is that the proposed service lines to the variable demand approach.
(B) operated more routes, but with lower length compared It can be observed that the best result in terms of total
to other services. cost per demand is associated with the proposed service (E)
Consequently, the operation cost presents the lowest with a value of $2.617 per passenger per trip. A total cost of
value relative to other services. An equally important evalu- $1,47,872 is realized and 73 bus routes are designed. The
ation parameter is the time component in general. The aver- range of service headways is from 2.5 to 30 minutes and the
age travel time was within the range of 27.1 and 30 minutes average headway is 10.2 minute. The provided total cost per
among all networks. The difference in average travel time demand presents 16% improvement compared to the exist-
between the existing and proposed service (B) shows a 10% ing network cost. The objective function values of the pro-
improvement. The existing network shows a relatively poor posed service (E) show 9% and 4% lower than the existing
result, with the highest total time of 30 min on average, services and redesigned existing service, accordingly. In
while the proposed service (B) has shown 27.1 min. terms of the user’s cost, the proposed service (G) (the third
The proposed service (B) performed the best in terms of solution in total cost per demand) was the lowest among
low travel time because of the larger number of routes com- other services and 13% and 10% lower than the existing
pared to other services. The total passenger travel time may services and redesigned existing service, respectively.
increase in transferred trips because of the additional time On the other hand, the operator cost was the lowest in
spent on board by continuing passengers at the transfer cen- the proposed service (F) (the second solution in total cost
ters. On the other hand, the in-vehicle travel time of trans- per demand), which was about 15% lower than the rede-
ferring passengers may decrease as they select the shortest signed existing service. Figure 8 shows the comparisons of
downstream path among all the competing paths at the the cost terms per satisfied demand. The difference in total
transfer points. cost per trip per passenger between existing service and pro-
To sum up for fixed demand scenarios, the proposed ser- posed service (E) shows about a 16% improvement. And
vice (B) gave the best solution in terms of the objective this improvement between redesigned existing service and
function value and fleet size, as well as the lowest average proposed service (E) is about 7.4%.
total trip time, and total routes-length of bus operations. The percentage of the total mode share demand for the
The proposed service (D) showed a solution with the highest proposed service (E) presents 28.12% of the total demand,
direct demand allocation, while the proposed service (C) while the percentage mode share value by the existing
16 M. H. ALMASI ET AL.

Figure 8. Cost terms per satisfied demand comparison in variable demand.

network shows 25.89% of total demand. Figure 9 shows the operated bus length and consequently, the lowest operator
total assigned demand and the percentage of transit mode costs. The proposed service (G) had the best solution in total
share for all scenarios in detail. cost, user cost, and number of bus routes.
The percentage of the direct demand satisfied ranges
from 55.1% to 62.2% and the proposed service (E) satisfied
5.3. Multi-objective approach
more direct demand than all other cases with 62.2% of
coverage level. The proposed service (F) gave the lowest It is critical for the transit authority to make decisions on
total route length and operated bus kilometers with a value how to assign for a new transit network to satisfy operators
of 767.8 kilometer and 5,371.4 kilometers per peak hour, and users with an optimal solution. The non-dominated sol-
respectively. utions are generally used as scenarios for further evaluations.
The reason for having these lowest values, in spite of the In fact, the three objectives including user cost, operator
largest number of routes, is that the proposed service (F) costs and assigned demand were contradicting with each
operated more routes, but with a lower length compared to other. Therefore, the improved multi-objective optimization
other services. Consequently, the operation cost presents the model was applied to explore the Pareto front set (non-
lowest value compared to other services. As shown in Table dominated solution) for the considered case study. Figures
8, the average travel time was within the range of 27.2 and 10 and 11 show the diagram of obtained Pareto solutions in
30 minutes for all networks. Moreover, the proposed service three-dimensional space obtained by the NSGA-II for fixed
(F) shows 10% lower rate than the existing service. This net- and variable demands, respectively.
work also performed the best in terms of the lowest average A three-dimensional illustration of the determined solu-
service headway of 9 minutes. In this regard, the existing tions can be utilized to visualize the tradeoffs between the
network performed poorly, with the highest average total user cost, operator cost, and number of assigned demands
trip time of 30 minutes. with the aim of supporting decision makers for evaluating
To summarize the variable demand scenarios, the pro- the effects of various transit network plans for the reported
posed service (E) gave the best solution in terms of total study area. Each point in Figures 10 and 11 represents a set
cost per demand, transit mode share, total direct demand of transit services. Decision makers can visualize and evalu-
allocation, and user cost, while the proposed service (F) ate the tradeoffs in order to organize an appropriate transit
showed the best results in service headway, total trip time, service. Also, Figure 12 denotes the values of the objective
 INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 17

Figure 9. Comparison of transit mode share.

Figure 10. Comparison of Pareto front obtained by the NSGA-II for the fixed demand, considering (a) all objectives, (b) operating vs. user costs, (c) satisfied demand
vs. user cost, and d) satisfied demand vs. operating cost.
18 M. H. ALMASI ET AL.

Figure 11. Comparison of Pareto front obtained by the NSGA-II for the variable demand, considering (a) all objectives, (b) operating vs. user costs, (c) satisfied
demand vs. user cost, and d) satisfied demand vs. operating cost.

functions obtained by Pareto solutions using the NSGA- interactive relationship of the three objectives and has
II algorithm. helped optimizing the multimodal transit network plans.
As can be seen in Figures 10–12, the range of hourly user Generally, costs increase with the increase of the assigned
cost, operator cost and assigned demand obtained for the demand. Since, there is no global solution methodology
fixed demand are between $26,797 to $41,275 $67,140 to found for the TNDP at the time of writing (i.e. solution
$1,11,628 and 35,697 to 52,035 passengers, respectively. methodology which minimizes operator and user costs and
Accordingly, the user cost, operator cost and assigned maximize assigned demand at the same time), the problem
demand range attained using the variable demand are, should be dealt with the multi-objective level approach.
respectively, $28,360 to $42,309 $79,175 to $1,06,039 and
42,202 to 55,533 passengers.
The results indicate that the variable demand offers a 6. Conclusions and future research
wider range of costs and satisfied demand compared with
the fixed demand approach. It can be concluded that the In this paper, an improved solution methodology for a sus-
variable demand approach performs more efficiently for tainable transit network design problem (TNDP) was pre-
large-size problems as well. These ranges are considered as sented with single and multi-objective approaches. The main
different tradeoffs between operator cost, user cost, and sat- goal of this research was to propose an integrated multi-
isfied demand-based solutions. modal transit model to design the efficient bus line network
Having obtained the Pareto optimal set, choosing the best connected to the existing railway line for fixed and variable
compromised solution is crucial for the decision-making demands while minimizing costs. This study represented an
process. In this paper, a descriptive approach was proposed optimum set of routes via metaheuristic optimization algo-
to help decision makers to find the best compromised solu- rithms including Genetic Algorithms (GAs) and the non-
tion. The developed model has revealed the quantitatively dominated sorting GA II (NSGA-II) that addresses chosen
 INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 19

Figure 12. Comparison of objective values for non-dominated solutions (a) fixed demand and (b) variable demand.

tradeoffs between user cost, operator cost, and unsatisfied optimization, and the descriptive approach showed the best
demand cost. The used input data was based on the actual compromised solution for the decision-making process. The
transit network in Daejeon City, South Korea, which is developed model revealed the quantitatively interactive rela-
located at a strategic central location and has potential to tionship of the three objectives and helped optimizing the
expand the transit networks reaching to other major cities multimodal transit network plans.
such as Seoul and Busan on which proposed model could be The results indicate that the variable demand offers a
implemented throughout the nation. wider range of costs and satisfied demand. It can be con-
The obtained optimization result from the single object- cluded that the variable demand approach performs more
ive approach has shown that the proposed optimized net- efficiently in large-size problems. When compared to single-
works acquired significantly higher performance measures objective optimization models, the multi-objective model
when compared with the existing network. The best solu- presents results with more valuable information to decision
tions obtained by the GAs with the minimum total costs, makers and is able to explore the realistic multimodal transit
are respectively 11% and 9% less than that of the existing network space with thorough examinations (Pareto optimal
service for the fixed and variable demand. In addition, a solutions). Therefore, the proposed methodology can be
case study of the transit network was performed with the considered as a potential alternative method to overcome
multi-objective optimization model. The three objectives current difficulties with the TNDP. This model may inspire
were integrated to evaluate the tradeoffs among them. The more realistic models in simulating real-life problems if fed
Pareto optimal set has been obtained through the by newly collected empirical data in the near future.
20 M. H. ALMASI ET AL.

The proposed framework can be easily expanded with Chakroborty, P. (2003). Genetic algorithms for optimal urban transit
additional constraints and objectives suggested by transit network design. Computer-Aided Civil and Infrastructure
Engineering, 18(3), 184–200.
operators, including limitation of transfer time, transfer Chew, J. S. C., Lee, L. S., & Seow, H.-V. (2013). Genetic algorithm for
demand, total or average length of lines, degree of accessibil- biobjective urban transit routing problem. Journal of Applied
ity, and distance gap between stops. These expansions will Mathematics, 2013, 1–15.
increase the practical applicability of this model to the Chien, S. (2005). Optimization of headway, vehicle size and route
choice for minimum cost feeder service. Transportation Planning
real world.
and Technology, 28(5), 359–380. https://doi.org/10.1080/
03081060500322565
Cipriani, E., Gori, S., & Petrelli, M. (2012). Transit network design: A
Author contributions procedure and an application to a large urban area. Transportation
Research Part C: Emerging Technologies, 20(1), 3–14. https://doi.org/
In this study, all of the authors contributed to the writing of
10.1016/j.trc.2010.09.003
the manuscript. Mohammad Hadi Almasi designed and per- Deb, K., Pratap, A., Agarwal, S., & Meyarivan, T. A. M. T. (2002). A
formed the experiments and analyzed the result. Yoonseok fast and elitist multiobjective genetic algorithm: NSGA-II. IEEE
Oh and Ali Sadollah contributed in building the model and Transactions on Evolutionary Computation, 6(2), 182–197. https://
developing the solution algorithm. Young-Ji Byon and doi.org/10.1109/4235.996017
Dell’Olio, L., Ibeas, A., & Cecin, P. (2011). The quality of service
Seungmo Kang advised the data analysis and coordinated desired by public transport users. Transport Policy, 18(1), 217–227.
the overall work. Desaulniers, G., & Hickman, M. D. (2007). Chapter 2, Public transit.In
C. Barnhart and G. Laporte (Eds.), Handbooks in operations research
and management science (Vol. 14, pp. 69–127). Elsevier B.V. https://
Funding doi.org/10.1016/S0927-0507(06)14002-5
Fan, W., & Machemehl, R. B. (2006). Optimal transit route network
This work was supported by Jungseok Logistics Foundation grant, the design problem with variable transit demand: Genetic algorithm
National Research Foundation of Korea (NRF) funded by the Ministry approach. Journal of Transportation Engineering, 132(1), 40–51.
of Science and ICT (2018R1A2B6005729) and the Ministry of https://doi.org/10.1061/(ASCE)0733-947X(2006)132:1(40)
Education (2020R1A6A1A03045059), and Abu Dhabi Department of Golberg, D. E. (1989). Genetic algorithms in search, optimization, and
Education and Knowledge (ADEK) Award for Research Excellence machine learning. Addison-Wesley.
(No. 8-434000104). Golub, A., Balassiano, R., Ara
ujo, A., & Ferreira, E. (2009). Regulation
of the informal transport sector in Rio de Janeiro, Brazil: Welfare
impacts and policy analysis. Transportation, 36(5), 601–616.
Guihaire, V., & Hao, J.-K. (2008). Transit network design and schedul-
ORCID ing: A global review. Transportation Research Part A: Policy and
Practice, 42(10), 1251–1273. https://doi.org/10.1016/j.tra.2008.03.011
Ali Sadollah http://orcid.org/0000-0002-7782-4126 Guti
errez-Jarpa, G., Laporte, G., Marianov, V., & Moccia, L. (2017).
Young-Ji Byon http://orcid.org/0000-0003-1209-1803 Multi-objective rapid transit network design with modal competi-
Seungmo Kang http://orcid.org/0000-0002-9435-5835 tion: The case of Concepci
on, Chile. Computers & Operations
Research, 78, 27–43.
Hassanat, A., Almohammadi, K., Alkafaween, E., Abunawas, E.,
References Hammouri, A., & Prasath, V. B. S. (2019). Choosing mutation and
crossover ratios for genetic algorithms - A review with a new
Almasi, M. H., Sadollah, A., Kang, S., & Karim, M. R. (2016). dynamic approach. Information, 10(12), 390.
Optimization of an improved intermodal transit model equipped with Holland, J. H. (1975). Adaptation in natural and artificial systems: An
feeder bus and railway systems using metaheuristics introductory analysis with applications to biology, control, and artifi-
approaches. Sustainability, 8(6), 537. https://doi.org/10.3390/ cial intelligence. University of Michigan Press.
su8060537 Horowitz, A. J. (1987). Extensions of stochastic multipath trip assign-
Almasi, M. H., Sadollah, A., Mounes, S. M., & Karim, M. R. (2015). ment to transit networks. Transportation Research Record, 1108,
Optimization of a transit services model with a feeder bus and rail 66–72.
system using metaheuristic algorithms. Journal of Computing in Ibarra-Rojas, O. J., Delgado, F., Giesen, R., & Mu~ noz, J. C. (2015).
Civil Engineering, 29(6), 04014090. https://doi.org/10.1061/ Planning, operation, and control of bus transport systems: A litera-
(ASCE)CP.1943-5487.0000418 ture review. Transportation Research Part B: Methodological, 77,
Arbex, R. O., & da Cunha, C. B. (2015). Efficient transit network 38–75. https://doi.org/10.1016/j.trb.2015.03.002
design and frequencies setting multi-objective optimization by alter- Ibarra-Rojas, O. J., Giesen, R., & Rios-Solis, Y. A. (2014). An integrated
nating objective genetic algorithm. Transportation Research Part B: approach for timetabling and vehicle scheduling problems to analyze
the trade-off between level of service and operating costs of transit
Methodological, 81, 355–376. https://doi.org/10.1016/j.trb.2015.06.014
networks. Transportation Research Part B: Methodological, 70,
Baaj, M. H., & Mahmassani H. S. (1991). An AI-based approach for transit
35–46. https://doi.org/10.1016/j.trb.2014.08.010
route system planning and design. Journal of Advanced Transportation,
Kepaptsoglou, K., & Karlaftis, M. (2009). Transit route network design
25(2), 187–209. https://doi.org/10.1002/atr.5670250205
problem. Journal of Transportation Engineering, 135(8), 491–505.
Barton, R. (2006). Estimation of costs of heavy vehicle use per vehicle-kilo-
https://doi.org/10.1061/(ASCE)0733-947X(2009)135:8(491)
metre in Canada. Transport Canada Economic Analysis Directorate final Kuah, G. K., & Perl, J. (1988). Optimization of feeder bus routes and
report. (File: T8080–05–0326). Logistics Solution Builders Inc. bus stop spacing. Journal of Transportation Engineering, 114(3),
Ceder, A., & Israeli, Y. (1998). User and operator perspectives in transit net- 341–354. https://doi.org/10.1061/(ASCE)0733-947X(1988)114:3(341)
work design. Transportation Research Record: Journal of the Transportation Kuah, G. K., & Perl, J. (1989). The feeder-bus network-design problem.
Research Board, 1623(1), 3–7. https://doi.org/10.3141/1623-01 Journal of the Operational Research Society, 40(8), 751–767. https://
Ceder, A., & Wilson, N. H. (1986). Bus network design. Transportation doi.org/10.1057/jors.1989.127
Research Part B: Methodological, 20(4), 331–344. https://doi.org/10. Kuan, S. N., Ong, H. L., & Ng, K. M. (2004). Applying metaheuristics
1016/0191-2615(86)90047-0 to feeder busnetwork design problem. Asia-Pacific Journal of
 INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 21

Operational Research, 21(04), 543–560. https://doi.org/10.1142/ Shrivastava, P., & Dhingra, S. L. (2006). Operational integration of sub-
S0217595904000382 urban railway and public buses—Case study of Mumbai. Journal of
Kuan, S. N., Ong, H. L., & Ng, K. M. (2006). Solving the feeder bus Transportation Engineering, 132(6), 518–522. https://doi.org/10.
network design problem by genetic algorithms and ant colony opti- 1061//asce/0733-947x/2006/132:6/518
mization. Advances in Engineering Software, 37(6), 351–359. https:// Shrivastava, P., & O’Mahony, M. (2007). Design of feeder route net-
doi.org/10.1016/j.advengsoft.2005.10.003 work using combined genetic algorithm and specialized repair heur-
Kwan, C. M., & Chang, C. S. (2008). Timetable synchronization of istic. Journal of Public Transportation, 10(2), 109–123. https://doi.
mass rapid transit system using multiobjective evolutionary org/10.5038/2375-0901.10.2.7
approach. IEEE Transactions on Systems, Man, and Cybernetics, Part Shrivastava, P., & O’Mahony, M. (2009). Use of a hybrid algorithm for
C (Applications and Reviews), 38(5), 636–648. modeling coordinated feeder bus route network at suburban railway
Lee, Y.-J., & Vuchic, V. R. (2005). Transit network design with variable stations. Journal of Transportation Engineering, 135(1), 1–8. https://
demand. Journal of Transportation Engineering, 131(1), 1–10. doi.org/10.1061//asce/0733-947x/2009/135:1/1
https://doi.org/10.1061/(ASCE)0733-947X(2005)131:1(1) Sonmez, R., & Ontepeli, B. (2009). Predesign cost estimation of urban
Mandl, C. E. (1980). Evaluation and optimization of urban public railway projects with parametric modeling. Journal of Civil
transportation networks. European Journal of Operational Research, Engineering and Management, 15(4), 405–409. https://doi.org/10.
5(6), 396–404. 3846/1392-3730.2009.15.405-409
Mateescu, G. D. (2006). On the application of genetic algorithms to dif- Szeto, W. Y., & Wu, Y. (2011). A simultaneous bus route design and
ferential equations. Romanian Journal of Economic Forecasting, 7(2), frequency setting problem for Tin Shui Wai, Hong Kong. European
5–9. Journal of Operational Research, 209(2), 141–155. https://doi.org/10.
Mauttone, A., & Urquhart, M. E. (2009). A multi-objective metaheuris- 1016/j.ejor.2010.08.020
tic approach for the transit network design problem. Public Thilakaratne, R. S., & Wirasinghe, S. C. (2016). Implementation of Bus
Rapid Transit (BRT) on an optimal segment of a long regular bus
Transport, 1(4), 253–273. https://doi.org/10.1007/s12469-010-0016-7
route. International Journal of Urban Sciences, 20(1), 15–29. https://
Metro, V. (2012). Transit Performance Report (TPR), FY 2012 (July 1,
doi.org/10.1080/12265934.2015.1133317
2011 - June 30, 2012). http://www.valleymetro.org/images/uploads/
Ul Abedin, Z., Busch, F., Wang, D. Z., Rau, A., & Du, B. (2018).
projects/2012_transit_performance_report_opt.pdf
Comparison of public transport network design methodologies using
Montgomery, D. C. (2005). Design and analysis of experiments. Wiley.
solution-quality evaluation. Journal of Transportation Engineering,
Nayeem, M. A., Rahman, M. K., & Rahman, M. S. (2014). Transit net-
Part A: Systems, 144(8), 04018036.
work design by genetic algorithm with elitism. Transportation
Verma, A., & Dhingra, S. L. (2006). Developing integrated schedules
Research Part C: Emerging Technologies, 46, 30–45. https://doi.org/
for urban rail and feeder bus operation. Journal of Urban Planning
10.1016/j.trc.2014.05.002
and Development, 132(3), 138–146. https://doi.org/10.1061/
Nikolic, M., & Teodorovic, D. (2013). Transit network design by bee col-
(ASCE)0733-9488(2006)132:3(138)
ony optimization. Expert Systems with Applications, 40(15), Wirasinghe, S., Chandana, Hurdle, V. F., & Newell, G. F. (1977).
5945–5955. https://doi.org/10.1016/j.eswa.2013.05.002 Optimal parameters for acoordinated rail and bus transit system.
Oguchi, T., Mitsuyasu, A., Oshima, D., & Imagawa, T. (2017). An evalu- Transportation Science, 11(4), 359–374. https://doi.org/10.1287/trsc.
ation study on advanced public transport priority system using traffic 11.4.359
simulation. International Journal of Urban Sciences, 21(sup1), 43–53. Wirasinghe, S. C. (1980). Nearly optimal parameters for a rail/feeder-
https://doi.org/10.1080/12265934.2016.1275750 bus system on a rectangular grid. Transportation Research Part A:
Pradhan, A., & Mahinthakumar, G. (2013). Finding all-pairs shortest General, 14(1), 33–40.
path for a large-scale transportation network using parallel Floyd- Xiong, J., Guan, W., Song, L., Huang, A., & Shao, C. (2013). Optimal
Warshall and parallel Dijkstra algorithms. Journal of Computing in routing design of a community shuttle for metro stations. Journal of
Civil Engineering, 27(3), 263–273. https://doi.org/10.1061/(ASCE)CP. Transportation Engineering, 139(12), 1211–1223. https://doi.org/10.
1943-5487.0000220 1061/(asce)te.1943-5436.0000608
Shih, M.-C., Mahmassani, H. S., & Baaj, M. H. (1997). Trip assignment Yao, B., Hu, P., Lu, X., Gao, J., & Zhang, M. (2014). Transit network
model for timed-transfer transit systems. Transportation Research design based on travel time reliability. Transportation Research Part
Record: Journal of the Transportation Research Board, 1571(1), C: Emerging Technologies, 43, 233–248. https://doi.org/10.1016/j.trc.
24–30. https://doi.org/10.3141/1571-04 2013.12.005
Shrivastav, P., & Dhingra, S. L. (2001). Development of feeder routes Yu, B., & Yang, Z. Z. (2011). An ant colony optimization model: The
for suburban railway stations using heuristic approach. Journal of period vehicle routing problem with time windows. Transportation
Transportation Engineering, 127(4), 334–341. https://doi.org/10.1061/ Research Part E: Logistics and Transportation Review, 47(2),
(ASCE)0733-947X(2001)127:4(334) 166–181. https://doi.org/10.1016/j.tre.2010.09.010

View publication stats