Transportation Research Part C 171 (2025) 104967

Contents lists available at ScienceDirect

Transportation Research Part C

journal homepage: www.elsevier.com/locate/trc

A composite transportation network design problem with land-air

coordinated operations☆
Honggang Zhang a,b , Jinbiao Huo a , Churong Chen a,c , Zhiyuan Liu a,*
a
Jiangsu Key Laboratory of Urban ITS, Jiangsu Province Collaborative Innovation Center of Modern Urban Traffic Technologies, School of
Transportation, Southeast University, Nanjing, China
b
Department of Logistics and Maritime Studies, Faculty of Business, The Hong Kong Polytechnic University, Kowloon, Hong Kong, 999077, China
c
Department of Civil and Environmental Engineering, University of Michigan, Ann Arbor, MI 48109, USA

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

Keywords: With the advent of electric vertical-takeoff-and-landing (eVTOL) vehicles, it becomes imperative
Network planning to assess their impact on transportation network efficiency. Hence, this paper introduces a novel
Urban aerial mobility model for the composite transportation network design problem (CTNDP). Specifically, the model
Vertiport location and capacity
is designed to inform the planning of composite transportation networks by considering travel
behaviors of land-air coordinated mobility. This innovative model is formulated as a bi-level
programming problem, in which the upper-level model focuses on strategically planning verti­
port locations and capacities, thereby reducing the total travel time within the transportation
system. Upon determining decisions at the upper level, the composite transportation network can
be established by constructing new links that interconnect the vertiports and existing road net­
works. The lower-level model addresses the land-air collaboration network equilibrium (LAC-NE)
problem through mathematical programming. To solve the bi-level programming problem, a
customized Bayesian optimization method, namely the mixed-integer Bayesian optimization (MI-
BO) method, is proposed. Specifically, MI-BO is developed by incorporating the branch-and-
bound algorithm into the Bayesian optimization framework. Additionally, an advanced path-
based improved gradient projection (IGP) algorithm is developed to efficiently resolve the LAC-
NE problem. The efficacy of the proposed model and solution algorithms are substantiated
through numerical experiments, and the results illustrate the impact of land-air coordinated
operations on the performance of the composite transportation network.

1. Introduction

Transportation is a vital facilitator of human activities, and with a growing population, the demand for efficient transportation is
escalating. The resulting surge in road traffic contributes to more and more mobility issues like congestion, air pollution, and accidents;
particularly in bustling city centers. Addressing these challenges necessitates strategic transport network planning and traffic control
strategies. These concerns are conventionally tackled within the urban transportation network design problem (UTNDP) (Dantzig et al.
1979; Yang and Wang 2002; Cantarella et al. 2006; Farahani et al. 2013; Gu et al. 2018; Alkaabneh et al. 2019; Jiang and Nielsen 2022;

☆
This article belongs to the Virtual Special Issue on “Transportation Research Part C”.
* Corresponding author.
E-mail address: [email protected] (Z. Liu).

https://doi.org/10.1016/j.trc.2024.104967
Received 8 December 2023; Received in revised form 20 November 2024; Accepted 7 December 2024
Available online 18 December 2024
0968-090X/© 2024 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

Zhang et al. 2023c).

The advent of electric vertical-takeoff-and-landing (eVTOL) vehicles is reshaping transportation systems and introducing novel
mobility options. Different from traditional helicopters, this innovative mode offers distinct advantages, including reduced noise,
heightened safety, cost-effectiveness, and enhanced environmental sustainability. These qualities position eVTOL vehicles as key
players in the future of urban transportation. Notably, both high-tech companies and aviation industries are actively developing
business models utilizing eVTOL vehicles for daily commuting, fostering the concept of urban aerial mobility (UAM). Globally,
countries including China, Singapore, the United States, and Japan, along with organizations such as the European Union Aviation
Safety Agency, Joby Aviation, and EHang, are making substantial investments in UAM infrastructure and technologies. These de­
velopments are critical to the integration of eVTOL vehicles into daily commuting. Given the swift evolution of UAM technologies, it is
expected that UAM will soon become a widely adopted mode of urban transportation systems. With the growth of eVTOL vehicles, a
key concern is understanding the role that the UAM plays in an integrated transportation system and how it can effectively coordinate
with traditional road transportation.
It is worth noting that drones have recently been employed for parcel delivery, effectively shifting urban logistics into the airspace
(Bombelli et al. 2020; Tseremoglou et al. 2022). In light of this evolving landscape, it is increasingly clear that eVTOL vehicles will be
pivotal in passenger transportation, propelling urban aerial mobility into an inevitable future. Notably, due to the unique properties of
eVTOL vehicles, integrating them directly into existing road networks presents challenges, as this could potentially disrupt the regular
traffic flow of conventional vehicles. This highlights the need to address traffic assignment challenges within a three-dimensional
space. A crucial strategy involves establishing several vertiports, serving as virtual nodes strategically positioned to alleviate these
concerns. Building on this, it becomes evident that the transition to UAM necessitates the development of new transportation in­
frastructures designed to accommodate this innovative mode. Among the critical factors, the strategic placement and capacity
configuration of vertiports are crucial, significantly influencing traveler behavior. Moreover, empowering sustainable mobility
through advanced analytics and optimization is crucial in shaping the future of transportation into a green, flexible, and well-
synchronized mobility ecosystem that meets the needs of society. However, note that existing studies on UTNDP primarily focus on
traditional ground-based modes of transport, such as automobiles, subways, and buses–within a two-dimensional framework, often
overlooking the integration of advanced low-altitude (ALT) transportation. To realize a green, efficient, and seamless experience for
travelers, it is essential to optimize transportation planning and operations by incorporating this emerging transportation mode.
Therefore, this paper focuses on addressing the composite transportation network design problem with land-air coordinated opera­
tions, aiming to provide a comprehensive analytical framework for its modeling and solution algorithm.

1.1. Literature review

The UTNDP has been under consistent investigation for the past six decades, with a growing number of publications over time
(Dantzig et al. 1979; Yang and Wang 2002; Mesbah et al. 2008; Scarinci et al. 2017; Pu et al. 2022; Kim et al. 2023). This can be
attributed to the problem’s inherent complexity, theoretical intrigue, and practical significance. The UTNDP is typically formulated as
a bi-level problem. The upper-level problem pertains to the decision-maker, often a government authority, responsible for planning or
managing the transport network. This problem involves practical policy discussions, measurable goals (e.g., reducing total travel
time), and design decisions (e.g., construction of new roads). The assumption is that the decision-maker can predict travelers’ travel
behaviors. In contrast, the lower-level problem is the traveler’s problem, where individuals decide whether to travel and if so, select
travel modes and routes based on decisions made at the upper level. The bi-level structure allows the decision-maker to make
transportation planning or management decisions by considering traveler reactions and improving network performance.
The upper-level problems of UTNDP can be categorized based on various network design policies and decisions. Traditionally, they
are divided into road network design problems and transit network design problems. Among them, the former focuses on street
networks without distinguishing between public transit and private vehicles (Wang et al. 2013; Hosseininasab and Shetab-Boushehri
2015; Yin et al. 2017; Di et al. 2018; Wang et al. 2020a; Mansourianfar et al. 2021; Chai et al. 2023; Kim et al. 2023), while the latter
considers public transit network topology, service frequency, and timetables (Mesbah et al. 2008; Gallo et al. 2011; Szeto and Jiang
2014; Yu et al. 2015; Iliopoulou and Kepaptsoglou 2021; Lee et al. 2022). Notably, these types of problems mentioned above typically
focus on a single mode (e.g., car, metro, or transit) and frequently confine themselves to a single transportation network. In practice,
however, multiple modes coexist, and travel demands within different modes are interconnected. A more fitting classification for this
scenario is the multi-modal network design problem, representing a category that incorporates at least two distinct modes within the
UTNDP framework (Yao et al. 2012; Rashidi et al. 2016; Huang et al. 2018; Pinto et al. 2020; Du et al. 2022).
The lower-level problem takes different forms based on considered choice dimensions and traveler assumptions. Road network
design problems allocate traffic flows to road networks based on predefined transport policies. For transit network design problems,
the lower-level problem focuses on assigning passengers to transit vehicles, taking into account a predetermined flow of transit ve­
hicles resulting from the specific transit design scenario. Furthermore, the lower-level model is generally employed to describe route
choices of travelers, which is known as trip assignment problems, encompassing traffic assignment problems (LeBlanc et al. 1975; Yao
et al. 2019; Zhang et al. 2023b) and transit assignment problems (Wahba 2004; Ryu et al. 2017; Xu et al. 2020). Notably, trip
assignment assumes route choice behavior principles, such as Wardrop’s principle (Wardrop 1952), representing traveler non-
cooperation and selection of the shortest path, resulting in user equilibrium (UE) pattern (Beckmann et al. 1956; Bell and Cassir
2002; Guo et al. 2013; Khani 2019; Ma et al. 2022). The extension includes stochastic user equilibrium (SUE) assignment for perceived
travel time–based route choice (Daganzo and Sheffi 1977; Xie and Liu 2014; Gu et al. 2022). Opposite to the UE assumption, coop­
erative traveler behavior leads to system optimal (SO) assignment (Wie et al. 1995; Chow 2009; Rambha et al. 2018).

2
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

In what follows, we summarize the problems studied and solution methods used in existing literature, as shown in Table 1. It
becomes evident that most UTNDP studies assume fixed demand, with some considering elastic demand. Additionally, a majority focus
on a single objective (denoted by S), while a limited number explore multi-objectives (denoted by M). In terms of traffic assignment
and modes, the UE principle has received considerable attention. Compared with discrete and mixed design decisions, continuous
design decisions attract significant UTNDP research due to the flexibility of variables allowing various modeling approaches and
solution methodologies.
In terms of solution methods, the genetic algorithm (GA) has been used in existing literature to tackle mixed-integer decision
variables within UTNDP. However, the performance of GA is usually not guaranteed. Recently, the surrogate-based optimization
methods have received increasing attention. Surrogate-based optimization fundamentally involves fitting surrogate models—these are
cost-effective approximations of complex objective functions—by strategically evaluating the objective function at selected points,
known as samples. These models help efficiently pinpoint the optimum’s location, significantly reducing the need to compute objective
functions directly (Hong and Zhang 2021). Among the various algorithms for surrogate-based optimization, Bayesian optimization
(BO) algorithms are prominent for their use of Bayesian statistics in constructing surrogate models (Frazier 2018). BO is noted for its
exceptional data efficiency and has been increasingly employed across various transportation issues (Huo et al. 2023; Tay and Osorio
2022; Yin et al. 2022a). This study attempts to develop a BO method for the network design problem. BO is predominantly applied to
solve optimization problems with continuous decision variables. However, discrete decision variables widely exist in practical ap­
plications. To tackle discrete decision variables, previous studies have adapted the ’naïve approach’ for discrete optimization chal­
lenges (Garrido-Merchán and Hernández-Lobato 2020; Luong et al. 2019). Specifically, the BO algorithm is directly applied to discrete
structures. In this approach, new samples that adhere to integer constraints are created in each iteration by first relaxing these con­
straints and then rounding to the nearest integer solution. However, this method has been noted in prior studies to potentially result in
inefficient sampling (Luong et al. 2019). Some limited studies have explored BO methods for mixed-integer decision variables (Baptista
and Poloczek 2018; Cuesta Ramirez et al. 2022). Their focus lies in developing efficient surrogate models to handle mixed-integer
decision variables. The sampling method over a feasible set is rarely discussed. This study attempts to develop a tailored BO
method for mixed-integer decision variables.
On the other hand, UAM is rapidly gaining momentum, aiming to revolutionize urban transportation by shifting from a two-
dimensional to a three-dimensional paradigm. This transformation is largely driven by advancements in eVTOL technology. As
these vehicles continue to evolve, UAM is increasingly seen as a promising solution for alleviating ground traffic congestion (Du et al.
2021; Tripathi et al. 2022; Wang et al. 2023; Wang and Qu 2023; Jin et al. 2024; Shon et al. 2024; Yan et al. 2024). However, despite its
potential, UAM development faces significant challenges, including infrastructure limitations, regulatory barriers, and safety concerns.
Nevertheless, there is growing interest in the planning and operational aspects of UAM, with several studies focusing on path planning
for aerial mobility in urban environments (Tang et al. 2021; Pinto Neto et al. 2022). Notably, surveys indicate strong public interest in
using UAM for airport shuttle services, highlighting its appeal as a favored mode of transport. Some studies have validated the
feasibility of integrating multiple service routes connecting airports using agent-based simulation methods, projecting positive out­
comes for the efficiency, safety, and scalability of eVTOL services (Guo et al. 2022; Brunelli et al. 2023b; Lv et al. 2024). Moreover,
UAM is expected to operate at high densities and on a large scale, which demands carefully designed operational plans to ensure its safe
and efficient implementation. To address this, numerous studies have explored optimization techniques to guide critical decisions in

Table 1
A summary of the studies of UTNDP.
Reference Single/Multiple Traffic Demand Decision Modes Solution method
Objective assignment variables

Dantzig et al. (1979) S SO Fixed Continuous Car Lagrangian multiplier

Leblanc and Boyce (1986) S UE Fixed Continuous Car Bard’s algorithm
Chen and Alfa (1991) S SUE Fixed Discrete Car Branch and Bound
Yang and Wang (2002) M UE Fixed Continuous Car Simulated annealing
Poorzahedy and Abulghasemi S UE Fixed Discrete Car Genetic algorithm
(2005)
Cantarella et al. (2006) S UE Fixed Mixed Car Genetic algorithm
Mesbah et al. (2008) S UE Fixed Discrete Car and transit Enumeration method
Mathew and Sharma (2009) S UE Fixed Continuous Car Genetic algorithm
Long et al. (2010) S SUE Fixed Discrete Car Sensitivity analysis
Gallo et al. (2011) S UE Fixed Continuous Car, transit and Genetic algorithm
metro
Hosseininasab and Shetab- S UE Fixed Discrete Car Genetic algorithm
Boushehri (2015)
Scarinci et al. (2017) S UE Fixed Continuous Car Metaheuristic methods
Li et al. (2019) S UE Fixed Discrete Car Genetic algorithm
Ye et al. 2021 S UE Elastic Continuous Car, metro, and Successive linear
bike programming method
Wang et al. (2022b) S UE Fixed Continuous Car, transit, and Kriging-surrogate-based
bike algorithm
Luan et al. (2023) S UE Fixed Continuous Car Branch and Bound
This paper S UE Fixed Mixed Car and eVTOL Mixed-integer Bayesian
optimization

3
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

UAM operations, taking into account factors such as energy availability and weather conditions, which significantly impact flight
safety (Reiche et al. 2021; Schweiger et al. 2023; Lee et al. 2024).
In addition to vehicle technology and operations, vertiports play a pivotal role in UAM by providing essential infrastructure for
take-off, landing, and maintenance. The strategic selection of vertiport locations is crucial for optimizing travel routes, improving
service quality, and minimizing urban disruption. Several studies have explored vertiport location optimization at both the network
and system levels. For instance, Wu and Zhang (2021) examined the network design problem of UAM on-demand services, focusing on
integer programming to determine optimal vertiport locations. Wang et al. (2022a) investigated vertiport placement while considering
the interdependencies between strategic deployment, tactical operations, and passenger demand. Macias et al. (2023) developed a
linear placement model with vehicle sizing constraints to identify optimal vertiport configurations based on eVTOL performance.
Additionally, Brunelli et al. (2023a) conducted a systematic review of vertiport characteristics, identifying key features that influence
location and capacity while highlighting research gaps in the field.
Despite these advancements, there remains a need for a more integrated approach to vertiport planning that better examines the
relationships between infrastructure, route choice, and travel demand distribution. In this context, our paper focuses on macroscopic
aspects, particularly the integration of low-altitude flights with traditional road networks and their impact on commuters’ route
choices (i.e., the distribution of travel demand). We further investigate how the introduction of UAM influences the overall perfor­
mance of the transportation system, evaluating its potential to enhance efficiency, reduce congestion, and reshape urban mobility.

1.2. Objectives and contributions

In consideration of the aforementioned points, we contend that the majority of literature addressing the UTNDP primarily con­
centrates on ground-based transportation modes (e.g., car, transit, and metro). Within this context, numerous efforts have been un­
dertaken to plan an urban transportation network, encompassing strategies like toll setting, parking pricing, and the expansion of
existing streets or the construction of new ones. However, crucial distinctions between advanced low-altitude transportation and
conventional road transportation underscore the limited applicability of management policies outlined in previous studies. Addi­
tionally, in urban environments, the competition for limited land space with other essential facets of city life significantly contributes
to persistent and challenging traffic congestion. Thus, the transition from roads to the air assumes particular significance, given the
heavy reliance of traditional urban transportation on road infrastructure.
In light of these considerations, we argue that there is an urgent need for the exploration of novel planning and management
policies to facilitate seamless land-air coordinated operations, which became especially crucial before the widespread adoption of
eVTOL vehicles. In specific, the primary contributions of this paper are threefold. Firstly, we introduce a novel bi-level programming
model to formulate the composite transportation network design. In this model, the authority seeks to reduce the total travel time by
designing the vertiports’ location and capacity (i.e., the upper-level problem). Secondly, based on decisions derived from the authority,
the land-air collaboration composite transportation network is established, and commuter route choices within this composite network
are characterized following the classical user equilibrium theory, thereby developing a land-air collaboration network equilibrium
model. Thirdly, we introduce a tailored algorithm to address the proposed model. Specifically, we have developed a customized mixed-
integer Bayesian optimization (MI-BO) method that integrates a modified Gaussian process model and the branch-and-bound tech­
nique into the BO framework. This MI-BO approach optimizes the bi-level programming model by strategically selecting feasible
solutions, with each solution being assessed through the lower-level problem using the improved gradient projection (IGP) algorithm.
Finally, numerical experiments are conducted to validate the applicability of the proposed model, with results offering decision-
making support for government authorities.
The subsequent sections of this paper are organized as follows: Section 2 outlines the characterization of a composite transportation
network integrating urban aerial mobility. Section 3 delves into the specific network design problem, presenting a bi-level model
formulation for planning the land-air coordinated operations. Section 4 provides a customized solution algorithm framework for the
proposed model. Section 5 presents numerical examples to demonstrate the effectiveness of the proposed model. Finally, Section 6
concludes this study.

2. The characterization of the land-air collaboration composite network

Given the above, electric vertical-takeoff-and-landing (eVTOL) vehicles are catalyzing the transformation of urban aerial mobility
(UAM), offering a more convenient, efficient, and sustainable option for urban commuting. Notably, eVTOL pertains to a unique land-
air amphibious vehicle, capable of ground travel akin to conventional vehicles, with the added capability of transitioning to advanced
low-altitude flight by accessing the vertiports. Before delving into the specifics of the model, it is essential to provide a summary of the
key notations employed in this paper, which are outlined as follows:

Sets:
NR The set of nodes of the road network, and ∀n ∈ NR ;

AR The set of links of the road network, and ∀a ∈ AR ;

AQ The set of incoming links of the vertiport (termed queue links), and ∀a ∈ AQ ;
AE The set of outgoing links of the vertiport (termed evacuation links), and ∀̃
a ∈ AE ;
AC The set of links connecting any two vertiports (termed flight links), and ∀̂
a ∈ AC ;
(continued on next page)

4
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

(continued )
Sets:
NR The set of nodes of the road network, and ∀n ∈ NR ;

W The set of OD pairs, and ∀w ∈ W;

Kw The set of paths between OD pairs w in the land-air collaboration composite network;
P The set of candidate vertiports, and ∀p ∈ P.
Parameters:
Vh The horizontal cruising speed of the eVTOL vehicles;
Vv The average speed of the vertical take-off and landing of the eVTOL vehicles;
Vf The free-flow speed on both the road links and the evacuation links;
Lã The length of the evacuation links, and ∀̃a ∈ AE ;
Lâ The length of the flight links, and ∀̂
a ∈ AC ;
hlp The flight altitude between any two vertiports, and ∀l, p ∈ P;
qw The total demand for OD pair w, and ∀w ∈ W;
d The safe spacing of the eVTOL vehicles during flight;
B The maximum number of vertiports allowed for construction;
C The maximum capacity of the constructed vertiports.
Intermediate variables:
va The traffic flows on the road link a, and ∀a ∈ AR ;
va The traffic flows on the queue link a, and ∀a ∈ AQ ;
vã The traffic flows on the evacuation link ̃ a, and ∀̃
a ∈ AE ;
vâ The traffic flows on the flight link ̂
a , and ∀̂
a ∈ AC ;
ta The travel time on the road link a, and ∀a ∈ AR ;
ta The travel time on the queue link a, and ∀a ∈ AQ ;
tã The travel time on the evacuation link ̃ a, and ∀̃
a ∈ AE ;
tâ The travel time on the flight link ̂a , and ∀̂
a ∈ AC .
Decision variables:
ca The capacity of the queue links, and ∀a ∈ AQ ;
zp The binary variables equal 1 when the vertiport is established at the node p, and ∀p ∈ P; otherwise 0;
Xnp The binary variables equal 1 when the road node n ∈ NR is connected to the vertiport p ∈ P; otherwise 0;
Ylp The binary variables equal 1 when the vertiport l ∈ P is connected to the vertiport p ∈ P; otherwise 0;
fkw The traffic flow on the route k ∈ Kw between OD pair w ∈ W.

To enhance a more comprehensive understanding, we present a simplified composite transportation network to illustrate the land-air
coordinated operations, as depicted in Fig. 1. Herein, l and p denote the vertiports, serving as the locations where land-air amphibious
vehicles execute vertical take-off and landing operations.
Based on the preceding analysis, we posit that the determination of travel time on road links can be modeled using the classical
Bureau of Public Road (BPR) function. However, when commuters choose the land-air collaboration travel scheme, as illustrated in
Fig. 2, the whole process unfolds in three stages: (i) eVTOL vehicles queue to enter vertiport, execute vertical take-off operation, and
reach a safe altitude; (ii) eVTOL vehicles fly to another vertiport near the destination through flight links; (iii) eVTOL vehicles land on
the vertiport and then enter road networks via evacuation links. Note that we chose to illustrate a one-way UAM system to demonstrate
its integration with traditional road networks, offering an intuitive understanding of UAM operations. However, the subsequent nu­
merical experiments consider a multi-way UAM system, analyzing how the number of vertiports affects total travel time and traffic
network performance.
During the first stage, considering vertiport capacity constraints, eVTOL vehicles need to queue to enter vertiports and execute
vertical take-off. The travel time for queue links can be characterized based on the classical point queuing theory. Specifically,
assuming arrival rates follow a Poisson distribution and service rates for vertical take-off points follow an exponential distribution, the
process preceding eVTOL vehicles reaching a safe altitude is akin to an M/M/1/C queuing problem. Consequently, the travel time for
queue links is determined by the average stay time in the queuing system, and it can be formulated as (Cho et al. 2020; Liu et al. 2022;
Cheng et al. 2022; Zhou et al. 2022; Zhang et al. 2024):

Fig. 1. Diagram of the land-air collaboration composite transportation network.

5
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

Fig. 2. Schematic diagram of the flight process for eVTOL vehicles.

Lqa
ta = ( ), ∀a ∈ AQ (1)
μ 1 − P0a

where μ represents the average service rate. Additionally, Lqa and P0a indicate the average queue length and the probability of the initial
state, respectively, defined as follows:
1 − ρa
P0a = , = 1, ∀a ∈ AQ
ρa ∕ (2)
1 − (ρa )ca +1

ρa (ca + 1)(ρa )ca +1

Lqa = − , ∀a ∈ AQ (3)
1 − ρa 1 − (ρa )ca +1

where ρa denotes the ratio of the average arrival rate to the average service rate for the specific queue link, and ρa = λa /μ. Herein, λa
denotes the average arrival rate of commuters, which can be determined based on traffic flows corresponding to queue links, namely,
∑∑
λa = fkw χ wa,k , ∀a ∈ AQ (4)
w∈Wk∈Kw

where χ wa,k equals 1 when the path k uses the queue link a between the OD pair w, and 0 otherwise. Given the average speed of eVTOL
vehicles during vertical take-off and landing phases, the average service rate of each vertiport is defined as:
Vv
μ= (5)
d

where d denotes the safe spacing of the eVTOL vehicles during the flight phase, indicating that when an eVTOL reaches a given safe
altitude, the next eVTOL vehicle can be ready for vertical take-off.
In contrast to the first stage, the characterization of travel time is relatively straightforward for both the second and third stages. To
ensure the safety and efficiency of eVTOL vehicle operations throughout the flight phase, such as avoiding collisions and hovering over
the vertical landing point, it is assumed that eVTOL vehicles maintain a certain cruising speed during the horizontal flight stage, and
there is no airspace conflict among various routes. The travel time for flight links comprises two parts: (i) time taken by eVTOL vehicles
for vertical take-off and landing (i.e., the first term in Eq. (6)); (ii) time taken by the eVTOL vehicles to fly horizontally (i.e., the second
term in Eq. (6)). Thus, the travel time corresponding to flight links is defined as follows:
2hlp − d Lâ
tâ = + , a ∈ AC , l, p ∈ P
∀̂ (6)
Vv Vh
In this setting, eVTOL vehicles shall sequentially land at the vertiport, effectively avoiding congestion on evacuation links. For
simplicity, we assume that eVTOL vehicles travel at free-flow speeds on evacuation links, and hence the corresponding travel time for
these links can be calculated as follows:
Lã
tã = , a ∈ AE
∀̃ (7)
Vf

6
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

3. The network planning of land-air coordinated operation

In this section, we elucidate the composite transportation network design problem featuring land-air coordinated operations,
followed by the development of an innovative bi-level programming model that integrates the UAM concept.

3.1. The problem description

Building on the insights from Section 2, the characterization of impedance functions for diverse link types (e.g., road links, queue
links, flight links, and evacuation links) varies within the land-air collaboration composite transportation network. This approach
enables the developed lower-level model not only to simulate traffic flow distribution but also to accurately describe the state of the
traffic system under a land-air cooperative operation scenario—a capability that is lacking in traditional traffic assignment models.
Furthermore, it is crucial to underscore that both the vertiport locations and the capacities of the queue system play pivotal roles in
facilitating the land-air coordinated operations. These factors have a significant impact on the travel time of eVTOL vehicles.
Consequently, to fully leverage the advantages of this emerging transportation mode and effectively mitigate urban traffic congestion,
there is an urgent need to design the land-air collaboration composite transportation network.
Specifically, we formulate a composite transportation network design problem with land-air coordinated operations as a bi-level
programming model (as detailed in Subsection 3.2). This model can be construed as exhibiting the structure of a Stackelberg game
between the system policymaker and the network users (Yang and Bell 1998; Yin et al. 2022b). Herein, the objective of the authority (i.
e., the upper-level problem) aims to minimize the total travel time (TTT) by determining the vertiport locations and capacities, and the
commuters (i.e., the lower-level problem) aim to regulate the flow pattern in the land-air collaboration composite transportation
system. The lower-level problem takes the form of a novel land-air collaboration network equilibrium (LAC-NE) model, presented as a
mathematical programming problem. In developing this lower-level traffic assignment model, we accounted for the transfer of traffic
congestion between ground and low-altitude areas. This consideration is primarily reflected in the characterization of the impedance
function for different types of links, which is a key factor influencing the distribution of travel demand.
In general, the upper-level model, representing the authority, is responsible for determining the optimal locations and capacities of
vertiports, while the lower-level model, representing the commuters, regulates the distribution of travel demand within the network
established by the authority. The upper-level decision on vertiport locations and vertiport capacity shapes the overall topology of the
composite network, directly influencing the mode choices and route choices of commuters, as different topologies provide varying sets
of feasible paths. Once the upper-level authority determines the vertiport locations and capacities, the lower-level commuters choose
their travel paths (i.e., distribute travel demand) based on user equilibrium conditions within the given composite network. The results
are then fed back to the upper-level decision-makers, who adjust the vertiport locations and capacities to minimize the total travel time
of the transportation system. Commuters subsequently update their travel paths according to the revised network topology, and this
iterative process continues until equilibrium is achieved. The bi-level model for planning the land-air collaboration composite network
is illustrated in Fig. 3.

3.2. The bi-level model formulation

Given the abovementioned analysis, this section presents the mathematical formulation of the land-air collaboration composite
transportation network design problem (LAC-CTNDP), detailed as follows:
[Upper Level].
∑ ∑ ∑ ∑
minf(c, z) = va ta + va ta + vâ tâ + vã tã (8)
a∈AR a∈AQ a∈AC a∈AE

subject to.

Fig. 3. The bi-level model for planning the land-air collaboration composite transportation network.

7
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

∑
zp ⩽B (9)
p∈P

ca ⩽C, ∀a ∈ AQ (10)

zp ∈ {0, 1}, ∀p ∈ P (11)

ca > 0, ∀a ∈ AQ (12)
[Lower Level].
∑∫ va ∑∫ va ∑ ∑
min Z(f) = ta (w)dw + ta (w)dw + vâ tâ + vã tã (13)
a∈AR 0 a∈AQ 0 â∈AC ã∈AE

subject to.
∑
Xnp = zp , ∀p ∈ P (14)
n∈NR

Ypl = zp zl , ∀p, l ∈ P : p ∕
=l (15)
∑∑
va = fkw χ wa,k ∀a ∈ AR (16)
w∈Wk∈Kw

∑∑ ∑∑
va = fkw χ wa,k Xnp , ∀a ∈ AQ (17)
w∈Wk∈Kw n∈NR p∈P

∑∑ ∑∑
vã = fkw χ wã,k Xnp , a ∈ AE
∀̃ (18)
w∈Wk∈Kw n∈NR p∈P

∑ ∑ ∑∑
vâ = fkw χ wâ,k Ypl , a ∈ AC
∀̂ (19)
w∈Wk∈Kw l∈P p∈P

∑
fkw = qw , ∀w ∈ W (20)
k∈Kw

fkw ⩾0, ∀w ∈ W, k ∈ Kw (21)

Xnp ∈ {0, 1}, ∀n ∈ NR , p ∈ P (22)

Ypl ∈ {0, 1}, ∀p, l ∈ P (23)

In the upper level, the objective function (8) minimizes the total travel time of the transportation system. Constraint (9) indicates
the upper bound on vertiports planned by the authority. Constraints (10) represent the maximum capacity constraints applicable to the
queue links (or vertiports). The binary domain of the vertiport location variables is defined in (11). Constraints (12) ensure that the
capacity of the queue link remains non-negative.
In the lower level, the objective function (13) is dedicated to minimizing individual travel time within the composite transportation
network. Constraints (14) ensure that at least one road node is connected to the constructed vertiports, enabling eVTOL vehicles to
enter the vertiports and execute flight operations. Furthermore, to ensure the return of eVTOL vehicles from the vertiports to the road
networks, it is crucial to establish a two-way link between the road node and the vertiports. Constraints (15) further guarantee the
interoperability of any two constructed vertiports, ensuring that eVTOL vehicles can seamlessly complete low-altitude flights between
them. Constraints (16)–(19) define the incidence relationships between path flows and various types of link flows. Constraints (20)
ensure flow conservation for travel demand, and the non-negative constraints (21) are imperative for maintaining the physical
meaningfulness of the program’s solution. Finally, the domains of the binary variables governing the connection between road nodes
and vertiports, as well as the connection between vertiports, are defined in (22) and (23).
Challenges for solving the bi-level programming problem mainly lie in the implicitly defined objective function. Specifically, the
relationship between the objective function f( ⋅ ) and the decision variables c and z are embedded in the lower-level traffic assignment
problem, in which each commuter makes route choices considering the decisions of all other commuters. Consequently, the objective
function is highly nonlinear, non-differentiable and intractable. Conventional optimization techniques (e.g., gradient-based methods)
are challenging to solve the problem, especially in networks with practical sizes. To tackle this issue, this study proposes a customized
Bayesian optimization to inform the design of land-air collaboration composite networks.

8
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

4. Solution algorithm

BO is widely used to solve expensive-to-evaluate black-box optimization problems (Frazier 2018). It consists of two iterative steps.
First, it constructs a surrogate model using sample points. Second, it suggests and evaluates a new sample for the enrichment of the
sample points and the refinement of the surrogate model. These two steps are typically executed by constructing Gaussian process
models and optimizing acquisition functions, respectively. This study aims to develop a customized BO method to inform the design of
land-air collaboration composite networks. The significant challenge lies in how to handle mixed-integer decision variables. Specif­
ically, two problems need to be addressed. First, the surrogate model should be able to approximate the objective function with mixed-
integer inputs. A key component within the Gaussian process model is the kernel function, which measures similarity and covariance
between solutions. However, the kernel functions are typically defined for continuous decision variables. The problem lies in how to
measure similarity and covariance among solutions when discrete decision variables are included. Second, the acquisition function
should be optimized considering integer constraints. Existing literature has shown that the rounding method (i.e., first generating
solutions by relaxing all constraints and then rounding to the nearest feasible solution) is inefficient. We aim to explore effective ways
to tackle integer constraints. To tackle these problems, a modified Gaussian process model is developed to deal with the mixed-integer
decision variables. The modified Gaussian process model is developed by modifying the distance measure in kernel calculation. In
addition, regarding the integer restrictions, the branch-and-bound method is incorporated with the commonly used acquisition
functions to sample feasible solutions. In BO algorithms, each solution is assessed by solving the lower-level traffic assignment
problem. The evaluation can be computationally expensive especially when the networks are of practical scales. To facilitate the
evaluation, we incorporate an IGP method into the solution algorithm to efficiently resolve the lower-level LAC-NE problem.
A flowchart summarizing the solution algorithm is depicted in Fig. 4. As illustrated, the MI-BO process unfolds iteratively. Initially,
a set of feasible solutions is selected and evaluated at random, creating a dataset that serves as the input for the Gaussian process model
tailored for mixed-integer decision variables. This model is used to develop the integer-restricted Expected Improvement (EI) function,
a widely adopted acquisition function. A new sample is then generated by optimizing this integer-restricted EI function through the
branch-and-bound method. Subsequently, the objective function value of the new sample is assessed. The Improved Gradient Pro­
jection (IGP) method is utilized to resolve the lower-level problem. This iterative process continues until the predetermined stopping
criteria, which are detailed in Section 4.2, are satisfied.

4.1. Gaussian process for mixed-integer decision variables

Let xi = (zi , ci ) denote a realization of the decision variable, which is also referred to as a sample point. Suppose we have m (m⩾1)
sample points, denoted as Xm = {x1 , ..., xm }. Given xi ∈ Xm , the associated total travel time of the LAC composite transportation
network can be obtained by solving the lower-level traffic assignment problem. In this study, we adopted the improved gradient
projection (IGP) method to solve the lower-level problem (Xie et al. 2018). Let fi denote the objective function value associated with
( )
variables xi . Let Fm = f1 , ..., fm denote the vector of the objective function values. The dataset is defined as Dm = (Xm , Fm ). The
Gaussian process surrogate model is formulated based on Dm .
A Gaussian process is a set of random variables, any finite number of which have a joint Gaussian distribution. Thus, a Gaussian
process f(x) is completely specified by two components: the mean function g(x) and the covariance function κ(x, xʹ). The two com­
ponents are defined as:

Fig. 4. A flow chart of the MI-BO algorithm.

9
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

g(x) = EEf(x)F (24)

κ(x, xʹ) = EE(f(x) − f(x))(f(xʹ) − f(xʹ))F (25)

The Gaussian process is written as

f(x) ∼ G P (g(x), κ(x, xʹ)) (26)

While constructing Gaussian process surrogate models for f( ⋅ ), a Gaussian process prior is initially applied to f( ⋅ ), which es­
tablishes a joint Gaussian prior distribution. The surrogate model is then derived by conditioning this joint prior distribution on
observations at sample points. In the Gaussian process prior, g(x) is typically set to zero. The covariance function is performed using
kernel functions. One widely adopted kernel function is the radial basis function (RBF).
( 2/ )
κ(x, xʹ) = σ 2 exp − ‖x − xʹ‖2 2η2 (27)

where σ and η are parameters of the RBF, which can be tuned using maximum likelihood estimation. Eq. (27) explicitly assumes that
nearby inputs are highly correlated with outputs. However, the assumption can be violated when dealing with mixed-integer decision
variables. In such cases, the covariance or similarity between objective function values cannot be properly captured by the Euclidean
distance. To tackle the problem, this study modified the distance measure in kernel calculation. For sample points x = (z, c) and xʹ =
(zʹ,cʹ), we propose to use the Hamming distance to measure the similarity in discrete spaces, and the Euclidean distance for continuous
spaces.
( / )
dz
∑ ( ( ))
κz (z, zʹ) = σ2z exp − 1 − δ zi , zʹi 2η2z (28)
i=1

( 2/ )
κc (c, cʹ) = σ2c exp − ‖c − cʹ‖2 2η2c (29)
( )
where δ( ⋅ ) is the Kronecker delta function. δ zi , zʹi equals one if zi = zʹi and zero otherwise. dz is the dimension of the decision vector z.
σz (σc ) and ηz (ηc ) are parameters of the kernel functions. Given Eqs. (28) and (29), the kernel function for mixed-integer decision
variables is formulated as follows:

κMI (x, xʹ) = κz (z, zʹ)κc (c, cʹ)

( / )
dz
∑ ( ( )) 2/ (30)
= σ2MI exp − 1 − δ zi , zʹi 2η2z − ‖c − cʹ‖2 2η2c
i=1

where σMI = σc σz . The parameters σ MI , ηz , and ηc can be tuned using maximum likelihood estimation.
Suppose the function f( ⋅ ) takes the abovementioned Gaussian process prior (i.e., g(x) = 0, and the covariance function takes κMI (x,
xʹ)), the joint prior distribution of f(x1 ), ..., f(xm ) and f(xʹ) is represented in (31), where xʹ is the decision vector without evaluation, and
we are interested in the associated f(xʹ).
[ ] ( [ ])
F m Km K*
∼ N 0, (31)
f(x )
ʹ
KT* K**
( )
where Fm = (f(x1 ), ..., f(xm )), K* = (κMI (x1 , xʹ), ..., κMI (xn , xʹ)), K** = κMI (xʹ,xʹ), and Km is an m × m matrix with entries κMI xi ,xj , i,j ∈
{1, ..., m}.
Given the observation fi of f(xi ), i ∈ {1, ..., m}. The predictive distribution of f(xʹ) can be derived by conditioning the prior dis­
tribution (the distribution (31)) on observations of f(x1 ), ..., f(xm ), which yields
( )
f(xʹ)|xʹ, Dm ∼ N μf (xʹ), σ2f (xʹ) (32)

where μf (xʹ) = KT* K−m1 Fm , σ2f (xʹ) = K** − KT* K−m1 K* .

4.2. Branch-and-bound method to solve integer-restricted EI functions

Given the surrogate model of f( ⋅ ), this section discusses the construction and optimization of acquisition functions, based on which
new samples can be selected regarding the integer constraints. In this study, the widely used expected improvement (EI) function
serves as the acquisition function (Zhan and Xing 2020).

4.2.1. Integer-restricted EI function

{ }
Let f ∗ = min f1 ,...,fm . The fundamental concept of EI is to sample a new solution x to maximize the expected improvement over
the current best sample (Kleijnen et al. 2012), that is,

10
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

maxEEmax(f ∗ − f(x), 0)F (33)

x

Given the predictive distribution (32), the optimization problem (33) is equivalent to.
( ) (f ∗ − μ (x)) ( ∗
f − μf (x)
)
f
max f ∗ − μf (x) Φ + σf (x)ϕ (34)
x σf (x) σ f (x)

where Φ( ⋅ ) and ϕ( ⋅ ) are the cumulative distribution function (CDF) and probability density function (PDF) of the standard normal
distribution, respectively.
While using EI functions to suggest new samples for the land-air collaboration composite transportation networks, constraints of
the original network design problem should be incorporated, such that feasible solutions can be sampled and evaluated. Therefore, an
integer-restricted EI function is formulated as follows:

4.2.2. A branch-and-bound method

( ) (f ∗ − μ (x)) ( ∗
f − μf (x)
)
f
max f ∗ − μf (x) Φ + σf (x)ϕ (35)
x σf (x) σ f (x)
subject to.
cl ⩽c⩽cu (36)

zi ∈ {0, 1}, i = 1, ..., P (37)

The integer-restricted EI function is solved in the branch-and-bound framework. The branch-and-bound (B&B) method employs a
tree diagram, consisting of nodes and branches, to structure the partitioning of solutions. Specifically, the algorithm begins by solving a
relaxed EI function (all binary constraints are relaxed). This results in the first node, which contains the relaxed solution. The feasibility
and objective function value of the relaxed solution are evaluated to determine: (i) the upper bound of the node, (ii) the lower bound of
the problem, and (ii) the further branching of the problem. Subsequently, for each branch of the problem, a relaxed EI function—with
certain specific relaxed binary constraints—is solved, yielding a relaxed solution. This process continues until a feasible solution is
reached, which possesses the highest upper bound value among all terminal nodes. Note that the EI function is highly nonlinear and
nonconvex. A challenge of adopting the B&B algorithm is how to derive the upper bound of the EI function. This study adopts the
method proposed by Jones et al. (1998) to estimate upper bounds. The B&B algorithm for solving the integer-restricted EI function is
summarized in Algorithm 1.

Algorithm 1. Branch-and-bound algorithm to solve the integer-restricted EI function

Input: The Gaussian process surrogate model of f( ⋅ ); the minimal objective function value f ∗ of all samples.
Output: The new sample xm+1 .
1: Let X denote the search space derived by relaxing all binary constraints*. Initialize L = {X }, fL* = − ∞.
2: While L ∕
= ∅ do
3: Select a search space S from L;
4: Optimize the EI function (35) in S. Let x∗S denote the optimal solution.
5: If x∗S is a feasible solution do
( )
6: If f x∗S > fL* do
( )
7: Update x∗ = x∗S , fL* = f x∗S .

8: ​ ​ End if
9: ​ Else
( )
10: ​ ​ If f x∗S > fL* do
11: ​ ​ ​ Select a variable xi for branching**.
12: ​ ​ ​ Partition S into S1 = {x|x ∈ S, xi = 1}, and S2 = {x|x ∈ S, xi = 0}. Let L = L ∪ {S1, S2}.
13: ​ ​ End if
14: ​ End if
15: ​ L = L\{S}
16: End while
17: Return x∗ .

Notes: *The binary constraints are replaced by 0⩽pi ⩽1, i = 1, ..., P. **This study randomly selects a binary variable, which has a
fractional part in x∗S , as the branching variable.
Subsequently, the MI-BO algorithm is applied to solve the proposed model, and the comprehensive procedure is summarized in
Algorithm 2.

Algorithm 2 Mixed-integer Bayesian optimization algorithm

Input: The objective function f( ⋅ ); the number m of initial sample points.
Output: A satisfactory solution x*
(continued on next page)

11
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

(continued )
Algorithm 2 Mixed-integer Bayesian optimization algorithm
Input: The objective function f( ⋅ ); the number m of initial sample points.
Output: A satisfactory solution x*
1: Randomly generate m initial sample points, Xm = {x1 , ..., xm }. Evaluate the samples by solving the lower-level problem using the IGP method. Denote the
( )
objective function values as Fm = f1 , ..., fm .
2: While the stopping criteria are not met do
3: Construct the Gaussian process surrogate model of f( ⋅ ) according to (32).

1: Randomly generate m initial sample points, Xm = {x1 , ..., xm }. Evaluate the samples by solving the lower-level problem using the IGP method. Denote the
( )
objective function values as Fm = f1 , ..., fm .
2: While the stopping criteria are not met do
3: Construct the Gaussian process surrogate model of f( ⋅ ) according to (32).

4: ​ Use Algorithm 1 to solve P1 and obtain a new solution xm+1 .

5: ​ Based on the newly determined solution xm+1 , use a path-based IGP algorithm to solve the lower-level problem (i.e., LAC-NE model).
6: ​ Solve the lower-level problem and obtain the objective function value fm+1 = f(xm+1 ) associated with xm+1 .
( )
7: ​ Update Xm+1 = Xm ∪ {xm+1 }, Fm+1 = f1 , ..., fm , fm+1 , m = m + 1.
8: End while
9: Let * = argmini=1,...,m fi . Return x*

Remark 1: To facilitate the construction of a composite network, upon establishing the decisions made by the upper-level model (i.e.,
vertiport locations and capacities), each vertiport is connected to the node in the road network that is nearest to it. Furthermore,
connections are established between any two vertiports. This process transforms the lower-level problem into a unique land-air
collaboration network equilibrium (LAC-NE) problem. It is crucial to note that the LAC-NE problem differs from the traditional
user equilibrium (UE) problem due to variations in the impedance function characterizations of various links. Consequently, the al-
gorithm conventionally used to solve the UE problem can be adapted to address the LAC-NE model. In this context, we opt for the
advanced path-based IGP algorithm (Xie et al. 2018) to effectively tackle the LAC-NE model.
Remark 2: There is randomness in Algorithm 2, which stems from the generating of initial samples. There are two ways to mitigate
this randomness. The first approach involves i) running multiple replications of the algorithm to generate various network design
schemes and ii) evaluating each one to select the scheme with the best performance. This method is commonly used in practical

Fig. 5. The road transportation network.

12
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

applications. The second is to increase the computational budget by selecting a larger number of samples. As the number of samples
increases, the randomness in results decreases.

5. Numerical examples

In this section, we provide several numerical examples to validate our proposed models and methods. In Section 5.1, experiments
are carried out on the Sioux Falls network, demonstrating that the land-air collaborative network design model can effectively alleviate
congestion. Section 5.2 features experiments on the large-scale Anaheim network. The objective here is to showcase the effectiveness
of the proposed methods within real-world networks and to confirm the efficiency of the MI-BO. Details about the experimental
networks are accessible at the website: https://github.com/bstabler/TransportationNetworks/tree/master. All experiments are per-
formed on a Windows 10 64-bit computer equipped with an AMD 4800H 2.9 GHz CPU and 16G RAM. The proposed algorithm is
implemented in C++ using Visual Studio 2022.

5.1. Experiments on the Sioux-Falls network

The first experiment is built from the Sioux Falls network. The tested road network, as depicted in Fig. 5, consists of 24 nodes and 76
links, and there are a total of 576 OD pairs. The link lengths are determined based on the geographical coordinates (longitude and
latitude) of the respective nodes. Other parameters incorporated into the optimization model are based on reasonable values from
existing studies. For instance, the free-flow speed on both the road links and the evacuation links is set at 60 km/h (Wang et al. 2020b).
The horizontal cruising speed for land-air amphibious vehicles is established at 300 km/h, while the average speed for vertical take-off
and landing is set to 50 km/h (Bacchini et al. 2021; Li and Liu 2022). The safe spacing for land-air amphibious vehicles during the flight
phase is maintained at 10 m. Notably, we assume that there are eight alternative vertiports available for the authority to choose.
Furthermore, the solution precision of the low-level problem (i.e., the LAC-NE model) is evaluated using the relative gap (RG) (Xie
et al. 2018; Zhang et al. 2023a), with the IGP algorithm concluding when the RG is below 10-12. The MI-BO algorithm is employed to
solve the bi-level programming model. The stopping criterion for MI-BO is defined as reaching a predetermined computational budget,
which in this case corresponds to 75 iterations.
As detailed in Section 3, we primarily focus on addressing the composite transportation network design problem with land-air
coordinated operations, where the central challenge is determining the optimal number of vertiports and their capacity limits. The
significance of these two parameters lies in two key aspects: (i) the number and location of vertiports shape the topology of the land-air

Fig. 6. Convergence performance of the MI-BO algorithm.

13
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

cooperative network, which directly influences the set of alternative routes available to travelers; (ii) the capacity of vertiports impacts
the impedance value of low-altitude travel, influencing travelers’ mode choices, which in turn affects the distribution of travel demand
and the total travel time of the transportation system. Based on this analysis, we conducted a comprehensive sensitivity analysis of
these two key parameters.
Fig. 6 evaluates, for different numbers of vertiports, the performances of the proposed method as a function of the iteration number.
The horizontal axis displays the iteration number, and the vertical axis denotes the objective function value (i.e., total travel time) of
the best point identified so far by the method. In all four cases, the proposed method could identify satisfactory solutions within the
first few iterations. This demonstrates the efficiency of the proposed method. In addition, Fig. 6 indicates that with the increase in the
number of vertiports, there is an observed decline in the total travel time. This highlights the necessity of constructing LAC composite
networks.
Subsequently, Table 2 provides a comprehensive overview of network performance, comparing scenarios with and without the
integration of UAM. It becomes evident from this table that the inclusion of UAM in the design of the composite transportation network
leads to a significant reduction in total travel time. Notably, the optimal capacity of vertiports exhibits variability, as both vertiport
capacity and location are determined by the regulatory authorities. Moreover, with an increase in the number of established vertiports,
there is a proportional increase in the sharing rate of land-air collaboration, leading to an additional decrease in the total travel time.
To illustrate, the strategic construction of five vertiports to facilitate the LAC operations yields a noteworthy 19.1 % reduction in the
total travel time of the transportation system when compared to the scenario without considering the concept of UAM. In simple terms,
when government authorities propose alternative plans for vertiport construction, the model developed in this paper can be used to
determine the optimal locations and capacities of the vertiports. Additionally, based on the transportation system’s operational ef-
ficiency and various cities’ development plans for the low-altitude economy, the land-air collaborative network design model can help
government departments identify the most suitable number of vertiports, while also analyzing the impact of UAM on travel demand
distribution and traffic congestion from a macro perspective. This would lead to more informed and effective planning and design of
the land-air cooperative transportation network.
In what follows, we focus on investigating the distribution of the ratio of link flow to link capacity (i.e., V/C) across diverse network
designs, as shown in Fig. 7. Insightful, the advent of eVTOL vehicles offers a potential avenue for alleviating urban congestion issues
through the strategic development of a land-air collaboration composite transportation network.
Notably, compared to the conventional urban transportation network that neglects consideration for the UAM, effectively deter-
mining the location and capacity of vertiport can yield a substantial increase in the number of links with V/C values below 1.0,
signifying a state of low congestion. In contrast, the count of links experiencing high congestion (i.e., V/C greater than 1.0) can be
reduced to a certain extent. Furthermore, with the establishment of more vertiports, such as five in number, the positive impact on
alleviating urban congestion becomes more pronounced. However, it is imperative to comprehensively evaluate this positive effect
against the economic costs associated with vertiport construction. Determining an equilibrium point requires a nuanced consideration,
a top that warrants further exploration in subsequent studies.

5.2. Experiments on the Anaheim network

The Anaheim network, as shown in Fig. 8, has 416 nodes and 914 links, with a total number of 1406 OD pairs. In this experiment,
we assume that there are 16 alternative vertiports available for the authority to choose. We aim to select eight vertiports out of 16.
Parameter settings on the network design problem and methods are the same as those in Section 5.1.
The MI-BO method is applied to solve the network design problem. For comparison, its performance is evaluated against the GA
method, which is widely used in network design studies. Both methods are run 10 times using the same initial solution. Fig. 9 shows the
convergence curves for each method (MI-BO in blue, GA in red). The x-axis represents the number of traffic assignment model in-
vocations, the most time-consuming aspect of solving network design problems, especially in large-scale networks. The y-axis shows
the total travel time of the traffic system. The solid line represents the mean value across 10 experiments, with the variance also
depicted. The results demonstrate that the proposed MI-BO method achieves both higher solution accuracy and efficiency. Starting
from the initial solution, MI-BO reduced the total travel time to 1,250,000 s, taking approximately 1/8 of the time required by GA.
Moreover, MI-BO shows stable performance across the 10 experiments, indicating its effectiveness in consistently identifying satis-
factory solutions.
We tested the impact of capacity values on total network travel time, based on the optimal network design solutions. As depicted in
Fig. 10, it shows the total travel time of the network under different capacity values. As illustrated, within the selected vertiports, the
variation in capacity values has a slight effect on total travel time. This suggests that in practical implementation; by selecting

Table 2
Network performance with and without incorporating the UAM.
Indicators Without UAM Two vertiports Three vertiports Four vertiports Five vertiports

Vertiport capacity _ 10000 10000 9511 10000

Auto share 100 % 96.4 % 94.9 % 92.9 % 90.4 %
LAC share 0.0 % 3.6 % 5.1 % 7.1 % 9.6 %
Total travel time 45114.9 41199.7 39891.0 38371.0 36524.9

Notes: LAC denotes the land-air collaboration.

14
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

Fig. 7. The distribution of V/C in different network design schemes.

Fig. 8. The layout of the Anaheim network.

appropriate vertiports, lower capacity values can be used to meet network travel demands.
We then explored the impact of UAM on ground traffic congestion under varying levels of traffic demand. Specifically, we scaled
the original traffic demand by factors of 0.4, 0.8, 1, 1.2, and 1.6. For each demand level, we solved the CTNDP to generate a corre-
sponding network design scheme. We then evaluated network performance before and after incorporating UAM. The results are
presented in Table 3, which summarizes the demand distribution, flow distribution over the ground network, and total travel time. For
the flow distribution, we calculated the percentage of links with varying V/C ratios.
Table 3 reveals the following insights: i) As traffic demand increases, the proportion of demand served by UAM slightly decreases.
ii) In terms of flow distribution, integrating UAM into the transportation network significantly reduces the proportion of congested
links on the original road network, particularly by decreasing the percentage of mildly congested links within the urban network.
Observing the percentage of congested links across different demand levels, we find that UAM’s congestion mitigation effect becomes
more pronounced under high-traffic demand conditions. These findings indicate that UAM can effectively alleviate road traffic
congestion. iii)This conclusion is further supported by the total travel time: as traffic demand grows, the reduction in total travel time
due to UAM implementation becomes more substantial.

6. Conclusions

This paper developed an innovative framework for modeling the land-air coordinated operations, where the emerging concept of
UAM is integrated with conventional travel modes. Focusing on the land-air collaboration composite transport networks, we proposed

15
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

Fig. 9. Convergence curves of each method.

Fig. 10. Sensitivity analysis of capacity values.

a bi-level modeling framework for the addressed topic. The objective of the upper level is to determine optimal vertiport locations and
capacities, minimizing the total travel time of the transportation system. The lower-level problem, capturing commuters’ route choice
behavior in the composite transport network, is solved using a mathematical programming model. A tailored MI-BO method, incor-
porating an efficient IGP algorithm for the LAC-NE problem, was developed to solve the proposed model. Numerical experiments
validated the effectiveness of the analysis framework and the solution method. With the emergence of eVTOL vehicles, the proposed
model can provide decision support for authorities and the formulation of relevant management policies. In summary, the proposed
model exhibits a higher degree of foresight in developing land-air coordinated operations, offering broad potential applications as a
planning tool.
Notably, the proposed optimization model serves as a foundational framework and represents an initial exploration of the UAM
concept within urban transportation networks. Future extensions could include: (i) considering additional travel modes (e.g., metro,
transit, and bike) when investigating the LAC-CTNDP. (ii) addressing the stochastic nature of travel demands or travel times to enhance
the robustness and reliability of the composite transportation network. (iii) incorporating realistic factors related to eVTOL vehicles in
the behavioral component of the lower-level problem to better characterize travelers’ behavior when choosing eVTOL vehicles for
daily commuting. (iv) regulating the LAC-CTNDP based on the time-dependent transportation services.

16
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

Table 3
Network Performance with and without UAM across Traffic Demand Levels.
Demand level 0.4 0.8 1.0 1.2 1.6

Anaheim network without UAM

Demand distribution (%) Auto 100 100 100 100 100
LAC 0 0 0 0 0
Flow distribution (%) 0–0.5 95.95 78.75 75.36 71.85 64.73
0.5–1 3.94 18.84 18.18 17.31 17.96
1–1.5 0.11 2.19 6.13 9.75 13.14
1.5–2 0.00 0.22 0.22 0.88 3.50
2- 0.00 0.00 0.11 0.22 0.66
Total travel time 501641 1061180 1418920 1868570 3263760
Anaheim network with UAM
Demand distribution (%) Auto 97.05 97.29 97.35 97.42 97.56
LAC 2.95 2.71 2.65 2.58 2.44
Flow distribution (%) 0–0.5 97.83 84.47 80.13 75.46 67.21
0.5–1 1.95 13.79 15.85 17.59 20.20
1–1.5 0.22 1.52 3.47 5.97 8.90
1.5–2 0.00 0.22 0.54 0.87 3.04
2- 0.00 0.00 0.00 0.11 0.65
Total travel time 428474 900598 1194270 1567260 2757140

CRediT authorship contribution statement

Honggang Zhang: Writing – original draft, Validation, Methodology, Investigation, Formal analysis, Conceptualization. Jinbiao
Huo: Writing – original draft, Validation, Methodology, Investigation, Formal analysis. Churong Chen: Visualization, Validation,
Investigation, Formal analysis. Zhiyuan Liu: Writing – review & editing, Validation, Supervision, Funding acquisition,
Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
influence the work reported in this paper.

Acknowledgments

This study is supported by the Key Project (No. 52131203) of the National Natural Science Foundation of China; the Natural Science
Foundation of Jiangsu Province (No. BK20232019); the Jiangsu Provincial Scientific Research Center of Applied Mathematics (No.
BK20233002).

References

Alkaabneh, F., Diabat, A., Elhedhli, S., 2019. A Lagrangian heuristic and GRASP for the hub-and-spoke network system with economies-of-scale and congestion.
Transp. Res. Part C: Emerging Technol. 102, 249–273.
Bacchini, A., Cestino, E., Van Magill, B., Verstraete, D., 2021. Impact of lift propeller drag on the performance of eVTOL lift plus cruise aircraft. Aerosp. Sci. Technol.
109, 106429.
Baptista, R., Poloczek, M., 2018. Bayesian optimization of combinatorial structures. Int. Conference on Machine Learning (PMLR) 462–471.
Beckmann, M., Mcguire, C.B., Winsten, C.B., 1956. Studies in the Economics of Transportation Yale University Press. New Haven, Connecticut.
Bell, M.G.H., Cassir, C., 2002. Risk-averse user equilibrium traffic assignment: an application of game theory. Transp. Res. Part B: Methodological 36 (8), 671–681.
Bombelli, A., Santos, B.F., Tavasszy, L., 2020. Analysis of the air cargo transport network using a complex network theory perspective. Transp. Res. Part E: Logistics
and Transp. Rev. 138, 101959.
Brunelli, M., Ditta, C.C., Postorino, M.N., 2023a. New infrastructures for urban air mobility systems: A systematic review on vertiport location and capacity. J. Air
Transp. Manag. 112, 102460.
Brunelli, M., Ditta, C.C., Postorino, M.N., 2023b. SP surveys to estimate airport shuttle demand in an urban air mobility context. Transp. Policy 141, 129–139.
Cantarella, G.E., Pavone, G., Vitetta, A., 2006. Heuristics for urban road network design: Lane layout and signal settings. Eur. J. Oper. Res. 175 (3), 1682–1695.
Chai, S., Yin, J., D’ariano, A., Sama, M., Tang, T., 2023. Train schedule optimization for commuter-metro networks. Transp. Res. Part C: Emerging Technol. 155,
104278.
Chen, M.Y., Alfa, A.S., 1991. A network design algorithm using a stochastic incremental traffic assignment approach. Transp. Sci. 25 (3), 215–224.
Cheng, Q., Liu, Z., Guo, J., Wu, X., Pendyala, R., Belezamo, B., Zhou, X., 2022. Estimating key traffic state parameters through parsimonious spatial queue models.
Transp. Res. Part C: Emerging Technol. 137, 103596.
Cho, Y., Oh, S., Egger, B., 2020. Performance modeling of parallel loops on multi-socket platforms using queueing systems. IEEE Trans. Parallel Distrib. Syst. 31 (2),
318–331.
Chow, A.H.F., 2009. Properties of system optimal traffic assignment with departure time choice and its solution method. Transp. Res. Part B: Methodological 43 (3),
325–344.
Cuesta Ramirez, J., Le Riche, R., Roustant, O., Perrin, G., Durantin, C., Glière, A., 2022. A comparison of mixed-variables Bayesian optimization approaches. Adv.
Modeling and Simulation in Eng. Sci. 9 (1), 6.
Daganzo, C.F., Sheffi, Y., 1977. On stochastic models of traffic assignment. Transp. Sci. 11 (3), 83–111.
Dantzig, G.B., Harvey, R.P., Lansdowne, Z.F., Robinson, D.W., Maier, S.F., 1979. Formulating and solving the network design problem by decomposition. Transp. Res.
Part B: Methodological 13 (1), 5–17.

17
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

Di, Z., Yang, L., Qi, J., Gao, Z., 2018. Transportation network design for maximizing flow-based accessibility. Transp. Res. Part B: Methodological 110, 209–238.
Du, W., Guo, T., Chen, J., Li, B., Zhu, G., Cao, X., 2021. Cooperative pursuit of unauthorized UAVs in urban airspace via multi-agent reinforcement learning. Transp.
Res. Part C: Emerging Technol. 128, 103122.
Du, M., Zhou, J., Chen, A., Tan, H., 2022. Modeling the capacity of multimodal and intermodal urban transportation networks that incorporate emerging travel
modes. Transp. Res. Part E: Logistics and Transp. Rev. 168, 102937.
Farahani, R.Z., Miandoabchi, E., Szeto, W.Y., Rashidi, H., 2013. A review of urban transportation network design problems. Eur. J. Oper. Res. 229 (2), 281–302.
Frazier, P.I. (2018) A Tutorial on Bayesian optimization, arXiv preprint arXiv:1807.02811.
Gallo, M., Montella, B., D’acierno, L., 2011. The transit network design problem with elastic demand and internalisation of external costs: An application to rail
frequency optimisation. Transp. Res. Part C: Emerging Technol. 19 (6), 1276–1305.
Garrido-Merchán, E.C., Hernández-Lobato, D., 2020. Dealing with categorical and integer-valued variables in bayesian optimization with gaussian processes.
Neurocomputing 380, 20–35.
Gu, Y., Chen, A., Kitthamkesorn, S., 2022. Accessibility-based vulnerability analysis of multi-modal transportation networks with weibit choice models. Multimodal
Transportation 1 (3), 100029.
Gu, Z., Shafiei, S., Liu, Z., Saberi, M., 2018. Optimal distance- and time-dependent area-based pricing with the network fundamental diagram. Transp. Res. Part C:
Emerging Technol. 95, 1–28.
Guo, Z., Hao, M., Liu, J., Yu, B., Jiang, Y., 2022. Joint routing and charging optimization for eVTOL aircraft recovery. Aerosp. Sci. Technol. 126, 107595.
Guo, R., Yang, H., Huang, H., 2013. A discrete rational adjustment process of link flows in traffic networks. Transp. Res. Part C: Emerging Technol. 34, 121–137.
Hong L.J., Zhang X. (2021) Surrogate-based simulation optimization. Tutorials in Operations Research: Emerging Optimization Methods and Modeling Techniques with
Applications (INFORMS), 287-311.
Hosseininasab, S.M., Shetab-Boushehri, S.N., 2015. Integration of selecting and scheduling urban road construction projects as a time-dependent discrete network
design problem. Eur. J. Oper. Res. 246 (3), 762–771.
Huang, D., Liu, Z., Fu, X., Blythe, P.T., 2018. Multimodal transit network design in a hub-and-spoke network framework. Transportmetrica A: Transport Sci. 14 (8),
706–735.
Huo, J., Liu, Z., Chen, J., Cheng, Q., Meng, Q., 2023. Bayesian optimization for congestion pricing problems: a general framework and its instability. Transp. Res. Part
B: Methodological 169, 1–28.
Iliopoulou, C., Kepaptsoglou, K., 2021. Robust electric transit route network design problem (RE-TRNDP) with delay considerations: model and application. Transp.
Res. Part C: Emerging Technol. 129, 103255.
Jiang, Y., Nielsen, O.A., 2022. Urban multimodal traffic assignment. Multimodal Transp. 1 (3), 100027.
Jin, Z., Ng, K.K.H., Zhang, C., Wu, L., Li, A., 2024. Integrated optimisation of strategic planning and service operations for urban air mobility systems. Transp. Res.
Part A: Policy and Practice 183, 104059.
Jones, D.R., Schonlau, M., Welch, W.J., 1998. Efficient global optimization of expensive black-box functions. J. Glob. Optim. 13, 455–492.
Khani, A., 2019. An online shortest path algorithm for reliable routing in schedule-based transit networks considering transfer failure probability. Transp. Res. Part B:
Methodological 126, 549–564.
Kim, J., Tak, S., Lee, J., Yeo, H., 2023. Integrated design framework for on-demand transit system based on spatiotemporal mobility patterns. Transp. Res. Part C:
Emerging Technol. 150, 104087.
Kleijnen, J.P.C., Van Beers, W., Van Nieuwenhuyse, I., 2012. Expected improvement in efficient global optimization through bootstrapped kriging. J. Glob. Optim. 54
(1), 59–73.
Leblanc, L.J., Boyce, D.E., 1986. A bilevel programming algorithm for exact solution of the network design problem with user-optimal flows. Transp. Res. Part B:
Methodological 20 (3), 259–265.
Leblanc, L.J., Morlok, E.K., Pierskalla, W.P., 1975. An efficient approach to solving the road network equilibrium traffic assignment. Transp. Res. 9 (5), 309–318.
Lee, Y., Gwak, C., Bae, S., Park, K.R., Lee, H.J., Lee, J.W., 2024. Functional analysis and validation of a uam integrated flight management (UIFM) System. Int. J.
Aeronaut. Space Sci. 25 (2), 749–758.
Lee, K., Jiang, Y., Ceder, A., Dauwels, J., Su, R., Nielsen, O.A., 2022. Path-oriented synchronized transit scheduling using time-dependent data. Transp. Res. Part C:
Emerging Technol. 136, 103505.
Li, Y., Liu, M., 2022. Path planning of electric VTOL UAV considering minimum energy consumption in urban areas. Sustainability 14 (20), 13421.
Li, T., Sun, H., Wu, J., Gao, Z., Ge, Y.E., Ding, R., 2019. Optimal urban expressway system in a transportation and land use interaction equilibrium framework.
Transportmetrica A: Transport Sci. 15 (2), 1247–1277.
Liu, T., Tang, L., Wang, W., He, X., Chen, Q., Zeng, X., Jiang, H., 2022. Resource allocation in DT-assisted internet of vehicles via edge intelligent cooperation. IEEE
Internet Things J. 9 (18), 17608–17626.
Long, J., Gao, Z., Zhang, H., Szeto, W.Y., 2010. A turning restriction design problem in urban road networks. Eur. J. Oper. Res. 206 (3), 569–578.
Luan, M., Waller, S.T., Rey, D., 2023. A non-additive path-based reward credit scheme for traffic congestion management. Transp. Res. Part E: Logistics and Transp.
Rev. 179, 103291.
Luong, P., Gupta, S., Nguyen, D., Rana, S., Venkatesh, S., 2019. Bayesian optimization with discrete variables. In AI 2019: Advances in Artificial Intelligence: 32nd
Australasian Joint Conference.
Lv, D., Zhang, W., Wang, K., Hao, H., Yang, Y., 2024. Urban Aerial Mobility for airport shuttle service. Transp. Res. Part A: Policy and Practice 188, 104202.
Ma, J., Meng, Q., Cheng, L., Liu, Z., 2022. General stochastic ridesharing user equilibrium problem with elastic demand. Transp. Res. Part B: Methodological 162,
162–194.
Macias, J.E., Khalife, C., Slim, J., Angeloudis, P., 2023. An integrated vertiport placement model considering vehicle sizing and queuing: a case study in London. J. Air
Transp. Manag. 113, 102486.
Mansourianfar, M.H., Gu, Z., Waller, S.T., Saberi, M., 2021. Joint routing and pricing control in congested mixed autonomy networks. Transp. Res. Part C Emerging
Technol. 131, 103338.
Mathew, T.V., Sharma, S., 2009. Capacity expansion problem for large urban transportation networks. J. Transp. Eng. 135 (7), 406–415.
Mesbah, M., Sarvi, M., Currie, G., 2008. New methodology for optimizing transit priority at the network level. Transp. Res. Rec. 2089, 93–100.
Pinto, H.K.R.F., Hyland, M.F., Mahmassani, H.S., Verbas, I.O., 2020. Joint design of multimodal transit networks and shared autonomous mobility fleets. Transp. Res.
Part C: Emerging Technol. 113, 2–20.
Pinto Neto, E.C., Baum, D.M., De Almeida Jr, J.R., Camargo Jr, J.B., Cugnasca, P.S., 2022. A trajectory evaluation platform for urban air mobility (UAM). IEEE Trans.
Intell. Transp. Syst. 23 (7), 9136–9145.
Poorzahedy, H., Abulghasemi, F., 2005. Application of ant system to network design problem. Transp. 32 (3), 251–273.
Pu, F., Yin, J., Wang, Y., Su, S., Yang, L., Tang, T., 2022. Rolling stock allocation and timetabling for urban rail transit network with multiple depots. Transp. Res. Rec.
2676 (11), 422–435.
Rambha, T., Boyles, S.D., Unnikrishnan, A., Stone, P., 2018. Marginal cost pricing for system optimal traffic assignment with recourse under supply-side uncertainty.
Transp. Res. Part B: Methodological 110, 104–121.
Rashidi, E., Parsafard, M., Medal, H., Li, X., 2016. Optimal traffic calming: a mixed-integer bi-level programming model for locating sidewalks and crosswalks in a
multimodal transportation network to maximize pedestrians’ safety and network usability. Transp. Res. Part E: Logistics and Transp. Rev. 91, 33–50.
Reiche, C., Cohen, A.P., Fernando, C., 2021. An initial assessment of the potential weather barriers of urban air mobility. IEEE Trans. Intell. Transp. Syst. 22 (9),
6018–6027.
Ryu, S., Chen, A., Choi, K., 2017. Solving the combined modal split and traffic assignment problem with two types of transit impedance function. Eur. J. Oper. Res.
257 (3), 870–880.

18
H. Zhang et al. Transportation Research Part C 171 (2025) 104967

Scarinci, R., Markov, I., Bierlaire, M., 2017. Network design of a transport system based on accelerating moving walkways. Transp. Res. Part C: Emerging Technol. 80,
310–328.
Schweiger, K., Schmitz, R., Knabe, F., 2023. Impact of wind on eVTOL operations and implications for vertiport airside traffic flows: a case study of Hamburg and
Munich. Drones 7 (7), 464.
Shon, H., Kim, S., Lee, J., 2024. Optimal planning of urban air mobility systems accounting for ground access trips. Int. J. Sustain. Transp. 18 (4), 356–378.
Szeto, W.Y., Jiang, Y., 2014. Transit route and frequency design: Bi-level modeling and hybrid artificial bee colony algorithm approach. Transp. Res. Part B:
Methodological 67, 235–263.
Tang, H., Zhang, Y., Mohmoodian, V., Charkhgard, H., 2021. Automated flight planning of high-density urban air mobility. Transp. Res. Part C: Emerging Technol.
131, 103324.
Tay, T., Osorio, C., 2022. Bayesian optimization techniques for high-dimensional simulation-based transportation problems. Transp. Res. Part B: Methodological 164,
210–243.
Tripathi, M., Mandal, M., Wadhwa, R., 2022. Air taxis: a technological breakthrough to beat the traffic woes. Commun. Assoc. Inf. Syst. 50, 317–336.
Tseremoglou, I., Bombelli, A., Santos, B.F., 2022. A combined forecasting and packing model for air cargo loading: a risk-averse framework. Transp. Res. Part E:
Logistics and Transp. Rev. 158, 102579.
Wahba, M.M.A. (2004) A New Modelling Framework for the Transit Assignment Problem: A multi-agent Learning-based Approach University of Toronto (Canada), Ontario,
Canada.
Wang, G., Chen, A., Kitthamkesorn, S., Ryu, S., Qi, H., Song, Z., Song, J., 2020b. A multi-modal network equilibrium model with captive mode choice and path size
logit route choice. Transp. Res. Part A: Policy and Practice 136, 293–317.
Wang, K., Jacquillat, A., Vaze, V., 2022a. Vertiport planning for urban aerial mobility: an adaptive discretization approach. M&som-Manufacturing & Service
Operations Manage, 24 (6), 3215–3235.
Wang, K., Li, A., Qu, X., 2023. Urban aerial mobility: network structure, transportation benefits, and Sino-US comparison. The Innovation 4 (2), 100393.
Wang, Y., Liu, H., Fan, Y., Ding, J., Long, J., 2022b. Large-scale multimodal transportation network models and algorithms-Part II: network capacity and network
design problem. Transp. Res. Part E: Logistics and Transp. Rev. 167, 102918.
Wang, S., Meng, Q., Yang, H., 2013. Global optimization methods for the discrete network design problem. Transp. Res. Part B: Methodological 50, 42–60.
Wang, K., Qu, X., 2023. Urban aerial mobility: Reshaping the future of urban transportation. The Innovation 4 (2), 100392.
Wang, G., Xu, M., Grant-Muller, S., Gao, Z., 2020a. Combination of tradable credit scheme and link capacity improvement to balance economic growth and
environmental management in sustainable-oriented transport development: a bi-objective bi-level programming approach. Transp. Res. Part A: Policy and
Practice 137, 459–471.
Wardrop, J.D. (1952) Some theoretical aspects of road traffic research. Proceedings of the Institute of Civil Engineers 1(3):325-362.
Wie, B.W., Tobin, R.L., Bernstein, D., Friesz, T.L., 1995. A comparison of system optimum and user equilibrium dynamic traffic assignments with schedule delays.
Transp. Res. Part C: Emerging Technol. 3 (6), 389–411.
Wu, Z., Zhang, Y., 2021. Integrated network design and demand forecast for on-demand urban air mobility. Eng. 7 (4), 473–487.
Xie, C., Liu, Z., 2014. On the stochastic network equilibrium with heterogeneous choice inertia. Transp. Res. Part B: Methodological 66, 90–109.
Xie, J., Nie, Y., Liu, X., 2018. A greedy path-based algorithm for traffic assignment. Transp. Res. Rec. 2672 (48), 36–44.
Xu, Z., Xie, J., Liu, X., Nie, Y., 2020. Hyperpath-based algorithms for the transit equilibrium assignment problem. Transp. Res. Part E: Logistics and Transp. Rev. 143,
102102.
Yan, Y., Wang, K., Qu, X., 2024. Urban air mobility (UAM) and ground transportation integration: a survey. Frontiers of Eng. Manage. 18, 1–25.
Yang, H., Bell, M.G.H., 1998. Models and algorithms for road network design: a review and some new developments. Transp. Rev. 18 (3), 257–278.
Yang, H., Wang, J.Y.T., 2002. Travel time minimization versus reserve capacity maximization in the network design problem. Transp. Res. Rec. 1783, 17–26.
Yao, J., Shi, F., Zhou, Z., Qin, J., 2012. Combinatorial optimization of exclusive bus lanes and bus frequencies in multi-modal transportation network. J. Transp. Eng.
138 (12), 1422–1429.
Yao, J., Huang, W., Chen, A., Cheng, Z., An, S., Xu, G., 2019. Paradox links can improve system efficiency: an illustration in traffic assignment problem. Transp. Res.
Part B: Methodological 129, 35–49.
Ye, J., Jiang, Y., Chen, J., Liu, Z., Guo, R., 2021. Joint optimisation of transfer location and capacity for a capacitated multimodal transport network with elastic
demand: a bi-level programming model and paradoxes. Transp. Res. Part E: Logistics and Transp. Rev. 156, 102540.
Yin, R., Liu, Z., Zheng, N., 2022a. A simulation-based model for continuous network design problem using Bayesian optimization. IEEE Trans. Intell. Transp. Syst. 23
(11), 20352–20367.
Yin, R., Liu, X., Zheng, N., Liu, Z., 2022b. Simulation-based analysis of second-best multimodal network capacity. Transp. Res. Part C: Emerging Technol. 145,
103925.
Yin, J., Tang, T., Yang, L., Xun, J., Huang, Y., Gao, Z., 2017. Research and development of automatic train operation for railway transportation systems: a survey.
Transp. Res. Part C: Emerging Technol. 85, 548–572.
Yu, B., Kong, L., Sun, Y., Yao, B., Gao, Z., 2015. A bi-level programming for bus lane network design. Transp. Res. Part C: Emerging Technol. 55, 310–327.
Zhan, D., Xing, H., 2020. Expected improvement for expensive optimization: a review. J. Glob. Optim. 78 (3), 507–544.
Zhang, H., Liu, Z., Wang, J., Wu, Y., 2023a. A novel flow update policy in solving traffic assignment problems: successive over relaxation iteration method. Trans. Res.
Part E: Logistics and Trans. Rev. 174, 103111.
Zhang, Q., Liu, S.Q., D’ariano, A., 2023c. Bi-objective bi-level optimization for integrating lane-level closure and reversal in redesigning transportation networks.
Oper. Res. 23 (2), 12267.
Zhang, H., Liu, Z., Dong, Y., Zhou, H., Liu, P., Chen, J., 2024. A novel network equilibrium model integrating urban aerial mobility. Transp. Res. Part A: Policy and
Practice 187, 104160.
Zhang, K., Zhang, H., Dong, Y., Wu, Y., Chen, X., 2023b. An ADMM-based parallel algorithm for solving traffic assignment problem with elastic demand. Commun.
Transp. Res. 3, 100108.
Zhou, X.S., Cheng, Q., Wu, X., Li, P., Belezamo, B., Lu, J., Abbasi, M., 2022. A meso-to-macro cross-resolution performance approach for connecting polynomial arrival
queue model to volume-delay function with inflow demand-to-capacity ratio. Multimodal Transportation 1 (2), 100017.

19