Transportation Research Part C: Christopher Cummings, Hani Mahmassani

Downloaded from https://iranpaper.
ir
https://www.tarjomano.com https://www.tarjomano.com
Transportation Research Part C 159 (2024) 104467
Contents lists available at ScienceDirect
Transportation Research Part C

journal homepage: www.elsevier.com/locate/trc
Airspace Congestion, flow Relations, and 4-D fundamental

Diagrams for advanced urban air mobility☆
Christopher Cummings a, Hani Mahmassani b, *
a
Transportation Center, Northwestern University, 600 Foster Street, Evanston, IL 60208, USA
b
William A. Patterson Distinguished Chair in Transportation, Transportation Center, Northwestern University, 600 Foster Street, Evanston, IL
60208, USA
A R T I C L E I N F O A B S T R A C T
Keywords: This paper develops theoretical macroscopic air traffic flow models that relate vehicle density and
Urban air mobility spacing to traffic flow (throughput) measures under different operational parameters for un
Advanced air mobility structured airspace in the Advanced Air Mobility (AAM) context. Recognizing the role of conflicts
Airspace congestion
in air traffic flow, we relate vehicle density to the frequency of conflict occurrence in airspace
Macroscopic flow models
Fundamental diagrams
using a gas-kinetic analogy. The number of conflicts is then related to vehicle speeds using an
Traffic flow theory average speed loss per conflict. The effects of the speed reductions are coupled with density to
explore the fundamental diagram between flow rate and density. The theoretical models are
tested and validated with simulated results for a number of parameter levels. The models can be
applied for quick predictions of future traffic flow conditions, which will be especially useful for
operators or air traffic flow management systems. The theoretical and simulated findings also
provide operational and policy insights for AAM operators, planners, and modelers. Notable in
sights include the critical role of aircraft density in air traffic flow and the variable impact of that
density on traffic flow behavior. AAM operators and planners will need to closely manage the
airspace density to avoid large numbers of conflicts simultaneously and maintain acceptable
travel times and throughputs. Key operational parameters such as aircraft spacing requirements
and maximum aircraft speeds were also found to have significant impacts on traffic flow behavior,
and offer policy avenues for managing air traffic.
1. Introduction
Recent technological advances in battery technology, aircraft design, communications, and aircraft power systems have combined
to bring down small aircraft operating costs. Reductions in costs have opened up opportunities in new markets for air mobility services
such as cargo-carrying drones or passenger-carrying eVTOL vehicles. Collectively these areas are known as advanced air mobility
(AAM) and offer cheaper and more efficient alternatives to current services. While AAM services are still in their early stages, estimates
project that AAM could grow to serve tens or hundreds of thousands of trips daily (Alonso et al., 2017, Bulusu, 2019). To serve this
demand AAM operators will need to operate large fleets of aircraft in a finite area, placing a previously unimagined workload on
airspace. The presence of so many aircraft operating simultaneously and the need to enforce aircraft spacing requirements for safety
☆
This article belongs to the Virtual Special Issue on “special issue full title”.
* Corresponding author.
E-mail addresses: [email protected] (C. Cummings), [email protected] (H. Mahmassani).
https://doi.org/10.1016/j.trc.2023.104467
Received 9 March 2023; Received in revised form 26 September 2023; Accepted 24 December 2023
Available online 31 December 2023
0968-090X/© 2023 Elsevier Ltd. All rights reserved.
Downloaded from https://iranpaper.ir
C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467
purposes will create congestion issues for aircraft in the system. An ability to predict and manage congestion in airspace is critical.
Previous works in air traffic flows have attempted to estimate airspace capacity (Jardin, 2004, Sunil et al., 2018a, Bulusu, 2019),
but these works do not acknowledge that the effects of airspace congestion can significantly impact aircraft speeds, flow rates, and
travel times even before airspace capacity is reached. It is therefore insufficient to estimate airspace capacity only, a better under
standing of the patterns of congestion in air traffic flow is required for the entire range of aircraft densities. There is only a small
literature on air traffic flow models, and these models are incomplete in their ability to explain the factors of congestion and predict
traffic flow relationships based on key operating parameters.
The overall goal of this paper is to develop theoretical aviation traffic flow models that account for and explains the role of vehicle
density and operational parameters for unstructured airspace in the AAM context. The specific parameters of interest are vehicle
spacing requirements and maximum vehicle speed. We accomplish this goal by completing each of four objectives: (1) adapt the gas
law conflict prediction model to explain the impacts on conflict counts of vehicle density and vehicle parameters, (2) use the conflict
prediction model to create traffic flow models that directly use and explain the role of vehicle density and parameters, (3) test the
accuracy of the developed models through simulation of AAM vehicles in unstructured airspace, and (4) provide operational and policy
insights for AAM using the relationships found between vehicle density and parameters and traffic flow measures. These models
envision an advanced state of AAM service with decentralized and unstructured airspace along with nearly ubiquitous service that
creates uniformly random headings for aircraft. However, the models can also serve as a starting point for further development of air
traffic flow models for structured airspaces such as corridors.
The contributions of this paper will be important in further developing the theoretical understanding of low-altitude air traffic flow
in urban and regional settings. Other works focusing on air traffic flow either do not establish specific models with predictive capa
bilities, or only offer incomplete models that do not capture a full view of traffic flow variables. The models developed in this paper
extend and improve upon previous work by: (1) encompassing the full arc of airspace congestion development from aircraft density to
conflicts to impacts on average speeds and throughput, (2) reflecting the key roles that vehicle parameters such as maximum aircraft
speed and spacing play, (3) providing a direct and interpretable framework for testing the impacts of various vehicle and operating
contexts, and (4) reducing the number of variables and contexts requiring calibration for the models to be implemented. The models in
this paper provide a basis for quick approximations of air traffic flow and can be applied in other work on AAM networks and within air
traffic flow management systems, which rely upon future approximations of traffic flow conditions. The paper’s results are important
to AAM operators, air traffic controllers, and aviation planners in that the model can be used to predict airspace congestion for a given
context without the need for computationally intensive repetitive simulations. These predictions of airspace congestion will inform
expected vehicle speeds, flow rates, travel times, and airspace capacity for any given context of vehicle density and parameters. The
models will demonstrate the trade-offs made by operators and planners when choosing among the number of vehicles operating
simultaneously and vehicle parameters, greatly enhancing their decision-making processes.
2. Background review
The goal of this study is to develop a macroscopic theoretical model of aircraft traffic flow, specifically for the AAM context.
Previous work in air traffic flows has been largely focused on air traffic management (ATM), which is primarily concerned with aircraft
safety, tactical management of aircraft trajectories and conflict resolution between aircraft (Mao et al., 2001, Pallotino et al. 2016,
Stroe & Andrei, 2016, Liu, 2018, Ribeiro et al., 2020). Numerous conflict resolution algorithms have been proposed and tested in the
literature encompassing centralized and decentralized control, optimized and heuristic based solutions, simultaneous and sequential
solution orders, and heading-based, altitude-based, or speed-based detours (Ribeiro et al., 2020). The majority of these algorithms deal
with tactical solutions or path-planning instead of strategic solutions.
Other work has been done on air traffic flow management (ATFM), such as the development of the Eulerian model (Bayen et al.,
2006, Menon et al., 2006, Work & Bayen, 2008). These works commonly consider airspace as either a network of links and nodes
(Bayen et al., 2006, Work & Bayen, 2008) or set of zones (Menon et al., 2006). Each link or zone is capacitated and a model is
developed to optimize the flow of aircraft over the links or zones considering the capacities. These works generally do not consider
tactical management of aircraft trajectories or conflict resolution, and therefore could be considered strategic solutions to managing
airspace. In these papers traffic flow models that relate vehicle density to speeds and flow rate are not developed.
The authors are only aware of a small number of previous studies that begin to develop theoretical models of aircraft traffic flow.
Cummings and Mahmassani (2021) simulated AAM vehicles and with the results noted traffic flow patterns similar to traffic flow
models developed and used in highway traffic. The work identified that that one-dimensional highway traffic models are applicable to
multi-dimensional airspace but does not develop traffic flow models for explaining these relationships or predicting future traffic flow
conditions. Haddad et al. (2021) also conjectures that such macroscopic traffic flow models are applicable to multi-dimensional
airspaces. A flow-density curve representing the fundamental diagram is fit to simulated data and used within a delay-focused con
trol system for air traffic flow management. However, no models explaining the relationship were created thus limiting the models
ability to extend beyond its current narrow context, and the relationships with conflicts and aircraft speeds were ignored. An air traffic
flow management application was further explored in Safadi et al. (2023), which considered microscopic aircraft interactions and an
optimal control strategy. Tereshchenko et al. (2020) went further towards developing a theoretical traffic flow model for aviation by
linking density to conflict probability using a modified gas law model and conflict resolution to effective speed reduction and declining
returns for effective flow rate. The model was then tested in a fixed-wing aircraft simulator. However, this traffic model was prob
lematic because it assumed uniform and constant vehicle speeds, did not illustrate the direct relationships between vehicle operating
parameters and traffic flow relationships, and offered only limited testing of the effect of vehicle spacing on traffic flow, with no
2
exploration of the role of other factors in traffic flow. Furthermore, several terms in these models required extensive calibration for
each specific demand level and operational context, a limitation similar to the limitation in the Haddad et al. (2021) approach.
Requiring such calibration for each narrow operating context reduces the model applicability for predictions. This paper goes beyond
previous studies by developing more complete models of air traffic flow. These models explicitly consider the role of aircraft speed in
conflicts and explore the effects of vehicle parameters in order to provide further operational and policy insights relative to previous
work.
Two other areas of study outside of ATM are directly relevant to this work. Conflict probability models are a key component in this
study in relating vehicle density to speed reductions. There is a long history of aircraft conflict probability work with numerous ap
proaches (Ribeiro et al., 2020). The authors have chosen to use the gas law approach for its simplicity and directness in applying the
vehicle parameters and vehicle density to conflict probability. The details of the gas law method are described in section 3.3. A number
of other studies have developed and used the gas law model primarily for evaluation of safety in airspace (Mitici & Blom, 2018) or
estimations of airspace capacity (Jardin, 2004, Sunil et al., 2018a).
Aircraft simulators are also a useful tool for studying aircraft traffic (Shah et al., 2005), and in recent years a number of simulators
have been developed and tested (Shah et al., 2005, Alonso et al., 2017, Sunil et al., 2018a, Xue et al., 2018, Bulusu, 2019, Neto et al.,
2019, Sun et al., 2020). Of specific interest to this paper are simulators that apply to the AAM context (Xue et al., 2018, Bulusu, 2019,
Neto et al., 2019). NASA has developed a simulator known as Fe3 for high-fidelity modeling of UAM vehicles by considering the
physical forces applying to aircraft (Xue et al., 2018). The NASA simulator uses what is known as the six-degrees-of-freedom approach.
Josselson (1997) outlines a method for implementing a six-degrees-of-freedom aircraft simulator, which is known for its accuracy of
vehicle motion. However, the high degree of detail in these models comes at a computational cost, making room for a variety of other
simulators that operate with simplified vehicle motions and assumptions (Bulusu, 2019, Liu 2019, Neto et al., 2019). While the vehicle
trajectories in these simulators may not as accurately reflect real-world conditions, they can still be useful in studying macroscopic
questions for ATM and air traffic flow (Tereschenko 2020, Cummings & Mahmassani, 2021). The theoretical models in this work are
tested by using such a simulator. Initially developed in Cummings and Mahmassani (2021), this AAM-specific simulator moves vehicles
through three-dimensional airspace, plans and updates vehicle trajectories in real-time, and resolves conflicts using a non-linear
optimization program that seeks to minimize vehicle travel time. While it is limited in dealing with aircraft forces and accelera
tions, it nevertheless is useful for macroscopic traffic modeling of AAM vehicles.
3. Methodology: Theoretical models
3.1. Definition of variables
Before the models can be developed, first the definitions of the key traffic flow measures must be discussed, which is done in this
section. Equations for the measurement of these variables are included where appropriate. A conflict is defined as the predicted
intrusion of one aircraft into the spacing requirements of another aircraft. Conflicts are detected in a pair-wise manner and each
conflict counted affects two aircraft. If there are multiple predicted conflicts involving the same aircraft simultaneously, the conflicts
are still counted based on the pairs of aircraft affecting each other. The spacing requirements in this paper take the shape of an
ellipsoid, which has been proposed for the AAM context because it offers greater flexibility (Geister & Korn, 2018).
Density refers to the number of aircraft within a given section of airspace. In this work density is measured in terms of vehicles per
cubic mile or vehicles per square mile. The density of aircraft within a given airspace is found using a multi-dimensional expansion of
Edie’s definitions (Edie, 1965, Saberi & Mahmassani, 2014, Saberi et al., 2014). Equation (1) shows the calculation of the vehicle
density in an airspace of dimensions X and Y when measuring over a time T.
∑
n τn
k= where τn is the time vehicle n spends in the airspace (1)
T(XY)
The speeds and flow rates of aircraft are broken into two types: actual and effective. The actual speed refers to the magnitude of
Euclidean distance a vehicle is traveling in a given time in any direction. Actual speed values are important in the operation of vehicles
and are used in the conflict prediction models. Meanwhile the effective speed measures the magnitude of distance moved towards the
vehicle’s destination in a given time. Effective speeds are useful for predicting travel times or demonstrating the cost of deviations from
a direct trajectory. Both actual and effective speeds are measured by considering the sum of all distances (actual or effective) moved in
the airspace divided by the amount of time considered.
In a similar way there are two flow rates: actual flow rate and effective flow rate. The actual flow rate refers to the flow of aircraft
through airspace in any direction and is not particularly interpretable. This work focuses instead on the effective flow rate, which is the
flow of aircraft towards each aircraft’s goal. The effective flow rate refers to the number of vehicles that can move through airspace in a
given time towards each vehicle’s destination. This is an important operational output measure as it defines the throughput of airspace.
The flow rates were also calculated using an expansion of Edie’s definitions (Saberi & Mahmassani, 2014) as shown in equation (2).
Equation (2) measures the sum of all distances moved within an airspace of dimensions X and Y over the time window considered T.
∑
dn
Qe = n where dn is the effective distance within the airspace moved by vehicle n (2)
YXT
3
3.2. Three key traffic flow concepts
The air traffic flow models derived and tested in this work encompass three key concepts for the development of airspace
congestion. Firstly, aircraft density in airspace creates conflicts, because more vehicles in the same airspace increases the
chances of conflicts for each vehicle. This relationship has been established by a wide body of work in aircraft conflict prediction
models, which this study draws from (Mitici & Blom, 2018, Sunil et al., 2018b). Secondly, conflicts reduce effective and actual
speeds of aircraft. It is assumed that without conflicts vehicles would travel on the shortest path to their destination to minimize
travel time. Therefore, to resolve a conflict one or both vehicles must deviate from the fastest route along one or more dimensions. This
deviation will always decrease effective speed because the aircraft is no longer traveling towards its destination as fast as it could be.
Some conflict resolutions will also reduce actual speed. Therefore, aggregated over many conflicts the average actual speed will
decrease. Thirdly, the effective flow rate, or throughput, is impacted by changes to the aircraft density and speeds. This is
consistent with the fundamental identity of traffic flow, which holds that the flow rate is the product of the density and speed.
These three concepts, summarized in Fig. 1, form the order in which rising aircraft densities in airspace affect traffic flow, and also
the order in which the theoretical models are developed in this section. The next subsection will discuss the gas law model for pre
dicting numbers of conflicts and how it has been adjusted in this work. After that the relationships between aircraft actual and effective
speeds and density are derived, followed by the models of actual and effective flow rates dependent on density.
3.3. Gas law conflict prediction model
There is a rich literature of aircraft conflict prediction models (Mitici & Blom, 2018). These models include gas law models, Monte
Carlo simulations, and Markov chain approximation. The gas law model, named that way because it is inspired by the study of gas
molecule collisions in chemistry, has been commonly used to predict conflict probabilities and rates between aircraft (Sunil et al.,
2018b). The model has been adjusted and reformulated in a variety of ways depending on the specific context and needs of the work.
The same applies here, where the model is applied to an unstructured airspace that assumes random headings for aircraft. The gas law
model is applicable to multi-dimensional situations, but the models developed here are simplified for the two-dimensional case. This
simplification primarily affects the relative velocity adjustment term. The models are also written in a form so as to highlight the role of
vehicle parameters and vehicle density in the number of conflicts, which is a primary objective of this work.
The gas law model begins by trying to estimate the conflict rate for a single aircraft in airspace. To do so, the expected number of
conflicts in the volume or area of airspace ahead of the aircraft is calculated. Fig. 2 demonstrates the shape of the volume of the
airspace considered. The calculation of the expected conflict rate for a single aircraft is shown in equation (3). First the size of the
volume or area is found by multiplying the cross-sectional area of the airspace the aircraft occupies (defined by the separation re
quirements) and the relative speed of the aircraft compared to other aircraft. This model considers unstructured airspace with aircraft
√̅̅̅
having uniformly random headings relative to each other, and therefore the relative speed of the aircraft is 2*vi where vi is the actual
speed of the aircraft. For a model with less random headings (such as a structured corridor), the relative speed of aircraft term should
be changed, although further investigation of that is outside of the scope of this paper. The volume of airspace searched can then be
multiplied by the average density of aircraft in airspace to estimate the number of aircraft in the volume, which will correspond to the
number of conflicts. In equation (3), Ci is the expected conflict rate for aircraft, i, k is the density of aircraft, vi is the current actual
speed of aircraft, i, and Dsep is the cross-sectional area of the separation requirements.
√̅̅̅
Ci = 2*vi *Dsep *k (3)
Once the conflict rate for each aircraft is determined it is multiplied by T, the time length of the look ahead window which each
aircraft is searching for conflicts. The product is the number of expected conflicts for each aircraft. This number is then multiplied by
the number of aircraft considered, which can be rewritten as k*Vol, or the density of aircraft multiplied by the total volume of airspace
studied. Lastly the number of conflicts is divided by 2 so that conflicts are not double-counted. Applying these factors creates an
equation for the expected number of conflicts between all aircraft in the considered airspace, which is shown in Eq. (4). In this equation
C is the expected number of conflicts in the airspace, v is the average actual speed of the aircraft considered, Dsep cross-sectional area of
the separation requirements, T is the time length of the look ahead window considered for conflicts, and Vol is the total volume of the
airspace studied.
√̅̅̅
2
C= *v*Dsep *k2 *TLA *Vol (4)
2
The conflict prediction model yields several observations from the relationships between variables. First, the number of conflicts
increases linearly with two vehicle parameters: the cross-sectional area of the separation requirements and the aircraft speed. These
Fig. 1. The process by which aircraft density in a given airspace affects the traffic flow of aircraft.
4
Fig. 2. Depiction of the airspace each aircraft searches for conflicts.
relationships will be explored in further detail in sections 3.4 and 4. Second, the number of conflicts increases due to density according
to a power law. This insight reflects the first key concept: that aircraft density creates conflicts. It is this power relationship that is
critical to understanding the aircraft traffic flow, because the number of conflicts will not just increase at higher densities, but increase
5
rapidly. From here traffic flow models utilizing the conflict prediction equation will be derived to demonstrate the role of vehicle
parameters in the development of traffic flow.
3.4. Theoretical Speed-Density models
In the previous section a conflict prediction model dependent on vehicle parameters and vehicle density was developed using the
gas law methodology. In this section traffic flow models are derived that utilize the conflict prediction model in order to understand
how vehicle parameters affect the traffic flow.
After the first key concept in which aircraft density leads to conflicts is the second concept, that resolving these conflicts reduces
both the actual and effective speeds of aircraft. We will begin by deriving the actual speed- density relationship, because this will be
necessary for the other traffic flow models. Assuming that aircraft fly the most direct route at the maximum possible speed unless they
are impacted by a conflict it is possible to model the average actual aircraft speed as shown in equation (5). In this equation the average
actual aircraft speed, v, is the maximum actual aircraft speed, vmax , less the average speed loss per vehicle, where the speed loss is the
average actual speed loss per conflict, b, multiplied by the number of conflicts, C, divided by the number of aircraft, n.
b*C
v = vmax − , where n is the number of aircraft (5)
n
In Eq. (5), C is the total number of conflicts among all aircraft in the studied airspace, and the conflict prediction model developed
above describes this. However, the conflict prediction model depended in part on the actual aircraft speed, which is what is solved for
here, so the relationship between actual speed and density will not be linear, as we will see momentarily. By substituting Eq. (4) into
Eq. (5), we arrive at Eq. (6).
√̅̅
b* 22*v*Dsep *k2 *T*Vol
v = vmax − (6)
n
It is possible to rewrite n, the total number of aircraft in the airspace studied, as the product of k*Vol, the density of vehicles and the
volume of airspace. Making this substitution and performing some algebra arrives at an actual speed- density function as shown in Eq.
(7). Dsep represents the cross-sectional area of the spacing requirements, k is the average vehicle density, and T is the time window over
which the average speed is measured.
vmax
v= √̅̅̅ (7)
(1 + 2*b*Dsep *k*T)
It is immediately notable that this speed-density model is not linear as in Greenshields’ model for highway traffic (Greenshields
et al., 1935). Again, this is because the average actual speed is modeled as a function of the number of conflicts in airspace, and the
number of conflicts itself is dependent on the actual speed of aircraft. The full implications of this relationship will be explored in
section 4, but for now it is notable that the average actual vehicle speeds have an inverse relationship with both vehicle density and the
spacing requirements, as opposed to a negative linear relationship.
The derivation of the average effective vehicle speed as a function of density begins in a similar way. The average effective speed, e,
is considered to be the maximum effective speed, emax , minus the effective speed loss, which is found by multiplying the effective speed
loss per conflict, a, by the total number of conflicts, C, and dividing by the total number of vehicles, C. It should be noted that the rate of
effective speed loss per conflict, a, is different than the rate of actual speed loss per conflict, b, depending on the form of conflict
resolution used. These terms are further explained in Eq. (12) below. The average effective speed as a function of conflicts is set up in
Eq. (8).
a*C
e = emax − (8)
n
Again, we can substitute in Eq. (4) for the total number of conflicts, creating Eq. (9) as follows.
√̅̅
a* 22*v*Dsep *k2 *T*Vol
e = emax − (9)
n
However, unlike with the average actual speed, the average effective speed does not directly influence the number of conflicts,
which will keep the relationship between average effective speed and average actual speed and density linear. This can be seen once we
again substitute in k*Vol for n and simplify the equation with algebra to arrive at the effective speed- density model shown in Eq. (10).
√̅̅̅
e = emax − 2*a*Dsep *v*k*T (10)
From this model it appears that there is a linear relationship between effective speed and density, however the actual speed, v, is
included in the model and depends on density as well. In effect the effective speed-density relationship follows a shape similar to the
actual speed-density model. The actual speed term, v, is variable according to Eq. (7), and can be removed from the model by sub
stitution, with the result being an effective speed – density function dependent only on the density variable. This is shown in Eq. (11).
√̅̅̅ vmax
e(k) = emax − 2*a*Dsep * √̅̅̅ *k*T (11)
(1 + 2*b*Dsep *k*T)
6
Both actual and effective speed models are dependent on the average speed loss per conflict, term b in equation (5) and term a in Eq.
(8). These terms are important because they relate the number of conflicts to speed reductions which result from conflict resolutions.
Logically we can conclude that a, the effective speed loss per conflict, will always be larger than or equal to b, the actual speed loss per
conflict, because effective speed measures the amount of the actual speed that is oriented towards the goal. The average speed losses
per conflict are difficult to estimate in part because they can fluctuate greatly due to contextual details: including the conflict resolution
algorithm used, c, the relative headings of the two aircraft, h, the separation volume or area of the aircraft, Vsep , the density of other
nearby aircraft, k, and the speeds of the aircraft, v. The impact of these relationships on average speed loss and the relationship between
actual speed loss per conflict, b, and effective speed loss per conflict, a, are reflected in Eq. (12).
( ) ( )
a c, h, Vsep , k, v ≥ b c, h, Vsep , k, v (12)
A model relating variables such as aircraft density, separation volume, and maximum aircraft speed can be derived from several
conceptual patterns. Aircraft flying in higher density airspaces will on average see a higher speed loss per conflict because the other
nearby aircraft decrease the solution space for conflict resolution. A smaller and less flexible solution space will generate longer detour
paths, reducing the average aircraft speeds. Additionally, larger separation volumes between aircraft will increase the speed loss per
conflict by similarly requiring longer detours to avoid conflicts. And higher maximum aircraft speeds will generate higher average
relative speeds for aircraft involved in conflict, causing greater speed loss from conflicts. Based upon these variable relationships,
models for the average speed losses per conflict can be calibrated for each specific context according to initial observed results from
even a low-density scenario. The calibrated model allows the a and b parameters to scale according to the demand on the airspace.
More details on calibrating the average speed loss per conflict parameters can be found in section 5.3.
3.5. Theoretical flow rate – Density model
The effective flow rate of aircraft through airspace is also of interest as a measure of the productivity of the airspace. The
fundamental identity holds that flow rate is the product of speed and density, and this applies here to effective speed as well in Eq. (13):
Qe = k*e (13)
Expanding Eq. (13) by substituting in the effective speed-density model for e reveals a parabolic model in Eq. (14). This parabolic
model will reflect the shape of the fundamental diagram. In this pattern the effective flow rate will initially increase proportionally
with the density, but as density increases more conflicts occur and vehicle speeds decrease, which results in decreasing returns to
density before ultimately a critical density is reached and the flow rate actually begins to decline.
√̅̅̅
Qe = k*emax − 2*a*Dsep *v*k2 *T (14)
The model of effective flow rate can be reduced to a function of only the density variable, k, by substituting in Eq. (7), the actual
speed-density model. The resulting function is Eq. (15).
√̅̅̅ vmax
Qe (k) = k*emax − 2*a*Dsep * √̅̅̅ *k2 *T (15)
(1 + 2*b*Dsep *k*T)
This section has described the derivation of traffic flow models utilizing the conflict prediction model. The next section will explore
how the vehicle parameters in these models impact traffic flow patterns of aircraft under different parameter contexts and the
Fig. 3. Theoretical conflict-density relationships from the models for varying spacing requirements (Fig. 3, left) and maximum vehicle speeds
(Fig. 4, right).
7
implications of these relationships.
4. Patterns from traffic flow models
In the previous section the theoretical traffic flow models were derived. These models covered the relationships between number of
conflicts and density, average actual vehicle speed and density, average effective speed and density, and effective flow rate and
density. In each of these models terms are included representing the separation requirements between aircraft (Dsep ) and the speed of
the aircraft (v), which are both important vehicle parameters that impact the patterns of traffic flow. In this section the patterns of
traffic flow as a function of density are explored graphically using the theoretical relationships and insights are discussed. In addition,
the effects of varying spacing and speed parameters are studied. This section is primarily concerned with the shape of the patterns
arising from the theoretical relationships, rather than the numerical outputs of the patterns. Therefore, throughout this section
arbitrarily assigned values are used for the theoretical models to illustrate the patterns.
The first key concept introduced in the theoretical models was that vehicle density in airspace creates conflicts, as a higher number
of vehicles in the airspace raises the probability of conflicts for each vehicle. Eq. (4) developed a conflict prediction model, which for a
given density of aircraft and set of vehicle parameters could predict the number of conflicts in airspace. Figs. 3 and 4 use that model to
predict the number of conflicts in airspace for varying levels of density. Furthermore, in Fig. 3 the horizontal spacing requirements
between vehicles is adjusted, while in Fig. 4 the maximum vehicle speed is adjusted.
The conflict patterns in these figures demonstrate a number of insights. The first is that higher densities have higher rates of
conflicts in airspace, which is consistent with the first key concept that density creates conflicts. The relationship is not linear, but a
power relationship instead, reflecting that the expected marginal number of conflicts added per vehicle is higher at higher densities.
This is true because adding one vehicle to airspace does not just introduce the potential for that vehicle to have conflicts, but also
increases the potential for conflicts with all other vehicles already in the airspace. This finding is significant because it suggests that at
some point even small additions to density could create many more conflicts in airspace. Since each conflict requires some amount of
travel time for a detour to resolve, at some point the marginal cost of adding vehicles to the airspace will exceed the marginal benefit of
flying more aircraft simultaneously.
Varying the vehicle parameters of spacing and speed also create interesting patterns of conflicts. To begin with, higher spacing
requirements are shown to be associated with higher conflict rates for a specific density across all levels of density. This occurs because
larger spacing requirements between aircraft effectively means that each aircraft takes up more airspace, and therefore there is a larger
chance that an aircraft trajectory intersects with other trajectories. While spacing requirements will ultimately be determined by safety
considerations, these results indicate that even small improvements in operations that allow for the spacing needs to be lowered can
significantly decrease the number conflicts in airspace that require resolution.
Eq. (4) also demonstrated the role of vehicle speed in creating conflicts in airspace. Higher vehicle speeds lead to a higher conflict
rate between vehicles at the same density because each vehicle is traveling farther and searching more airspace in its look ahead to
identify conflicts and in that airspace will find more vehicles in conflict. This relationship between higher speeds and higher conflict
rates is reflected in the differences between series in Fig. 4, in which faster vehicles are expected to develop more conflicts at a given
density than slower vehicles. Based on this insight operators may consider operating aircraft at lower speeds, especially in higher
density sections of airspace, in order to avoid higher conflict rates and the need for more complicated conflict resolutions.
Figs. 5 and 6 were developed to explore the theoretical relationships between average actual vehicle speed and density as
developed in Eq. (7). In Fig. 5 the different series represent varying levels of vehicle spacing requirements, while in Fig. 6 the series
Fig. 4. Theoretical conflict-density relationships from the models for varying spacing requirements (Fig. 3, left) and maximum vehicle speeds
(Fig. 4, right).
8
represent varying levels of maximum vehicle speed. In both figures it is interesting to note that the speed-density relationship is not
linear, as Greenshields model assumes for highway traffic. Rather because this model determines conflicts from vehicle speeds and the
conflicts in turn determine the reduction in average speed (see section 3.4 above), the relationship between speed and density is
curvilinear. Initially at low densities the rate of change in actual speed is low, because the conflict rate does not increase greatly with
density. However, as density increases and the conflict rate increases, the actual vehicle speeds will slow more significantly with added
density. Ultimately this rate of change will slow again as vehicles slow down, also decreasing the conflict rate given density. This
relationship is interesting firstly in that it represents a key difference from previous traffic models that assumed a linear speed-density
relationship for highway traffic (Greenshields 1935). In practicality this result is also interesting because it indicates that AAM op
erators can add aircraft to low density airspace without seeing significant reductions in actual vehicle speeds from conflicts. However,
adding vehicles to higher density airspaces will create larger speed reductions that are desirable to avoid.
Based on the theoretical results demonstrated in Fig. 5, it can be seen that traffic flows with larger spacing requirements between
aircraft will have lower average actual vehicle speeds at any significant level of density. This is because the larger spacing requirements
will create more conflicts between vehicles (see Fig. 3 above). This result emphasizes the benefits of lower spacing requirements.
Fig. 6 demonstrates the differences in average actual vehicle speed given density for several levels of maximum vehicle speed. At
lower densities the faster vehicles will move faster than the slower vehicles, as is obvious. However, at higher densities the difference in
average actual vehicle speed becomes less pronounced. This occurs because traffic flows with higher maximum vehicle speeds will
create more conflicts (see Fig. 4 above), which in turn reduces the average speeds more than traffic flows with lower maximum speeds.
Figs. 7 and 8 represent the theoretical effective speed-density relationship that was developed in Eq. (10). The patterns of this
relationship are similar to the actual speed-density relationship seen in Figs. 7 and 8. In practical terms this is because both actual and
effective speeds follow similar patterns of speed reduction as higher densities create more conflicts between aircraft.
The patterns of effective speed relative to density are similar to the patterns of actual speed relative to density in that the marginal
change in effective speed at low densities is low due to a low rate of change in conflicts. As densities rise the effective speed is reduced
more significantly due to more conflicts being created which require detours to resolve, these detours reduce the effective speed by
decreasing the vehicle speed towards its destination. The impacts of spacing requirements and maximum vehicle speeds are also
similar, in that larger spacing requirements see lower effective speeds and a sharper decline in effective speeds while higher vehicle
speeds see higher effective speeds but a sharper decline in effective speeds. Because effective speed as a measure of speed towards the
destination is directly linked to vehicle travel time, these results are particularly meaningful for operators. Higher effective speeds will
directly indicate shorter travel times. Higher spacing requirements are related to longer travel times and a sharper increase in travel
times, incentivizing operational methods that allow for lower spacing requirements while maintaining safety. Higher maximum
vehicle speeds are related to shorter travel times, but also have a sharper increase in travel times due to rising density than lower
maximum vehicle speeds. This is notable because it indicates that for high density airspace scenarios operators may end up with similar
travel times regardless of maximum vehicle speeds.
Effective flow rate is a key measure because it measures the output productivity of the airspace, or how the rate at which airspace
can process aircraft. The theoretical model for effective flow rate depending on vehicle density was developed in Eq. (14) by joining the
effective speed model with density. The patterns of this model are shown graphically in Figs. 9 and 10 below. The broad pattern reveals
the familiar fundamental diagram from traffic flow models (Johari et al., 2021). The fundamental diagram demonstrates that at low
densities the flow rate increases proportionally with the density because vehicle speeds are not significantly impacted. However, at
higher densities more conflicts are created, which in turn reduces vehicle speeds and decreases the positive rate of change of flow rate
given density. Eventually a critical density is reached when the effective flow rate is at its maximum, and further increases in vehicle
Fig. 5. Theoretical actual speed-density relationships from the models for varying spacing requirements (Fig. 5, left) and maximum vehicle speeds
(Fig. 6, right).
9
Fig. 6. Theoretical actual speed-density relationships from the models for varying spacing requirements (Fig. 5, left) and maximum vehicle speeds
(Fig. 6, right).
Fig. 7. Theoretical effective speed-density relationships from the models for varying spacing requirements (Fig. 7, left) and maximum vehicle
speeds (Fig. 8, right).
density actually cause the flow to breakdown and the flow rate to decline. At this point the airspace capacity is reached. This pattern is
manifested in the effective flow rate-density relationship and demonstrates clearly that airspace density must be carefully managed by
operators in order to avoid breakdowns in flow that occur at high density. Because there are decreasing returns to density and increases
in density increase travel times as previously noted, operators must balance a cost-benefit decision of operating fewer vehicles
simultaneously at higher speeds or more vehicles simultaneously at lower speeds. The fundamental diagram depicts the trade-off
between density and throughput which forms a key part of that decision.
Fig. 9 also demonstrates the different patterns of effective flow rate developed for varying spacing requirements. At low densities all
spacing requirements will have the same effective flow rate because there are few aircraft to create conflicts and slow down vehicles.
As densities rise however different spacing requirements cause traffic flows to reach very different critical densities, which greatly
impacts the effective flow rate. It is seen in Fig. 9 that larger spacing requirements lead to lower critical densities and lower effective
flow rates of aircraft, creating a lower capacity of airspace overall. This further reinforces the previous relationships that demonstrated
how reducing the spacing requirements safely can significantly increase the potential traffic flow by encountering fewer conflicts,
maintaining higher speeds, and allowing for higher flow rates.
Fig. 10 similarly shows the flow rate patterns for varying maximum vehicle speeds. At low densities each scenario has a different
rate of change in the flow rate because of the different vehicle speeds. From low to high densities the higher maximum vehicle speeds
allow for higher flow rates and a higher airspace capacity. However, the lower maximum speeds lead to higher critical densities
because of the lower rate of conflicts given density. This insight is interesting because it indicates that higher vehicle speeds do allow
for a higher throughput of vehicles in airspace, but for high density airspace lower vehicle speeds can operate without reaching the
critical density and flow breakdown. Such a finding recognizes that under certain situations with high airspace density operators may
10
Fig. 8. Theoretical effective speed-density relationships from the models for varying spacing requirements (Fig. 7, left) and maximum vehicle
Fig. 9. Theoretical effective flow rate-density relationships from the models for varying spacing requirements (Fig. 9, left) and maximum vehicle
restrict vehicle speeds in order to avoid reaching a critical density and flow breakdown.
5. Simulation model results
5.1. Simulation setup
Validation of the theoretical models is an important step. The third objective of this paper is to test the accuracy of the theoretical
models and validate their relationships through simulation. Simulated data is necessary for validation instead of observed data because
there are currently very few instances of AAM or similar vehicles operating, and none of vehicles operating at high density. A simulator
of the motion of AAM vehicles through airspace was used to generate simulated traffic flow data. The simulator tool was described in
more detail in Cummings and Mahmassani (2021); a simplified description of it is provided here for completeness.
The simulator tracks the three-dimensional positions of aircraft within an experimental airspace. The flight of each aircraft is
composed of three phases: an ascent phase from the vertiport to an initial altitude, a transiting phase in which the vehicle moves to a
point above the destination, and a descent phase to the destination vertiport. The positions of aircraft are tracked throughout the
simulation, however only aircraft within a smaller, centralized study airspace are considered for traffic flow measures. A smaller study
airspace is used instead of the larger experimental airspace in order to ignore the traffic issues that can arise from ascending and
descending aircraft near vertiports. In this way the simulated data only captures aircraft in the middle transiting phase of the flight. A
representation of the trajectories of vehicles is shown in Fig. 11.
11
Fig. 10. Theoretical effective flow rate-density relationships from the models for varying spacing requirements (Fig. 9, left) and maximum vehicle
A set of origin vertiports is generated separately from a set of destination vertiports. The vertiports are generated in a circle around
the centralized study airspace, and origins alternate with destinations. Aircraft are randomly assigned to an origin vertiport and are
given a randomly assigned destination vertiport. It is important that the vertiports are set up in such a way as to approximate random
headings in the study airspace, which is an assumption of the models for unstructured airspace. Vehicles are also randomly assigned a
starting time according to a normal distribution. The normal distribution is used to create rising and falling density in the study
airspace to gather traffic flow measurements for a range of densities with each simulation run. While the simulator is built to handle
three-dimensional air traffic the simulations used in this work were simplified to effectively two-dimensions by assigning all aircraft to
the same altitude for the transiting phase and disallowing vertical detours during conflict resolution. This simplification decreased
computation times and avoided the complexities of layered airspace.
During the transiting phase, aircraft always travel on the shortest possible path between their current position and the destination.
During conflict-free times of flight this means that aircraft will move in a straight line towards a point located above their destination.
When conflicts do arise aircraft will seek to minimize the travel time by solving for a valid trajectory that moves closest to the goal
within the look ahead window. Conflict detection is done by searching the local airspace surrounding an aircraft for other aircrafts, and
then checking the planned trajectories of those aircraft with the planned trajectory of the current aircraft. Conflict resolutions are
Fig. 11. Three dimensional representation of the trajectories of AAM vehicles in the simulation. The trajectories represent the series of vehicle
positions, with more recent more positions shown as more opaque. Note the three phases of movement: ascent (vertical) - transiting (horizontal) -
descent (vertical).
12
handled in a decentralized manner for each individual aircraft. A non-linear optimization program is employed that solves conflicts
sequentially beginning with the most conflicted aircraft trajectories (Pallotino et al., 2002). As mentioned before, altitude-based
detours are disallowed, leaving conflict resolutions to be heading-based and speed-based. In extreme cases with large numbers of
conflicts locally it is possible for each vehicle to stop moving altogether and “hover” in position. This ability reflects the AAM context
dependent largely on rotors for lift which can keep the aircraft aloft even when it is not moving as opposed to fixed wing aircraft which
needs to maintain a given speed to remain aloft.
5.2. Conflict prediction simulation results
Based on the trajectories of aircraft moving through the study volume the traffic flow measures are calculated. Each simulation is
run for 5,000 s of simulated time, or 83 min. The traffic flow measures are aggregated into sets of 300 s, or 5 min. Multiple simulations
were run for each parameter level to better confirm the results and reduce the effects of randomness built into vehicle starting times,
origins, destinations, and headings.
The conflict prediction model was tested with conflict resolution turned off, but with conflict detection in order to count the
number of conflicts. Simulations were performed across multiple levels for each parameter. In addition to conflict counts the density of
vehicles in the study airspace was counted and accumulated in the time windows. The conflict-density relationship was then explored
by plotting as individual points the average vehicle density in the study airspace during the time window with the number of conflicts
occurring in the study airspace. These conflict-density plots are shown in Figs. 12 and 13 for multiple parameter levels, with each level
colored differently. The theoretical gas law conflict prediction model developed in Eq. (4) was used to independently create predictions
of the number of conflicts for the given vehicle density and parameters. The curve of these predictions was overlaid on the scatter plots
of simulated data to compare the two. As demonstrated in Figs. 12 and 13 the theoretical model does an excellent job of both capturing
the shape of the conflict-density relationship and predicting the number of conflicts. The R-square values found at each level are
generally high. These results indicate an excellent fit for the model with the simulated results and validate the conflict prediction
model for unstructured airspace. The plots also confirm that at similar density levels the conflict rates for larger spacing requirements
and faster speeds are higher.
5.3. Calibrating the average speed loss per conflict parameters
To apply the actual speed-density and effective speed-density models, first average speed loss per conflict parameters must be found
that reflect the sensitivity of average aircraft speeds to conflicts in the airspace. These average speed loss per conflict parameters vary
according to operational details that affect how conflict resolution arrives at trajectory detours, including aircraft density, and sep
aration volumes and maximum aircraft speeds. Models that relate the parameters and contexts are calibrated from observed results
such that the parameters can be predicted across a range of aircraft densities. Eq. (16) shows the model used for determining each of
Fig. 12. Comparison of simulated results (dots) with the theoretical predictions (curves) for multiple spacing requirement levels of the conflict-
density relationship.
13
Fig. 13. Comparison of simulated results (dots) with the theoretical predictions (curve) for multiple maximum speed levels of the conflict-density
relationship.
the average speed loss per conflict parameters for actual and effective speeds, with the constant value calibrated separately for each.
The constant value primarily captures the impact of the conflict resolution system used, and the average relative heading between
aircraft, among other context-specific considerations. The model uses a positive linear relationship between average speed loss per
conflict and density as well as maximum aircraft speed. The spacing requirement (a radius) is squared in value to reflect the impact of
the separation area that increases the average speed loss per conflict.
a, b = constant*k*D2sep *vmax (16)
Fig. 14 demonstrates how the average speed losses per conflict parameters vary by density and operating context. Figures for each
the average actual speed loss per conflict and average effective speed loss per conflict, varying by density, separation radii and
maximum aircraft speeds are provided. Individual points of the parameters (representing aggregated traffic results from a study
volume and time window) will vary according to the specifics of the conflicts encountered in the window. However, the overall
patterns (the model predictions are shown as lines) illustrate that higher aircraft densities, separation volumes, and maximum aircraft
speeds all increase the average speed loss per conflict.
Simulated results from low to high densities corroborate that the parameters change in different operating conditions and demand
levels according to predictable patterns. Considering the predictable nature of the parameters according to the operating context, by
calibrating a model of the parameters using a test set of observed data it then becomes possible to predict the parameters for the whole
possible range of values. These parameters are then applied within the traffic flow models. While the results herein include an entire
range of observed data from low to high density operations to validate the concept, the parameters can be calibrated from low density
operations alone. This approach to calibrating key model parameters at low density has been used in previous work on conflict
prediction and early approaches to air traffic flow models (Jardin, 2004, Sunil et al., 2018a, Tereschenko et al., 2020).
5.4. Traffic flow model simulation results
The remaining three theoretical traffic flow models were each tested against simulated results to validate their relationships. The
objectives of this paper look to develop models for a range of vehicle parameters, and therefore the models were tested against a range
of vehicle parameters. For each of the vehicle spacing requirements and maximum vehicle speeds five levels were chosen and multiple
simulations conducted of the air traffic flow for each parameter level to gather simulated results. These simulation runs were con
ducted with the same setup as above and with conflict resolution turned on, such that conflict resolution will affect vehicle speeds and
traffic flow. The simulated results are again plotted using a scatter plot technique, with each point representing the average traffic
measure values of the study airspace during one time window (5 min) of the simulation. The theoretical models are then used to predict
values using only the density and vehicle parameters. The predicted values are shown as curves overlaid on each figure. As multiple
parameter levels are plotted simultaneously, each parameter level is plotted in a different color, with separate R-squared values for
only results pertaining to that parameter level calculated and displayed.
The average actual speed-density model developed in Eq. (7) is first tested. The results are displayed in Fig. 15 (testing various
spacing requirements) and Fig. 16 (testing various maximum speeds). From the figures we can see that the theoretical model (rep
resented by the curves) generally does an excellent job of predicting the average actual vehicle speed in the study airspace. The R-
squared values generally indicate an excellent fit for the theoretical model compared to the simulated results. The results also confirm
14
Fig. 14. (a-d). Comparison of simulated results (dots) with the theoretical predictions (curves) for the average speed losses per conflict. (a) Average
actual speed loss per conflict vs density, varied by spacing requirement. (b) Average actual speed loss per conflict vs density, varied by maximum
vehicle speed. (c) Average effective speed loss per conflict vs density, varied by spacing requirement. (d) Average effective speed loss per conflict vs
density, varied by maximum vehicle speed.
the theoretical finding that for multi-dimensional traffic flow the speed-density relationship is not linear, but instead curved with speed
reductions at low density not as large as at higher densities. Also, the results in Fig. 15 show that larger spacing requirements will
create sharper declines in average vehicle speed from density. For different maximum speeds in Fig. 16 the traffic flows with higher
maximum speeds will allow for faster speeds with sharper speed reductions due to density than lower maximum speeds.
Next the theoretical model from Eq. (10) relating the average effective speed to vehicle density was tested based on the same
simulated results. The simulated results and theoretical predictions are plotted in Figs. 17 and 18. As with the actual speed results, the
effective speed-density results demonstrate a generally good fit for the theoretical model with the simulated results. The prediction
curves closely follow the simulated results points for each of the parameter levels. The R-squared values are also generally high,
indicating a good fit. The results in Fig. 17 confirm the theoretical finding that declines in effective speed due to density are larger for
traffic flows with larger spacing requirements than smaller ones. In Fig. 18 the results demonstrate that traffic flows with higher
maximum speeds do have higher effective speeds at low density, but that these effective speeds converge at higher densities.
Lastly the model of the effective flow rate (or throughput) and density from Eq. (14) was tested. The results for various spacing
levels are shown in Fig. 19 and various maximum speed levels are shown in Fig. 20. In each of the results we find the pattern of the
fundamental diagram, in which flow rate increases proportionally with density at low densities before having declining returns to
15
Fig. 15. Comparison of simulated results (dots) with the theoretical predictions (curves) for the actual speed-density relationship.
Fig. 16. Comparison of simulated results (dots) with the theoretical predictions (curves) for the actual speed-density relationship.
density at higher densities. The presence of critical densities where the flow rate has reached a maximum are also notable. The results
in both figures show a good fit with generally high R-squared values for the theoretical predictions and the simulated results. This
finding supports the validation of the traffic flow models developed in this paper. It is also demonstrated that for traffic flows with
larger spacing requirements there will be a lower critical density and maximum flow rate than traffic flows with smaller spacing
requirements. In Fig. 20 it can be seen that higher maximum speeds allow for higher flow rates than lower maximum speeds, albeit
with the challenge of a lower critical density.
Throughout the validation process each of the theoretical models demonstrated a strong ability to predict the traffic flow values
produced via simulated results. These results support the validation of the theoretical models and the findings based on the models.
They also demonstrate that the models could be used to make predictions based on vehicle densities and parameters for other contexts
in which unstructured airspace is used.
16
Fig. 17. Comparison of simulated results (dots) with the theoretical predictions (curves) for the effective speed-density relationship.
Fig. 18. Comparison of simulated results (dots) with the theoretical predictions (curves) for the effective speed-density relationship.
6. Implications of traffic flow results
The primary contribution of this paper is the development of macroscopic air traffic flow models. The models extend traffic flow
concepts into multiple dimensions, creating new relationships and informing on the behavior of air traffic flow. Using these air traffic
flow models, future traffic flow conditions can be predicted on the basis of known operational parameters and demand on the airspace.
The average aircraft speeds are useful for predicting trip travel times and energy needs, while the predicted throughput in the airspace
points towards the efficiency of the network and its maximum capacity. Validating the theoretical models is key to applying the
findings of the theoretical models. The simulated results across a number of runs and different parameter levels supported the
theoretical findings by demonstrating similar patterns and a good overall fit of the theoretical model with the simulated results.
There are several useful applications that the authors anticipate. The air traffic flow models both contribute valuable insights for
traffic flow behavior of air traffic and managing air traffic, which are further outlined below. It is possible to use the theoretical models
for air traffic flow predictions, such as predicting the conflict rate, average aircraft speeds, throughput, or total airspace capacity. Each
17
Fig. 19. Comparison of simulated results (dots) with the theoretical predictions (curves) for the effective flow rate-density relationship.
Fig. 20. Comparison of simulated results (dots) with the theoretical predictions (curves) for the effective flow rate-density relationship.
of these are key outputs of specific interest to AAM researchers, planners, and operators, who can quickly experiment with various
policy changes and observe the predicted traffic flow behavior. With only limited data the traffic flow models can be calibrated and set
up, to provide predictive capabilities without requiring extensive results from a full range of demand levels, which is much more
expensive. Models with predictive power for traffic flow can also be incorporated into air traffic flow management systems, giving the
systems a better understanding of the traffic flow conditions and enabling more accurate traffic flow management solutions. The air
traffic flow models can also be extended outside of the unstructured case to cover more structured airspace, such as corridors. By
adjusting the term for the relative speed between aircraft (in Eq. (3), it becomes possible to observe aircraft moving within similar
directions.
There are several important conceptual implications that arise from the models. The first implication is that aircraft density
generates conflicts and congests the airspace. This finding is consistent with previous literature (Sunil et al., 2018a, Tereschenko et al.,
2020) which has directly related aircraft density and the number of conflicts in airspace. Airspace is shown to be congestible when
under high demands, due to the generation of and impact from conflicts. The shape and scale of the conflict-density relationship
defines the rate of congestion along with the average speed loss per conflict. Eq. (4) and Figs. 3 and 4 demonstrated the power
18
relationship between vehicle density and conflict rate, with conflict rate increasing much more quickly at higher densities than lower
densities.
The second implication is that as conflicts increase, they reduce average vehicle speeds and impact throughput, until a maximum
throughput (or capacity of the airspace) is reached. This key concept defines how congestion affects the air traffic flow, and how users
of the system will experience it. The multi-dimensionality of the traffic causes the impact of conflicts on the traffic to vary. Conse
quently, different traffic flow management policies may be effective for different demand levels. For example, conflicts have a rela
tively low impact on traffic at low demand levels, so traffic flow management policies that offer more routing efficiency at the expense
of a higher conflict rate may be relatively worth the trade-off. The traffic flow models developed within begin to illustrate some of these
strategies and can be used in experimenting with others.
One important outcome variable from the model is airspace capacity, which is critical for determining how networks and oper
ations are conducted. The traffic flow models identify maximum achievable throughputs within an airspace, and also the critical
densities at which those capacities occur. Various operational factors such as demand levels, vehicle spacing requirements, speeds, and
others can have sizable impacts on the total airspace capacity, and also separate impacts on the critical density. By setting policies and
managing the operations of airspace it is also possible to manipulate to some extent the achievable airspace capacity and critical
density to suit operational needs.
Several insights come from the role of operating parameters in traffic flow behavior. Larger spacing requirements between aircraft
were found to increase congestion within airspace. The theoretical results considered varying spacing requirements and demonstrated:
1) that larger spacing requirements have a higher conflict rate at any specified density, 2) that larger spacing requirements lead to
lower average speeds and sharper declines in speed, and 3) that larger spacing requirements do not impact the throughput of airspace
at low densities, but at higher densities have lower throughput and lower critical densities. In total these results indicate that operators
and planners should in general seek to minimize the spacing requirements (or aircraft types with large spacing requirements) while
ensuring safety, as the benefits include fewer conflicts, lower travel times, and higher flow rates. However, if larger spacing re
quirements are required, then the resultant airspace will have a lower capacity and congest more quickly, and should be managed
accordingly.
Higher maximum vehicle speeds were also shown to congest airspace more readily. The impact of conflicts on the average aircraft
speed is larger for systems with higher maximum vehicles speeds in the medium-density range. This finding makes higher maximum
vehicle speeds relatively less-efficient at high densities. Higher maximum speed scenarios had higher throughputs under low demand,
but reached a critical density lower than the critical densities of lower maximum speed scenarios. Taken together these findings reflect
that higher maximum speeds may generally be better for lower travel times and higher throughputs. Yet, in high density airspaces the
difference in throughput may be small and slower maximum speeds can achieve the same throughputs with fewer conflicts, making
them more efficient. Under such conditions operators may restrict aircraft speeds to decrease conflicts and increase the critical density
to prevent traffic breakdown or help in recovery after traffic breakdown.
7. Conclusion
This paper has developed macroscopic air traffic flow models that relate vehicle density and spacing and speed parameters to traffic
flow measures for unstructured airspace in the AAM context. The paper recognized the key role of conflicts in air traffic flow and
related vehicle density to the number of conflicts in airspace using a gas law conflict prediction method. The accuracy of the theoretical
models were tested and validated with simulated results for a number of parameter levels. The theoretical and simulated findings were
used to provide operational and policy insights for AAM operators, planners, and modelers. While other works have studied air traffic
flow and identified factors that impact air traffic flow, this work goes further by creating defined traffic flow models that encompass all
key traffic flow variables, generate insights and can be applied for predicting air traffic flow conditions. Using the models created it is
possible for AAM researchers, planners and operators to model key outcomes such as vehicle travel times and airspace capacity.
There are several key insights offered by this work. One is the critical role of aircraft density in air traffic flow. The conflict pre
diction model demonstrated a power rule for the number of conflicts based on density, indicating that at higher densities the conflict
rate increases much more quickly. Density was also shown to decrease actual vehicle speeds, effective speeds, and throughput due to
conflict resolutions. Based on this finding AAM operators and planners will need to closely manage the airspace density of AAM ve
hicles to avoid large numbers of conflicts simultaneously and maintain acceptable travel times and throughputs. Another insight is the
importance of the aircraft spacing requirement. The theoretical models found and the simulated results supported that larger spacing
requirements have higher conflict rates, sharper reductions in speed from density, and lower critical densities than smaller spacing
requirements. Therefore, AAM operators and planners may wish to find the minimal spacing requirements that still allow for safe
operations. Lastly, aircraft speeds were shown to have a nuanced role in air traffic flow. Higher maximum vehicle speeds related to
higher average vehicle speeds. However, they also created a higher conflict rate, lower critical density, and showed a sharper reduction
in vehicle speeds that led to a convergence of speeds at high density. From these findings AAM operators and planners might conclude
that vehicle speed restrictions are appropriate in high density scenarios to avoid large numbers of conflicts and a lower critical density.
There were several limitations in the methodology of this paper. The first is that while the theoretical models went farther than
previous work in reducing the reliance on calibrated values, there is still a need to calibrate the average speed loss per conflict value. It
is possible though to calibrate this value as a function of parameters such as spacing and maximum speed to reduce the number of
contexts requiring calibration. The models in this paper also only apply to unstructured airspace, with the assumption of random
headings for aircraft. However, it is possible to adapt the presented models for structured airspaces and corridors by adjusting the
relative speed parameter between aircraft (Eq. (3). The means of producing simulated results was also limited to only two-dimensions.
19
This simplification was made to approximate flying conditions in which an aircraft largely avoids expensive altitude changes and
reduce computation times, however it does not reflect the ability for aircraft to make altitude changes. The simulator used is also a
simple AAM simulator that does not account for the physics and accelerations of aircraft, reducing the accuracy of individual aircraft
trajectories. However, the broader trends of air traffic flow are still applicable to this situation.
Future work could extend this model to other forms of airspace, specifically by testing more restricted airspace than unstructured or
reflecting varying altitude levels of vehicles, as recently addressed in (Safadi et al., 2023; Cummings and Mahmassani, 2023a, b). The
theoretical models could also be deployed in other AAM models as an approximation of air traffic flow trends. Models that estimate and
study travel times in the AAM network, network capacity, air traffic control systems, and optimization of trip planning and vehicle
deployment could all benefit from the better approximations of vehicle speeds, effective speeds, conflicts, and throughput offered by
the models in this paper.
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 work is based on research funded in part by a Dissertation Year Fellowship from the Northwestern University Transportation
Center to the first author.
References
Alonso, J., Arneson, H., Melton, J., Vegh, M., Walker, C., Young, L., 2017. System-of-Systems Considerations in the Notional Development of a Metropolitan Aerial
Transportation System. National Aeronautics and Space Administration. NASA/TM—2017–218356.
Bayen, A., Raffard, R., Tomlin, C., 2006. Adjoint-based control of a new Eulerian network model for air traffic flow. IEEE Trans. Control Syst. Technol. 14 (5).
Bulusu, V., 2019. Urban Air Mobility: Deconstructing the Next Revolution in Urban Transportation Feasibility, Capacity and Productivity. University of California,
Berkeley. PhD Dissertation,.
Cummings, C., Mahmassani, H., 2021. Emergence of 4-D system fundamental diagram in urban air mobility traffic flow. Transp. Res. Rec. 2675 (11), 841–850.
Cummings, C., Mahmassani, H., 2023b. Comparing urban air mobility network airspaces: Experiments and insights. Transp. Res. Rec. https://doi.org/10.1177/
03611981231185146.
Cummings, C., Mahmassani, H.S., 2023a. Measuring the impact of airspace restrictions on air traffic flow using four-dimensional system fundamental diagrams for
urban air mobility. Transp. Res. Rec. 2677 (1), 1012–1026.
Edie, L.C., 1965. Discussion of Traffic Stream Measurements and Definitions. In: Almond, J. (Ed.), 2nd Symposium on the Theory of Traffic Flow; Summary of
Communications. Organization for Economic Co-operation and Development, Paris, pp. 139–154.
Geister, D., Korn, B. (2018) Density based management concept for urban air traffic. 2018 IEEE/AIAA 37th Digital Avionics Systems Conference (DASC) 1-9.
Greenshields, B., Bibbins, J., Channing, W., Miller, H., 1935. A study of traffic capacity. Highway Research Board Vol 14. Part 1, 448–477.
Haddad, J., Mirkin, B., Assor, K., 2021. Traffic flow modeling and feedback control for future low-altitude air city transport: An MFD-based approach. Transportation
Research Part c: Emerging Technologies 133.
Jardin, M. (2004) Air Traffic Conflict Models. American Institute of Aeronautics and Astronautics 4th ATIO Forum.
Johari, M., Keyvan-Ekbatani, M., Leclercq, L., Ngoduy, D., Mahmassani, H.S., 2021. Macroscopic network-level traffic models: Bridging fifty years of development
toward the next era. Transportation Research Part c: Emerging Technologies 131, 103334.
Josselson, R., 1997. Introduction to 6-DOF simulation of air vehicles. Visual Solutions Inc.
Liu, Y., 2018. A Progression Motion-Planning Algorithm and Traffic Flow Analysis for High-Density 2D Traffic. Transp. Sci. 53 (6), 1501–1525.
Mao, Z., Feron, E., Bilimoria, K., 2001. Stability and performance of intersecting aircraft flows under decentralized conflict avoidance rules. IEEE Trans. Intelligent
Transportation Systems 2, 101–109.
Menon, P., Sweriduk, G., Lam, T., Diaz, G., 2006. Computer-aided Eulerian air traffic flow modeling and predictive control. J. Guid. Control Dynam. 29 (1), 12–19.
Mitici, M., Blom, H., 2018. Mathematical models for air traffic conflict and collision probability estimation. IEEE Trans. Intell. Transp. Syst. 20 (3), 1052–1068.
Neto, E., Baum, D., Almeida, J., Camargo, J., Cugnasca, P., 2019. Trajectory-Based Urban Air Mobility (UAM) Operations Simulator (TUS). Polytechnic School or
University of Sao Paulo.
Pallotino, L., Feron, E., Bicchi, A., 2002. Conflict resolution problems for air traffic management systems solved with mixed integer programming. IEEE Trans. Intell.
Transp. Syst. 3 (1), 3–11.
Ribeiro, M., Ellerbroek, J., Hoekstra, J., 2020. Review of Conflict Resolution Methods for Manned and Unmanned Aviation. Aerospace 7 (6), 79.
Saberi, M., Mahmassani, H., 2014. Exploring areawide dynamics of pedestrian crowds: Three-dimensional approach. Transportation Research Record: Journal of the
Transportation Research Board 2421, 31–40. https://doi.org/10.3141/2421-04.
Saberi, M., Mahmassani, H., Hou, T., Zockaie, A., 2014. Estimating network fundamental diagram using three-dimensional vehicle trajectories. Transportation
Research Record: Journal of the Transportation Research Board 2422.
Safadi, Y., Geroliminis, N., Haddad, J. (2023) Traffic demand management for low altitude air city transport based on macroscopic fundamental diagram. In
Proceedings of the 102nd Transportation Research Board Annual Meeting, Washington, D.C.
Safadi, Y., Fu, R., Quan, Q., Haddad, J., 2023. Macroscopic fundamental diagrams for low-altitude air city transport. Transp. Res. Part C: Emerg. Technol. 152,
104141.
Shah, A. P., Pritchett, A., Feigh, K. M., Kalaver, S. A., Jadhav, A., Corker, K. M., Bea, R. C. (2005) Analyzing air traffic management systems using agent-based
modeling and simulation. In Proceedings of the 6th USA/Europe Air Traffic Management Research and Development Seminar, 661-671.
Stroe, G., Andrei, I.C., 2016. Automated conflict resolution in air traffic management. International Conference of Aerospace Sciences.
Sun, J., Hoekstra, J., Ellerbroek, J., 2020. OpenAP: An open-source aircraft performance model for air transporation studies and simulations. Aerospace 7 (8), 104.
Sunil, E., Ellerbroek, J., Hoekstra, J., Maas, J., 2018b. Three-dimensional conflict count models for unstructured and layered airspace designs. Transportation
Research Part c: Emerging Technologies 95, 295–319.
Sunil, E., Ellerbroek, J., Hoekstra, J. (2018) CAMDA: Capacity Assessment method for Decentralized Air Traffic Control. International Conference for Research in Air
Transportation.
20
Tereschenko, I., Hanson, M., Zou, B. (2020) Macroscopic fundamental diagram for air traffic: Preliminary theoretic results and simulation findings. International
Conference for Research in Air Transportation (virtual).
Work, D., Bayen, A., 2008. Convex formulations of air traffic flow optimization problems. Proc. IEEE 96 (12), 2096–2112.
Xue, M., Rios, J., Silva, J., Ishihara, A., Zhu, Z., 2018. Fe3: An evaluation tool for low-altitude air traffic operations. American Institute of Aeronautics and Astronautics
ATIO Forum.
21

Transportation Research Part C: Christopher Cummings, Hani Mahmassani

Uploaded by

Copyright:

Available Formats

Transportation Research Part C: Christopher Cummings, Hani Mahmassani

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Transportation Research Part C: Christopher Cummings, Hani Mahmassani

Uploaded by

Copyright:

Available Formats

Downloaded from https://iranpaper.

Transportation Research Part C 159 (2024) 104467

Contents lists available at ScienceDirect

Transportation Research Part C

Airspace Congestion, flow Relations, and 4-D fundamental

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

3. Methodology: Theoretical models

3.1. Definition of variables

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

3.2. Three key traffic flow concepts

3.3. Gas law conflict prediction model

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

Fig. 2. Depiction of the airspace each aircraft searches for conflicts.

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

3.4. Theoretical Speed-Density models

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

3.5. Theoretical flow rate – Density model

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

implications of these relationships.

4. Patterns from traffic flow models

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

5. Simulation model results

5.1. Simulation setup

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

5.2. Conflict prediction simulation results

5.3. Calibrating the average speed loss per conflict parameters

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

a, b = constant*k*D2sep *vmax (16)

5.4. Traffic flow model simulation results

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

6. Implications of traffic flow results

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

Declaration of competing interest

C. Cummings and H. Mahmassani Transportation Research Part C 159 (2024) 104467

You might also like

a, b = constantkD2sep *vmax (16)