This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2021.DOI

Multi-objective takeoff time optimization

using cellular automaton-based
simulator
KATSUHIRO SEKINE1 , TOMOAKI TATSUKAWA2 , ERI ITOH3 , AND KOZO FUJII4
1
Tokyo University of Science, Katsushika, Tokyo, 125-8585, Japan (e-mail: [email protected])
2
Tokyo University of Science, Katsushika, Tokyo, 125-8585, Japan (e-mail: [email protected])
3
Air Traffic Management Department National Institute of Maritime, Port and Aviation Technology Electronic Navigation Research Institute, Chofu, Tokyo
182–0012, Japan (e-mail: [email protected])
4
Tokyo University of Science, Katsushika, Tokyo, 125-8585, Japan (e-mail: [email protected])
Corresponding author: Tomoaki Tatsukawa (e-mail: [email protected]).
This research was supported by MEXT, the Ministry of Education, Culture, Sports, Science and Technology, as “Exploratory Challenge 2
(ID : hp190163) on Post-K computer” (Construction of Models for Interaction Among Multiple Socioeconomic Phenomena, sub-issue:
Model Development and its Applications for Enabling Robust and Optimized Social Transportation Systems.

ABSTRACT Although schedule design has further potential to reduce airline operation costs and flight
delay, the effectiveness of the globally optimal schedule design integrating air traffic flow has not been
discussed thus far. This paper presents a global multi-objective takeoff time optimization to design efficient
flight schedules that lead to minimal congestion and provide sufficient resilience against traffic problems.
NSGA-II is adopted as the multi-objective optimization technique in this study. The objective functions
include minimization of the total arrival delay and total fuel consumption because these are key performance
indicators of air traffic management (ATM). The design variable used in this study is the takeoff time offset
of each flight landing at the Tokyo International Airport. 607 design variables were used in this study. The
range of the design variables was ±300 s to investigate the effect of a minor variation in the takeoff time. A
cellular automaton-based model was utilized to simulate the interaction of the flights with each other. The
results of the simulations demonstrated that the obtained optimal solutions could drastically reduce the total
arrival delay and total fuel consumption by 1500 min and 80 tons, respectively. The spacing adjustments
of one of the optimum flight schedules, in comparison to the original flight schedule, were reduced by
80% in the en-route and terminal airspaces. Additional analyses suggest that it is preferable to have longer
takeoff time intervals for flights originating from the same point during congestion hours than those during
non-congestion hours. This indicates that the optimization of ground movements in airports improves the
efficiency of air traffic operations.

INDEX TERMS Air traffic control, air traffic management, cellular automaton, flight scheduling, ground
holding program, NSGA-II, schedule design

I. INTRODUCTION TBO requires collaborations among the industries, academia

and governments, which was encouraged by the promotion
IR traffic demand has rapidly grown across the world
A in previous decades [1]. Despite the sharp reduction in
revenue passenger kilometers due to COVID-19, the air traf-
of industrial and academic techniques in local areas, such as
the Next-Generation Air Transportation System (NextGen)
[4] in the U.S., Single European Sky Air Traffic Manage-
fic demand is expected to recover in the next 4 years, albeit ment (SESAR) [5] in European countries, and Collaborative
with uncertainty [2]. The International Civil Aviation Orga- Actions for Renovation of Air Traffic Systems (CARATS) [6]
nization (ICAO) has proposed a unique operational concept in Japan. Various studies have been conducted as part of these
for air traffic management (ATM), so-called trajectory-based programs in collaboration with ATM stakeholders to enhance
operations (TBO), to meet the increasing demand for air traf- ATM performance.
fic [3]. This involves the management of the trajectories of all
flights from their departure to arrival. The implementation of Air traffic flow management (ATFM) is one of the most

VOLUME xx, 2021 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

crucial operations in the ATM framework. ATFM generally the various fare classes [24]. Kenan et al. (2018) developed an
involves strategic regulation of flights to capacity-constrained integrated scheduling model, including schedule design, fleet
airports or flights that will pass through capacity-constrained assignment, and aircraft routing, in consideration of delay
airspace areas at a national scale. Traffic regulation at airports propagation and deadhead flights under the uncertainty in
with limited capacity has been widely discussed in previous demand [25]. Wei et al. (2019) proposed an integrated op-
studies and is commonly referred to as the ground holding timization approach for schedule design and fleet assignment
problem (GHP). Several GHP mathematical programming focusing on the potential revenue obtained from the customer
methods have been developed to determine the optimal al- satisfaction, attractiveness of each itinerary, and passenger
location of ground delays and minimize the expected cost of repurchasing determinants [26].
the overall delays [7] [8] [9] [10] [11] [12] [13]. Existing These robust flight scheduling studies, however, ap-
models are able to decide the allocation of ground delays plied heuristic optimization approaches, although the flight
once at the beginning of the planning period, assuming that scheduling problem is NP-Hard in nature. This suggests that
the all airspace capacities are known [7] [8]. Those models the obtained flight schedules are not necessarily globally
have been updated to incorporate the uncertainties in the optimal. Further, assuming a linearity, the objective function
airspace capacity [9] [11] [13], and to develop dynamic de- formulated with a single weighted linear sum was a min-
cision making by utilizing the updated capacity information imization of the overall costs and/or maximization of the
during the GHP planning period [10] [12]. Machine learning overall profit. This indicates that the relationships among the
(ML) techniques have recently been implemented by using terms in the function had not been quantitatively investigated.
historical data of actual operations to improve the accuracy In the field of ATM research, including flight scheduling,
in the ground delay prediction [14] [15] [16] [17]. These there have been few studies that conducted multi-objective
studies have investigated the effect of various factors, related optimization with meta-heuristic approaches and analyzed
to the traffic demand and weather conditions on the GHP, the obtained optimal solutions in detail with the data-mining
understood the mechanism underlying GHP, and predicted approaches [27] [28] [29], although significant design knowl-
necessary ground delays. Those mathematical and machine edge can be potentially extracted from those optimal or sub-
learning models have focused on delaying the takeoff time optimal solutions. Furthermore, optimal schedules obtained
using a reactive approach. in the related works were not created in consideration of the
Developing optimal flight schedules beforehand to adjust weather conditions and air traffic flow although they have a
the takeoff time falls under a different paradigm. Flight large impact on flight delay, which is one of the four dis-
scheduling traditionally comprises of the following four ruption types [19]. The other three disruptions, namely, flight
stages: (1) schedule design, (2) fleet assignment (3) route cancellation, fleet availability, and airport disruption, occur as
construction, and (4) crew pairing [18]. These stages are irregular events, complicating the mitigation of thereof prob-
closely associated with each other, which indicates that dis- abilities, and are thereby often managed through recovery-
turbances in any of the stages often affects the other stages based approaches [30] [31] [19]. In contrast, flight delays,
negatively. As noted in [19], disruptions are mainly catego- whose causes vary depending on the locations [32], could
rized into flight delays, flight cancellation, fleet availability, relatively be reduced using a proactive approach such as
and airport disruption. To minimize the overall probability simulation-based approaches and delay analyses, including
of the aforementioned disruptions, several researchers have the aforementioned GHP studies. Therefore, it would be of
developed robust flight scheduling models integrating two interest (1) to evaluate the effectiveness of the multi-objective
or three of the flight scheduling stages. Sherali et al. (2010) flight schedule global optimization combined with air traffic
proposed an integrated fleet assignment and schedule design flow simulation and (2) to analyze the optimal solutions
model in which the characteristics of passenger demand to enhance our understanding of the factors contributing to
were considered [20]. Jiang and Barnhart (2013) developed a ground holding operation and flight scheduling.
robust schedule design model, which dynamically scheduled This study aims to develop efficient and environmen-
the departure time in response to the fluctuations in demand tally friendly flight schedules with minimal congestion. We
in a de-banking environment [21]. Pita et al. (2014) proposed utilized a multi-objective optimization technique based on
the fleet assignment and schedule design model to minimize NSGA-II [33]. NSGA-II is one of the most used genetic
the overall costs in air transportation network, and they algorithms (GA). The objective functions include (1) mini-
implemented network welfare analysis [22]. Cadarso and mization of the total arrival delay and (2) minimization of the
Celis (2017) presented an airline planning model integrating total fuel consumption, which correspond with the demands
schedule design, fleet assignment, and passenger use, consid- of the customers and airlines, respectively. The difference
ering the stochastic demand value and uncertain operational between the scheduled and actual takeoff time of each flight
conditions, to minimize the number of passengers who miss arriving at Tokyo International Airport (RJTT) was used as
the connecting flight [23]. Faust et al. (2017) proposed the the design variable. The number of design variables used
schedule design and aircraft maintenance routing model to in this study was 607. The design variables were varied by
maximize the profit of medium-sized point-to-point airlines ±300 s to investigate the effect of the minute takeoff time
with a homogeneous fleet while considering the demand for difference in consideration of takeoff time error with a few
2 VOLUME xx, 2021

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

minutes in the flight schedules. Although true optimality

is not guaranteed as the obtained sub-optimal solutions are
computed based dependent on the GA variant, this study
emphasizes the importance of analyzing the obtained sub-
optimal solutions for efficient schedule design through the
data-mining approaches. A cellular automaton-based simula-
tor was adopted to simulate air traffic flow because it consists
of various complex rules and is heavily dependent on the
FIGURE 1. Basic Rules
installed equipment and weather.
The remainder of this paper is organized as follows. Sec-
tion II provides a brief overview of the simulator used in this Several CA models [54] [55] [56] [57] [58] [59] [60] have
study. The existing simulator, which has been constructed been proposed to analyze air traffic flow. Kim et al. [54]
by the author’s research group, is based on the cellular au- proposed a two-dimensional CA model to demonstrate an air-
tomaton model. Section III describes the problem statement borne self-separation system by describing aircraft that move
for the multi-objective optimization of the schedule design. in all directions within adjacent airspaces. Yu et al. [55] and
Section IV presents the computational configuration of the He et al. [56] utilized a one-dimensional CA model that can
simulator. The optimization framework used to solve the simulate the aircraft landing process for scheduling aircraft
multi-objective schedule design problem is mentioned in landing time in real time. Wang et al. [57] and Lim et al. [58]
Section II and Section III. Section V analyzes and discusses adopted a two-dimensional CA model for en-route airspace
the results of a case study wherein the proposed multi- by combining optimization techniques, to search for flight
objective optimization model is adopted for inbound traffic paths that avoid prohibited airspace, restricted airspace, and
at the RJTT. Finally, our conclusions are then presented in dangerous airspace. Enayatollahi et al. [59] [60] constructed
Section VI. a two-dimensional CA model considering a standard terminal
arrival route (STAR) to investigate the impact of weather on
II. MODEL DESCRIPTION AND FORMULATION delays and to optimize the flight path following performance-
Air traffic flow has been simulated previously in demand based navigation (PBN). Most studies have not focused on
capacity balancing (DCB) studies, congestion analysis, and constructing a CA model in the global airspace, such as
other ATM-related studies, using mathematical models, such Japan and Europe, but in terminal airspace or specific en-
as queue [34] [35] [36] [37], the differential equation [38] route airspace.
[39] [40] [41] [42] [43], and network [44] [45] [46] [47]. To address this gap, we propose a one-dimensional CA
Many strategic ATM concepts based on these models have model with step back movement (SBCA) that was con-
been proposed, and these concepts enable a low computa- structed considering Tokyo International Airport (RJTT) [61]
tional cost. However, mathematical and theoretical models and Japan [62]. Our objective was to contribute to the pro-
are not necessarily able to explain air traffic flow precisely duction of a fast-time simulation and optimization toolset
because they are based several complex rules and are strongly in which ATM stakeholders can quickly and easily change
affected by aircraft facilities and the weather. Agent-based airspace and routes for verifying new ATM concepts. There-
modelling is a different approach that has also been utilized fore, the unique characteristics of SBCA is its ability to
in the field of ATM by several researchers [48] [49] [50] [51] model air traffic flow with only a few parameters, simple
[29]. Although multiagent models can describe phenomena input files, and simpler rules than those followed in previous
in detail, it is difficult to develop a highly accurate model related studies. Our models [61] [62] were updated according
due to the high degree of freedom required. The cellular to the aircraft characteristics using the Base of Aircraft Data
automaton model is similar to the multiagent-based model, (BADA) [63] and data on international flights arriving at
but is simpler. As mentioned by Sun et al. [52], the cellular RJTT.
automaton model is computationally friendly compared with
the Eulerian flow theory and PDE models, which are theoret- B. BASIC RULE
ical models. Thus, we focused on a cellular automaton model SBCA consists of three basic rules that utilize two parame-
that can simulate complex phenomena by following simple ters, d BD, and aircraft speed s. d is the minimum in-trial
rules. separation between aircraft, and BD is the step-back dis-
tance. s [cells/timestep] is the ground speed of each aircraft.
A. CELLULAR AUTOMATON s is calculated following the BADA model [63] according to
A cellular automaton model (CA) is a discrete computational the status of each aircraft, such as climb, cruise, and descent.
model [53] that consists of a regular grid of cells, with each Then, air speed closest to the position of each aircraft is
cell existing in one of the finite number of states, such as 0 added to the s calculated based on the BADA model. An
and 1. The state of each cell is updated after each discrete aircraft travels according to the rules listed below (Fig. 1).
time interval in accordance with simple rules. 1) If there is no aircraft in the front for a distance of d
VOLUME xx, 2021 3

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

To North

Arrivals
Route Information
Flight path toward runway 34R at RJTT
Departures Flight path toward runway 34L at RJTT
34R
16L
16R

34R
34L 05 From North 16L
22
16R 40
To North

Latitude [deg]
To South

Arrivals
From South To South Departures

(a) North wind operation (b) South wind operation

FIGURE 2. Runway configuration of departures (blue lines) and arrivals (red

lines) at RJTT 30

cells, then the aircraft moves forward by s cells (step

forward).
2) If there are one or more aircraft in the front within a
distance of d cells, and there is no aircraft in the rear 120 130 140 150
Longitude [deg]
for a distance of BD cells, then the aircraft moves
backward by BD cells (step back). FIGURE 3. Flight path to Tokyo International Airport (RJTT)
3) If there are one or more aircraft in the front within a
distance of d cells and in the rear within a distance
of BD cells, then the aircraft remains in its current runway 16R in combination with the arrival aircraft.
position (stop). Climate and seasonal fluctuations change runway configu-
Rules (2) and (3) correspond to spacing adjustments, such as rations. According to the operational records between 6:00
vectoring and holding. Although Rule (3) temporarily allows AM and 11:00 PM for the years from 2016 to 2018, the
for the introduction of a conflict, by violating the separation northerly wind operation accounted for 70% of the total
minima d, the aircraft to which Rule (3) is applied will stop operations, indicating a larger rate over execution of the
by Rule (2) or step back by Rule (3) in the next time step, southerly wind operation. Based on these statistics, this paper
thereby solving the conflict immediately. focuses on the northerly wind operation.
Figure 3 shows the flight paths to RJTT, which is indicated
C. CASE STUDY DATA DESCRIPTION - TOKYO by a black square in Fig. 3. The flight paths are created by
INTERNATIONAL AIRPORT connecting the flight paths written in flight plan (FP) of the
This study focuses on the domestic and international flights flights arriving at RJTT in a single day. The flights landing
arriving at the Tokyo International Airport (RJTT). Tokyo on 34R do not affect the flights landing on 34L because the
International Airport (RJTT is the busiest airport in Japan, flight paths ending at 34L, depicted by red lines in Fig. 3,
and the fourth busiest airport in the world, which witnessed and the flight paths ending at 34R, depicted by blue lines in
85,408,975 passengers in 2017 [64]. Fig.3, are independent. SBCA is constructed between con-
The airport makes use of four runways: a set of paral- fluence points. Each domestic flight departs from a domestic
lel north-south runways (34L/16R and 34R/16L) and two airport in Japan, while each international flight starts at the
southwest-northeast crosswind runways (22/04 and 23/05). designated fix. The scheduled takeoff or departure time is
Two major runway configurations (see Fig. 2) are operated extracted from the time written in the first line in each FP.
depending on the wind direction. At the northerly operating These flights merge into other routes continuously until they
wind runway, an aircraft arrives at either runway 34L or arrive at RJTT. The first-come-first-serve protocol is applied
runway 34R, and a departing aircraft takes off from runway to each merging point as a confluence rule. If more than
05 or runway 34R, depending on their origin/destination two aircraft arrive at a confluence point, the flight that is
airports (see Fig. 2(a)). Basically, the northbound traffic scheduled to arrive at RJTT earlier is prioritized over the
utilizes runway 34R for both departure and arrival, while the other. This is the procedure followed in simulating air traffic
southbound traffic makes use of runway 05 and runway 34L flow between taking off and landing.
for departure and arrival, respectively. The southerly operated The meso-scale model grid point value (MSMGPV), pro-
wind runway, at which an aircraft arrives at runway 16L and vided by the Japan Meteorological Agency, is employed
16R (see Fig. 2(b)) was started in 2020. In this operation, to study the impact of weather. The data were collected
departure aircraft to south-western destination takeoff from by the Research Institute for Sustainable Humanosphere,
runway 22, and the others depart from runway 16L and Kyoto University (http://database.rish.kyoto-u.ac.jp/index-e.
4 VOLUME xx, 2021

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

TABLE 1. Definitions of the parameters used in the optimization problem

Begin
Parameter Definition
Nf the number of flights in a day’s schedule
Nd the number of delayed flights in Nf
Initialize Population TSAf the scheduled landing time of flight fi
i
TAAf the actual landing time of flight fi
i
gen = 0 TSDf the scheduled takeoff time of flight fi
i
TADf the actual takeoff time of flight fi
i
Evolution Assign Fitness Fi,j the fuel consumption of fi per time step [kg/timestep]
OSDf takeoff time offset of flight fi
i
Ol lower bound of takeoff time offset
Ou upper bound of takeoff time offset
No
Cond?
Reproduction
gen = gen + 1 Yes B. PROBLEM SETTING
Stop Crossover We attempted to develop a flight schedule with minimal
congestion and fuel consumption in this study. Thus, the
Mutation objective functions are (1) minimization of the total arrival
delay and (2) minimization of the total fuel consumption. The
FIGURE 4. Flowchart of the general multi-objective evolutionary algorithm optimization problem is mathematically defined by using the
(MOEA) parameters listed in Table 1 as follows:

Nd
html). The data include atmospheric properties such as wind
X
minimize (TAAfi − TSAfi )
and temperature mapped on a three-dimensional grid. These i=1
data are published every 3 hours. The grid points are located Nf TAAf
after 0.125° of longitude, 0.1° of latitude, and 50–100 hPa of X Xi (1)
Fi,j
pressure of altitude. In this study, the data of the grid point i=1 j=TADf
closest to the aircraft were used to add the wind speed to the i

true air speed in the traveling direction. subject to Ol ≤ OSDfi ≤ Ou

Arrival delay is defined as the difference between the actual
III. DESCRIPTION AND FORMULATION OF THE landing time (TAAfi ) and the scheduled landing time (TSAfi )
OPTIMIZATION PROCESS written in the flight plans. Total arrival delay is calculated
A. MULTI-OBJECTIVE OPTIMIZATION
by summing up the arrival delay of each flight, fi , to the
number of delayed flights, Nd (i ∈ 1, 2, · · · , Nd ). The taxi
The multi-objective evolutionary algorithm (MOEA), as time durations at the departure and arrival airports are not
shown in Fig. 4, is used to find the optimal solution of the accounted for in the simulation. Therefore, the arrival delay
optimization problem. The MOEA is capable of calculating for each flight is calculated by subtracting the scheduled
multiple optimal solutions in a single simulation run due landing time from the actual landing time. With respect
to its population-based approach, as shown in Fig. 4. The to fuel consumption, the performance table files (PTF) of
MOEA creates initialized populations, which are also known BADA are utilized to calculate the amount of the in-flight fuel
as “parent solutions” and serve as design variables. The algo- burned based on the aircraft type. In the PTF, the nominal fuel
rithm then assigns fitness values by calculating the objective consumption [kg/min] is described in each FL between FL0
function’s values. Certain populations are selected to repro- and approximately FL430 based on the status of the aircraft
duce the new populations, which are termed as “offspring (i.e., climb, cruise, or descent). These values are converted
solutions.” The crossover and mutation techniques are used in into the values using time step in SBCA [kg/timestep]. Thus,
the reproduction phase. These are the ways to stochastically the total fuel consumption is calculated by summing up
generate new solutions from an existing population, imitating the fuel consumption per time step, Fi,j , from the actual
the crossover that happens during sexual reproduction in departure time (TADfi ) to the actual arrival time (TAAfi ) for
biology. Thus, a certain number of solutions are selected all flights, Nf (i ∈ 1, 2, · · · , Nf , j ∈ TADfi , · · · , TAAfi ).
from the offspring solutions to form the parent solutions of In addition, we investigated the effects of the takeoff time
the next generation. This process is iterated until the end of difference on the arrival time and fuel consumption by fixing
the number of generations in MOEA. Finally, non-dominated the arrival time and offsetting the takeoff time. Thus, the
solutions, so-called “pareto front,” are obtained as the result design variable of the simulation is the takeoff time offset of
of MOEA. each flight (OSDfi ) between the lower bound (Ol ) and upper
bound (Ou ).
VOLUME xx, 2021 5

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

IV. COMPUTATIONAL CONDITIONS were used to construct the SBCA in this simulation. The
cell length was 70 m in the lateral direction and 20 m in
TABLE 2. Optimization configuration the vertical direction. A CA analysis with a time step of 1
MOEA NSGA-II second was then implemented to update the state of each
# of Population size 200 flight. The values of d and BD were adjusted on the basis
# of Generation 200 of a previous study [61] to account for the arrival interval
# of design variable (# of flights) 607
Range of offset [s] [-300, 300]
of the actual traffic data (CARATS Open Data [65]) on
Trials 1 11/15/2016 at runways 34L and 34R (see Fig. 5). These
values determine the in-trail separation in each flight phase
and virtually control the arrival capacity at RJTT. 40,000
simulations were performed by using the parameters listed
TABLE 3. Parameter setting in SBCA
in Table 2 and Table 3 to calculate the optimal takeoff time
Parameter Value and thereby minimize the total arrival delay and total fuel
Date 15th, November, 2016 consumption.
# of flights landing at 34R 140
# of flights landing at 34L 467
# of departure airport 51
# of starting fix for international flights 17 V. RESULTS AND DISCUSSION
# of routes constructed for SBCA 119
lx : Length of a cell in the lateral direction 70 [m]
This section discusses the results of the numerical simula-
ly : Length of a cell in the vertical direction 20 [m] tions from the four perspectives mentioned in Section V-A
∆t: Timestep 1 [s] (i.e., the validation result), Section V-B (i.e., the arrival delay
d 34R arrivals climb 5 [NM] / (lx /1852) and fuel consumption), Section V-C (i.e., the attribution of
cruise 16 [NM] / (lx /1852)
descent 10 [NM] / (lx /1852) the reduction of the arrival delay and fuel consumption), and
34L arrivals climb 5 [NM] / (lx /1852) Section V-D (i.e., the takeoff time adjustments). Section V-A
cruise 7 [NM] / (lx /1852) focuses on generating trust in our model by comparing the
descent 6 [NM] / (lx /1852)
BD 34R arrivals 5 [NM] / (lx /1852) nominal simulation result with the radar track by calculating
34L arrivals 2 [NM] / (lx /1852) four metrics. After the validation, Section V-B focuses on
the impact of the objective functions of the multi-objective
optimization by quantifying the benefits of controlling the
Table 2 shows the parameter setting of the optimization minute variations in the takeoff time. Analyzing the optimal
framework. Multi-objective air traffic flow optimization is solutions in detail, along with optimizing the schedule de-
conducted by using the non-dominated sorting genetic al- sign, is crucial for efficient ground delay operation and flight
gorithm II (NSGA-II) [33]. This algorithm has proven to be scheduling. Therefore, Section V-C analyzes the attribution
an effective and efficient multi-objective search technique in of the successful reduction in the arrival delay and fuel
various engineering applications. The design variables are the consumption. Section V-D discusses the efficient control of
takeoff time offsets of each flight arriving at RJTT. Therefore, the takeoff time by analyzing the design variables and their
the number of design variables is equal to the number of effects.
flights arriving at RJTT, which is equal to 607. The design
variables vary between −300 s and +300 s to investigate
the impact of a minute difference in the takeoff time on the A. VALIDATION RESULT
arrival delay and fuel consumption. The value interval for the
decision variables is 1 s corresponding to the time step in Figure 6 presents a comparison between the simulation result
SBCA, ∆t, as shown in Table 3. and the radar track. Four comparison metrics are calculated.
Additional distance is the difference between the scheduled
As described in Section II, SBCA is the rule-based deter-
flight path and the actual traveling distance. Arrival delay is
ministic simulator. Therefore, the computational parameters
the difference between the scheduled landing time and actual
of SBCA listed in Table 3 determine the behavior of each
landing time. Considering that the statistical distributions
aircraft. The simulation was performed on 11/15/2016, which
of the calculated metrics are quite close, it is implied that
is one of the nominal dates when the north wind operation
our model can capture the dynamics of individual flights
was applied at RJTT. The MSMGPV data were collected
accurately. With respect to the additional distance in Figure
on that date. The number of flights landing on runways
6(a), some flights in the radar track take a shortcut. Although
34R and 34L were 140 and 467, respectively. There are
our model cannot represent the shortcut mechanism, the other
51 departure airports, which include 49 domestic airports,
calculated metrics of our model are in agreement with those
the Taiwan Taoyuan International Airport (RCTP), and the
of the radar track, as shown in Figs. 6(b), 6(c), and 6(d).
Taipei Songshan Airport (RCSS). The flights departing from
Further comparative analyses will be provided in our future
these airports follow the climb phase specified by the BADA
work.
model. However, the other international flights start with the
cruise phase at 17 designated fixes. 119 unique flight paths
6 VOLUME xx, 2021

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

(a) Runway 34L (b) Runway 34R

FIGURE 5. Arrival interval

(a) Additional Distance (b) Traveling Distance

(c) Flight Duration (d) Arrival Delay

FIGURE 6. Calculated metrics for validation

B. ANALYZING ARRIVAL DELAY AND FUEL represent the solutions obtained by using the MOEA. A total
CONSUMPTION of 40,000 solutions are generated because the number of
generations and populations are 200 each. The red circles
Figure 7 depicts the objective results of the original sched- are non-dominated solutions, while the green square is the
ule and the solutions obtained by using the multi-objective solution corresponding to the original schedule. A total of 18
evolutionary algorithm MOEA. From the engineering per- non-dominated solutions were obtained in the current set of
spective, this figure enables to analyze the tradeoff between simulations. The non-dominated solutions are superior to the
the objective functions and the degree of improvement of the solution corresponding to the original schedule (hereinafter
optimal solution from the original solution. The horizontal referred to as the “original solution”), as shown in Fig. 7.
axis represents the total arrival delay, whereas the vertical The total arrival delay and total fuel consumption decreased
axis represents the total fuel consumption. The blue circles
VOLUME xx, 2021 7

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

TABLE 4. Representative solutions of the results of multi-objective optimization

Calculated Metrics Original Solution Minimum Delay Solution Minimum Fuel Solution
Total arrival delay [min] 2940.63 1427.88 1451.07
Averaged arrival delay [sec] 320.796 220.239 226.729
Total number of delayed flights [ac] 550 389 384
Percentage of the number of delayed flights [%] 90.61 64.09 63.26
Total fuel consumption [t] 3236.63 3164.35 3157.82
Averaged fuel consumption [kg] 5332.18 5213.09 5202.33
3260
3240

100

Frequency
Total Fuel Consumption [t]
3220

50
3200

0
3180

−1.0 −0.8 −0.6 −0.4 −0.2 0.0 0.2 0.4 0.6 0.8 1.0
Dominated Solution
Correlation Coefficient
Non−Dominated Solution
Min. Delay Solution FIGURE 8. Histogram of the correlation coefficient for all pairs of the
3160

Min. Fuel Solution non-dominated solutions

Solution with Original Schedule

1500 2000 2500 3000 3500

Total Arrival Delay [min] replaced by one objective function by considering the minute
takeoff time difference to be a design variable in the problem
FIGURE 7. Scatter plot in objective function space setting of the multi-objective optimization.
There is a significant difference between the original so-
lution and the non-dominated solutions, as shown in Figs. 7
by approximately 1500 min and 80 tons, respectively. As and 8. However, the non-dominated solutions are identical
shown in Table 4, averaged values of the reduced arrival delay to the original solution in the objective space and the design
and fuel consumption are approximately 100 s and 130 kg, variables space. Therefore, we compared the original solution
respectively. Thus, the current optimization technique suc- with one of the non-dominated solutions. The impact of
cessfully generates flight schedules that minimize delay and the total arrival delay and total fuel consumption were both
fuel consumption. Further, the values of objective functions accounted for while choosing the non-dominated solutions.
with non-dominated solutions are almost the same although As shown in Table 4, the averaged arrival delay and averaged
there is a quite weak tradeoff relationship between the total fuel consumption values of the non-dominated solutions was
arrival delay and total fuel consumption, as shown in Fig. 7. approximately 6 s and 10 kg, respectively. These values indi-
The degree of similarity should be investigated in the space of cate that the impact of fuel consumption is greater than that of
the design variables, as well as that of the objective functions. the arrival delay from customers’ and airlines’ perspectives.
Figure 8 depicts a histogram of the correlation coefficient Therefore, we selected the solution that provided minimum
for all pairs of the non-dominated solutions. As the number total fuel consumption (hereinafter referred to as “minimum
of obtained non-dominated solutions is 18, the total number fuel solution”), which is indicated by the light blue square in
of pairs is 153 (18 C2 ). The pairs are distributed between 0.90 Fig. 7.
and 1.00, with most being located between 0.95 and 1.00, Figure 9 depicts the arrival delay of each flight. The
as shown in Fig. 8. This indicates the existence of a strong horizontal and vertical axes represent the arrival delay and
positive correlation between each non-dominated solution. flights, respectively. The blue and red lines indicate the flights
Thus, each non-dominated solution is similar in the space landing at runway 34R and 34L, respectively. The arrival
of the design variables, as well as the objective space. This delay values of several flights in the original solution were
result suggests that the appropriate adjustment of the takeoff greater than 5 min, as shown in Fig. 9(a). In addition, a few
time contributes to the reduction of both the arrival delay and flights reported arrival delays in excess of 10 min at both 34L
fuel consumption. Thus, these two objective functions can be and 34R. However, the arrival delays of several flights were
8 VOLUME xx, 2021

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

34R Arrival 34L Arrival 34R Arrival 34L Arrival TABLE 6. Total number of flights in each air route cluster
600

600
Cluster # of flights Main Flights Direction
1 192 Short-haul Southwest
500

500
2 114 Long-haul Southwest
3 152 Long-haul&Short-haul Southwest
400

400
4 140 Long-haul&Short-haul North
Flights

Flights
300

300
200

200
to their origin points. Table 6 summarize the characteristics
100

100 of each air route cluster depicted in Fig. 10. Although only
598 flights have been mentioned in Table 6, the other 9
0

−10 −5 0 5 10 15 20 −10 −5 0 5 10 15 20
Arrival Delay [min] Arrival Delay [min]
international flights fly on the other route cluster depicted as
black lines in Fig. 10 with few interactions. Air route clusters
(a) Original Solution (b) Minimum Fuel Solution 1, 2, and 3 include the traffic arriving at RJTT from the
southwestern direction, whereas air route cluster 4 includes
FIGURE 9. Arrival delay of each flight
the traffic coming from the northern direction. Air route
cluster 1 includes short-haul flights, while air route clusters
drastically reduced in the minimum fuel solution, as shown 2 and 3 are predominantly composed of long-haul flights.
in Fig. 9(b). In particular, the minimum fuel solution success- Air route cluster 3 had more pop-up flights than air route
fully reduced the arrival delays of several flights landing on cluster 2. The blue and red circles represent the geographical
runway 34L. In addition, the arrival delays of several flights points where the “Step Back” and “Stop” rules were applied,
landing at 34L were reduced to less than 5 min. Some of the respectively. The concentric purple circles are plotted after
flights even touched down approximately 5 min before the every 10 NM within a radius of 200 NM. The “Step Back”
flights on the original schedule. Hence, we can state that the movement is implemented across the area close to RJTT,
minute takeoff time adjustments made in the minimum fuel while the “Stop” movement is conducted primarily at the
solution contributed to relieving traffic congestion. confluence points, as shown in Fig. 10(a). These spacing
Several factors contribute to the simultaneous reduction of adjustments are largely implemented in the region between
arrival delay and fuel consumption. An additional analysis longitudes of 130° E and 140° E and latitudes of 30° N
demonstrated that the reduction of these parameters coin- and 35° N. However, the number of the spacing adjustments
cided with the reduction of conflicts in the en-route airspace. implemented in the minimum fuel solution is significantly
Therefore, Section V-C analyzes the impact of spacing ad- less than that in the original solution, as shown in Fig. 10(b).
justments, such as the “Step Back” and “Stop” CA rules. Specifically, the number of geographical points representing
the application of the “Step Back” in the area between the
40 NM and 200 NM radii in the southwestern direction is
C. ANALYZING SPACING ADJUSTMENT
reduced in the minimum fuel solution. Further, the number
TABLE 5. Total number of the spacing adjustments
of “Step Back” adjustments made in the region between the
100 NM and 200 NM radii on air route clusters 1 and 2 was
Movement Original Solution Minimum Fuel Solution
significantly reduced in the minimum fuel solution. Similarly,
Step Back 2687 544
Stop 28238 5564
the number of “Step Back” adjustments made in the region
beyond the 200 NM radius on air route clusters 3 and 4 was
less in the minimum fuel solution than that in the original
Table 5 lists the number of spacing adjustments, such as solution. In addition, the geographical points of application
the “Step Back” and “Stop” movements, made in each solu- of the “Step Back” rule were absent in the regions beyond
tion. The total number of “Step Back” and “Stop” movements the 100 NM radius on air route cluster 3. Several pop-up
in the minimum fuel solution are significantly less than those flights merged on this route cluster. Thus, the conflicts in
in the original solution, as presented in Table 5. The total the en-route airspace are substantially reduced due to the
number of “Step Back” and “Stop” movements in the original minute takeoff time differences, as shown in Fig 14. Notably,
solution are 2687 and 28238, respectively. These values the number of spacing adjustments in the minimum fuel
are reduced by approximately 80 % in the minimum fuel solution is reduced not only in the en-route airspace but also
solution. The minor takeoff time adjustments for each flight in the terminal airspace enclosed within the 100 NM radius
drastically decreased the number of conflicts in the en-route centered at RJTT.
air space. Therefore, the arrival delay and fuel consumption Figure 11 depicts the number of spacing adjustments made
were successfully reduced, as mentioned in Section V-B. after 10 NM radius segments. The horizontal axis represents
Figure 10 depicts the points of implementation of the spac- the radius of entry for the aircraft, and the vertical axis
ing adjustments according to the CA rules. Air routes leading represents the number of spacing adjustments. The flights
to RJTT are divided into four representative groups according coming from the northern and southwestern directions are
VOLUME xx, 2021 9

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

Air Route Cluster 1 Air Route Cluster 1

Air Route Cluster 2 Air Route Cluster 2
Air Route Cluster 3 Air Route Cluster 3
Air Route Cluster 4 Air Route Cluster 4
Other Route Cluster Other Route Cluster
Concentic Circles around RJTT Concentic Circles around RJTT

Step Back Point Step Back Point

Stop Point Stop Point
40 40
Latitude [deg]

Latitude [deg]
30 30

120 130 140 150 120 130 140 150

Longitude [deg] Longitude [deg]

(a) Original solution (b) Minimum fuel solution

FIGURE 10. Mapping of spacing adjustments

Category Category
# of Spacing Adjustment

# of Spacing Adjustment

180 540
Original Original
Fuel 480 Fuel
420
120 360
300
240
60 180
120
60
0 0
10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200
Concentric Circle Radius when Entering [NM] Concentric Circle Radius when Entering [NM]

(a) North traffic (b) South traffic

FIGURE 11. Number of spacing adjustments for different radii of entry

independent of each other because their routes are separated. radii in the southwestern direction is significantly less in the
Hence, Fig. 11(a) includes the flights arriving from the north- minimum fuel solution than that of the original solution, as
ern direction, whereas Fig. 11(b) includes the flights arriv- shown in Fig. 11(b). The maximum mean number of spacing
ing from the southwestern direction. There are significant adjustments in the airspace between the 50 NM and 60 NM
differences among the number of spacing adjustments made radii is approximately equal to 60, which corresponds to the
in the original solution in the regions enclosed between the last merging point before arrival. The maximum mean and
70 NM and 100 NM radii, the 130 NM and 140 NM radii, variance in the airspace between the 50 NM and 60 NM radii
and the 170 NM and 180 NM radii in the northern direc- in the southwestern direction are significantly greater than
tion. The maximum mean number of spacing adjustments the corresponding values in the airspace between the 70 NM
is approximately 30. These adjustments occur in the region and 80 NM radii. Therefore, we can state that the reduction
between the 70 NM and 80 NM radii, which corresponds to in the number of spacing adjustments around the confluence
the last merging point before arrival. However, the variation points that are located before the terminal maneuvering area
among the number of spacing adjustments made in these reduces both the arrival delay and fuel consumption.
areas is considerably reduced in the minimum fuel solution. The most remarkable result to emerge from the data com-
The mean number of spacing adjustments is significantly parison is the significant reduction in the number of spacing
reduced as well. These areas correspond to the confluence adjustments in the terminal airspace. Terminal airspaces wit-
points and surroundings in the northern direction, as shown ness a large number of spacing adjustments because of the
in Fig. 10. Similarly, the variation of the number of spacing simultaneous arrival of the inbound traffic at RJTT, as shown
adjustments in the area between the 50 NM and 70 NM in Fig. 11(a). According to the specified CA parameters d
10 VOLUME xx, 2021

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

and BD, these flights attempt to maintain a sufficient in- hours is higher than it is during other hours according to the
trial separation when they come close to each other. Feasible marginal distribution of the actual arrival time. Further, most
solutions to avoid conflicts can be automatically generated flights with high benefits have recovered time delays status
by using the evolutionary algorithm. In particular, the “Step and are depicted by the symbols having a larger size than
Back” movement would have a more negative impact than the rest, as shown in Fig 12. The number of flights having
the “Stop” movement because the former would reduce the a takeoff time earlier than that mentioned in the original
space between the proceeding and succeeding flights. Hence, schedule is slightly higher than the number of flights having
the number of “Step Back” movements in the en-route space a delayed takeoff, as shown in the marginal distribution of
are reduced, as depicted in Fig 10. These results indicate that the takeoff adjustment time. Notably, some flights arriving
the operation wherein the flights takeoff in order to maintain during congested hours recovered from their delayed statuses
sufficient separation in the terminal airspace significantly by delaying their takeoff time. Further, the flights whose
reduces the arrival delay and fuel consumption as a result of delays are recovered during the morning hours generally
the reduced conflicts. In reality, there are many flights exclud- belong to air route cluster 1 and are represented by a circle
ing the arrival flights for RJTT. However, this study focuses in Fig 12(b). This is because the number of flights belonging
to adjust the takeoff time at the level of delay, which gets to air route cluster 1 is higher than that of the other air route
overlapped with the in-flight phase. Therefore, the airspace clusters, as shown in Table 6.
demand in the en-route sectors would not change drastically. The above results indicate that controlling the takeoff
In this situation, air traffic controllers in those sectors could time during congested hours would be highly effective, as
safely provide directions to the pilots to avoid any conflicts shown in Fig. 12. Several perspectives could be considered
in the flight travels. Additional analyses are conducted in to adjust the takeoff time appropriately. We further studied
Section V-D to clarify the possibility of implementing this two perspectives, namely, the takeoff time difference at the
operation. same origin and the time interval at specific fixes that are
close to the entry points of the terminal airspace. According
D. ANALYZING DESIGN VARIABLES AND THE TAKEOFF to the perspective of the takeoff time difference at the same
TIME origin, the number of flights departing from the same origin
Figure 12 shows the scatter plot of the design variable. The to RJTT tends to increase within a few minutes due to the
horizontal axis represents the actual arrival time obtained in high demand during congested hours. Therefore, the takeoff
the minimum fuel solution. The vertical axis represents the time interval at such origins is likely to be less than that
value of the design variable. The marginal distributions are of other low-demand origins. However, this narrow interval
additionally depicted according to each axis. The design vari- could negatively impact the in-trail separation in the terminal
ables are allotted different shapes according to the air route region. In addition, pop-up flights could disturb the tailored
clusters. The symbols representing the flights whose delays separation. Thus, according to the current perspective, the
have been recovered have larger sizes than those representing time interval at the point of merging of the pop-up flights
other statuses. Flights arriving 5 min after the scheduled is important.
arrival time are classified as delayed flights. Benefit refers to Figure 13 depicts the scatter plot of the takeoff time
the quantitative benefit of the design variables. Benef it(i) interval at the same origin. The horizontal axis represents
(i = 1, 2, · · · , 607) is calculated as follows: the takeoff time of successive flights at the same origin. The
vertical axis represents the takeoff time interval between the
Benef it(i) = Delayorg (i) − Delayf uel (i) + V ar(i) (2)
proceeding and succeeding flights at the same origin. The
where i denotes the number allocated to a flight. Further, marginal density distributions are additionally depicted for
Delayorg (i) represents the arrival delay reported in the orig- each axis.
inal schedule, and Delayf uel (i) represents the arrival delay The flights obtained from the original solution and the
reported in the minimum fuel solution. V ar(i) is the takeoff minimum fuel solution are represented by red and blue
adjustment time for each flight. A large benefit indicates that symbols, respectively. The size of the symbol corresponds
the takeoff time adjustment of the flight is more efficient to the number of spacing adjustments. In addition, the shape
than that of a flight with a small benefit. As the benefit of the symbol varies according to the distance of the flight to
values of almost all flights are greater than 0, as shown in RJTT (e.g., “>0200 NM” indicates that the distance to RJTT
Fig. 12, it can be stated that the takeoff time adjustments is greater than 200 NM and less than 300 NM). The number
are successfully implemented in general. The red-colored of spacing adjustments in the original solution increased
design variables, which represent high benefit values, were significantly during the morning and evening hours in air
clustered between 13 and 18 hours for traffic inbound from route clusters 1–3, while it increased during the afternoon in
the northern direction, as shown in Fig. 12(a). However, the air cluster route 4. This corresponds to a congested time when
traffic inbound from the southwestern direction reported high the benefit depicted in Fig 12 is remarkably high. In contrast,
benefit design variables between 8 and 11 hours and 17 and the number of spacing adjustments in the minimum fuel
22 hours. Flights with such design variables arrive during solution depicts a reduction during these congested hours.
congested hours because the number of arrivals during these In addition, the marginal density distribution of the original
VOLUME xx, 2021 11

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

5
5
Benefit
4
Origin Air Route 4
10 [min]
Take−off Adjustment Time [min]

Take−off Adjustment Time [min]

3 Cluster 4 3 5 [min]
2 0 [min]
2
Benefit
1 1
5 [min] Origin Air Route
0 0
Cluster 1
−1 0 [min] −1 Cluster 2
Cluster 3
−2 −2 Other Routes

−3 Arrival Status
−3
Delay Recovered Arrival Status
−4 Other Status −4 Delay Recovered
Other Status
−5 −5

2 4 6 8 10 12 14 16 18 20 22 2 4 6 8 10 12 14 16 18 20 22 24
Actual Arrival Time [hr] Actual Arrival Time [hr]

(a) Northern inbound traffic (b) Southwestern inbound traffic

FIGURE 12. Sscatter plot of the design variable

Category
10 10 ●
●
●

●
●
Original Category
●
●
Fuel ● ●

● ● ●

● ●
● ●
Original
● Fuel
Take−off Time Interval [min]

Take−off Time Interval [min]

8 8 ● ●

NumAdjustment
●
0 ●● ●
● ●
DtG
● 50 ●

● 100 ● > 0200NM

6 ● 150
6 ●
● ● ●● ● ● ●
● > 0550NM
●

● 200 ●
● > 0850NM
● 250 ● ●

●
●

4
4 ● ●
NumAdjustment
DtG
● ●
● ●
●
●

●
●
●
● ●●
●
0
● > 0250NM ●
● ● 50
● > 0300NM ● 100
> 0350NM 2 ●
● ● ●
● 150
2
> 0400NM ● 200
●
> 0500NM ●
> 0600NM
8 10 12 14 16 18 20 22 8 10 12 14 16 18 20 22
Take−off Time of Succeeding Flights [hr] Take−off Time of Succeeding Flights [hr]

(a) Air route cluster 1 (b) Air route cluster 2

10 ●● ●
Category 10 ●

● ●
Original ●
●
Category
● ●
Fuel ●
●
●
Original
Take−off Time Interval [min]

Take−off Time Interval [min]

● ●

8 ●
●
Fuel
8 ●
●

DtG
> 0450NM ● ● DtG
● > 0550NM ●
6 > 0900NM > 0400NM
●
6 ●
●
● > 0450NM
> 1150NM
> 1200NM ●
● NumAdjustment
4 ● ●
NumAdjustment 4 ●
● ● ●
● ●
●
●
0
● ●
0 ● ●
●
● 50
● 100
● 50
● 100
● ●
● ● 150
2 ●
● 150
● ● 200
●
● ● 200 2 ● ● ● ● ● ● ●

4 6 8 10 12 14 16 18 20 22 8 10 12 14 16 18 20
Take−off Time of Succeeding Flights [hr] Take−off Time of Succeeding Flights [hr]

(c) Air route cluster 3 (d) Air route cluster 4

FIGURE 13. Scatter plot of the takeoff time interval at the same origin

12 VOLUME xx, 2021

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

solution is higher than that of the minimum fuel solution route cluster 4 are greater than those of air route clusters
during these hours. This indicates that the minimum fuel 1, 2, and 3 because the parameters of the model, namely d
solution reduces the number of flights that have a takeoff time and BD, are greater for air route cluster 4 than they are for
interval greater than 10 min. However, the marginal density the other air route clusters. In addition, the distribution of
distribution of the original solution during non-congested the time interval of air route cluster 4 is similar to a normal
hours is lower than that of the minimum fuel solution. Hence, distribution because the final merge point of air route cluster
the minimum fuel solution adjusts the takeoff time such that 4 is identical to the final merge point of the traffic coming
the interval is as long as possible during congested time from the northern direction, as depicted in Fig. 15(d). The
and as short as possible during non-congested hours. The peak of the distribution increases, and its variance decreases,
peak of the density distribution of the takeoff time interval is as shown in Table 7.
relatively higher in the minimum fuel solution, as shown in Barring a few exceptions, these results demonstrate that
the marginal density distribution of the takeoff time interval. the time interval at the final merge point that is close to the
The peak of the marginal density distribution of the takeoff terminal increases with minimal variation in the minimum
time interval in the original solution was approximately 4 fuel solution, as shown in Fig 15 and Table 7. This result
min, whereas the peal of the minimum fuel solution was implies that the spacing adjustments can be reduced by not
greater than 4 min. Therefore, the takeoff time interval of only maintaining a sufficient in-trial separation while also
flights traveling to the same destination from the same origin minimizing the deviation of this time interval.
should be relatively higher.
Figure 15 and Table 7 depict the histogram and the sum- VI. CONCLUSION
mary of the time interval before the final merge point of each We conducted the multi-objective air traffic optimization of
air route cluster that is depicted in Fig. 14. The concentric schedule design by using a step-back cellular automaton in
purple circles are plotted after 60 NM within a radius of this study. The objective functions were to minimize (1)
180 NM. The results of air route clusters 1, 2, and 3 are the total arrival delay and (2) total fuel consumption. The
discussed separately from that of air route cluster 4 because design variable used in the study was the time difference
the parameters of the model, namely, d and BD, are different, between the scheduled and actual takeoff time within ±300
as shown in Table 3. The mean time interval of the final s. The number of the design variables was equal to 607 and
merge point of air route cluster 1 increased by 16.9 s, and included both domestic and international flights landing at
its standard deviation decreased by 18.8 s. The frequency of Tokyo International Airport (RJTT).
the time interval of the original solution, which is depicted in The results of the multi-objective optimization produced
red, is higher than that of the minimum fuel solution, which several non-dominated solutions that were superior to the
is depicted in blue, at approximately 1 min, as shown in Fig. solution corresponding to the original schedule. The total
15(a). In contrast, the frequency of the minimum fuel solution arrival delay and total fuel consumption were reduced by
at approximately 3 min is higher than that of the original approximately 1500 min and 80 tons, respectively. This study
solution. Therefore, more succeeding flights are sufficiently highlights the importance of making minor adjustments in
separated from proceeding flights at the final merge point of the takeoff time. The analyses of the objective function and
air route cluster 1. Similarly, the mean time interval of the design variable indicate that both objective functions can
final merge point of air route cluster 2 increases by 17.6 s, be minimized simultaneously by performing takeoff time
as shown in Table 7. The time interval of this cluster was adjustments in an identical manner.
greater than air route clusters 1 and 2 because the number Detailed comparisons between the original solution and
of flights in this cluster was the lowest in air route clusters the minimum fuel solution presented two significant findings.
1, 2, and 3. Therefore, the mean time interval of air route Minor takeoff adjustments contribute to a significant reduc-
cluster 2 obtained from the minimum fuel solution is greater tion in the number of spacing adjustments. These adjustments
than those of air route clusters 1 and 3. In contrast, the mean lead to the reduction of arrival delay and fuel consumption.
and standard deviation of the time interval of air route cluster Further, the takeoff time adjustment is highly effective during
3 before the final merge point reduced drastically declined. periods of high traffic demand in the terminal space. The
It is noteworthy that several pop-up and long-haul flights quantitative benefit provided to several flights is expected to
merge at the final merge point in air route cluster 3. The be more than 5 min during the congested time periods.
standard deviation of the time interval in air route cluster 3 Two effective operations can be proposed on the basis of
obtained from the original solution is greater than the other the current findings and additional analyses. The takeoff time
clusters, as shown in Table 7. However, this variation is interval of flights originating from the same point during
suppressed in the minimum fuel solution. Consequently, the congestion hours should be longer than those during non-
values of the resulting mean and standard deviation of the congestion hours. Pop-up flights should merge with suffi-
time interval are similar to the values of air route cluster 1. cient in-trail separation and minimum variation of the time
The mean time interval of air route cluster 4 increases by interval. Therefore, we can conclude that the optimization
24.7 s, while its standard deviation reduces by 18.9 s. The of airport operations, such as takeoff time, spot assignment,
absolute values of the mean and standard deviation of air and taxi routes, has a significant impact on air traffic flow.
VOLUME xx, 2021 13

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

38 Air Route Cluster 1

Air Route Cluster 2
Air Route Cluster 3
Air Route Cluster 4
Latitude [deg]
Other Route Cluster
36 Concentic Circles around RJTT

FinalMerge
34 Routes 1
Routes 2
Routes 3
Routes 4
32

135.0 137.5 140.0 142.5

Longitude [deg]

FIGURE 14. Air route and final merging point around RJTT

Category 10.0 Category

Original Original
15 Fuel Fuel

7.5
Frequency

Frequency

10
5.0

5
2.5

0 0.0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Time Interval at 7NM before Final Confluence Point [min] Time Interval at 7NM before Final Confluence Point [min]

(a) Air route cluster 1 (b) Air route cluster 2

6 Category Category
Original Original
Fuel Fuel

15

4
Frequency

Frequency

10

2
5

0 0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9
Time Interval at 7NM before Final Confluence Point [min] Time Interval at 16NM before Final Confluence Point [min]

(c) Air route cluster 3 (d) Air route cluster 4

FIGURE 15. Histogram of the time interval before the final confluence point of each air route cluster

14 VOLUME xx, 2021

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

TABLE 7. Summary of the time interval at 7 NM before the final confluence point of each air route cluster

Cluster Meanorg [sec] Meanf uel [sec] Difference of Mean [sec] SDorg [sec] SDf uel [sec] Difference of SD [sec]
1 221.1 238.0 +16.9 156.9 138.1 −18.8
2 251.8 269.4 +17.6 147.5 153.5 +6.0
3 256.5 239.2 −17.3 183.8 146.9 −36.9
4 200.2 224.9 +24.7 124.2 105.3 −18.9

Although this paper focuses on traffic arriving at RJTT on [15] A. Mukherjee, S. Grabbe, and B. Sridhar, “Predicting ground delay pro-
a specific date, the proposed approach is applicable for any gram at an airport based on meteorological conditions,” 2014.
[16] M. Bloem and N. Bambos, “Ground delay program analytics with behav-
type of traffic on different dates because NSGA-II is a meta- ioral cloning and inverse reinforcement learning,” Journal of Aerospace
heuristic optimization method, which has the potential to Information Systems, vol. 12, no. 3, pp. 299–313, 2015.
provide sufficient solutions to the problem of takeoff time [17] K. D. Kuhn, “A methodology for identifying similar days in air traffic
flow management initiative planning,” Transportation Research Part C:
adjustments for each flight. Emerging Technologies, vol. 69, pp. 1–15, 2016.
In the future, we intend to conduct optimization studies [18] Y. Xu, S. Wandelt, and X. Sun, “Robust integrated airline scheduling with
considering various scenarios, such as runway changes and chance constraints,” in 9th International Conference for Research in Air
Transportation, 2020.
inclement weather, to provide an ATM with sufficient re- [19] L. Hassan, B. Santos, and J. Vink, “Airline disruption management: A lit-
silience. We hope that it will eventually be possible to im- erature review and practical challenges,” Computers Operations Research,
prove the SBCA model to consider flights excluding arrival vol. 127, p. 105137, 2021.
[20] H. D. Sherali, K.-H. Bae, and M. Haouari, “Integrated airline schedule de-
traffic at RJTT. sign and fleet assignment: Polyhedral analysis and benders’ decomposition
approach,” INFORMS Journal on Computing, vol. 22, no. 4, pp. 500–513,
ACKNOWLEDGMENT 2010.
[21] H. Jiang and C. Barnhart, “Robust airline schedule design in a dynamic
The authors would like to thank the Ministry of Land, In- scheduling environment,” Computers Operations Research - CoR, vol. 40,
frastructure, Transport and Tourism (MLIT) for providing pp. 831–840, 03 2013.
access to the CARATS Open Data. The authors would also [22] J. P. Pita, N. Adler, and A. P. Antunes, “Socially-oriented flight scheduling
and fleet assignment model with an application to norway,” Transportation
like to thank the Electronic Navigation Research Institute for Research Part B: Methodological, vol. 61, pp. 17–32, 2014.
supplying the information that was crucial in developing the [23] L. Cadarso and R. de Celis, “Integrated airline planning: Robust update of
route and timetable for the simulations. scheduling and fleet balancing under demand uncertainty,” Transportation
Research Part C: Emerging Technologies, vol. 81, pp. 227–245, 2017.
[24] O. Faust, J. Gönsch, and R. Klein, “Demand-oriented integrated schedul-
REFERENCES ing for point-to-point airlines,” Transportation Science, vol. 51, no. 1,
[1] “ICAO long-term traffic forecasts passenger and cargo,” International Civil pp. 196–213, 2017.
Aviation Organization, 2016. [25] N. Kenan, A. Jebali, and A. Diabat, “The integrated aircraft routing
[2] “Five years to return to the pre-pandemic level of passenger demand,” problem with optional flights and delay considerations,” Transportation
International Air Transport Association, 2019. Research Part E: Logistics and Transportation Review, vol. 118, pp. 355–
[3] “Global air traffic management operational concept,” International Civil 375, 2018.
Aviation Organization, 2005. [26] K. Wei, V. Vaze, and A. Jacquillat, “Airline timetable development and
[4] “Concept of operations for the next generation air transportation system fleet assignment incorporating passenger choice,” Transportation Science,
ver. 3.2,” Federal Aviation Administration Joint Planning Development, vol. 54, no. 1, pp. 139–163, 2020.
2010. [27] L. H. Lee, C. U. Lee, and Y. P. Tan, “A multi-objective genetic algorithm
[5] “European atm master plan edition 2015,” SESAR Joint Undertaking, for robust flight scheduling using simulation,” European Journal of Oper-
2015. ational Research, vol. 177, no. 3, pp. 1948–1968, 2007.
[6] “Long-term vision for the future air traffic systems,” Japan Civil Aviation [28] K. Ng, C. Lee, F. T. Chan, and Y. Lv, “Review on meta-heuristics ap-
Bureau, 2010. proaches for airside operation research,” Applied Soft Computing, vol. 66,
[7] A. R. Odoni, “The flow management problem in air traffic control,” in pp. 104–133, 2018.
Flow control of congested networks, pp. 269–288, Springer, 1987. [29] A. Pellegrini, P. Di Sanzo, B. Bevilacqua, G. Duca, D. Pascarella,
[8] M. Terrab and A. R. Odoni, “Strategic flow management for air traffic R. Palumbo, J. J. Ramos, M. À. Piera, and G. Gigante, “Simulation-based
control,” Oper. Res., vol. 41, p. 138–152, Feb. 1993. evolutionary optimization of air traffic management,” IEEE Access, vol. 8,
[9] O. Richetta and A. R. Odoni, “Solving optimally the static ground-holding pp. 161551–161570, 2020.
policy problem in air traffic control,” Transportation science, vol. 27, no. 3, [30] G. Froyland, S. J. Maher, and C.-L. Wu, “The recoverable robust tail
pp. 228–238, 1993. assignment problem,” Transportation Science, vol. 48, no. 3, pp. 351–372,
[10] O. Richetta and A. R. Odoni, “Dynamic solution to the ground-holding 2014.
problem in air traffic control,” Transportation research part A: Policy and [31] Z. Liang, F. Xiao, X. Qian, L. Zhou, X. Jin, X. Lu, and S. Karichery,
practice, vol. 28, no. 3, pp. 167–185, 1994. “A column generation-based heuristic for aircraft recovery problem with
[11] M. O. Ball, R. Hoffman, A. R. Odoni, and R. Rifkin, “A stochastic integer airport capacity constraints and maintenance flexibility,” Transportation
program with dual network structure and its application to the ground- Research Part B: Methodological, vol. 113, pp. 70–90, 2018.
holding problem,” Operations research, vol. 51, no. 1, pp. 167–171, 2003. [32] M. Zanin, Y. Zhu, R. Yan, P. Dong, X. Sun, and S. Wandelt, “Characteri-
[12] A. Mukherjee and M. Hansen, “A dynamic stochastic model for the single zation and prediction of air transport delays in china,” Applied Sciences,
airport ground holding problem,” Transportation Science, vol. 41, no. 4, vol. 10, no. 18, 2020.
pp. 444–456, 2007. [33] K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast and elitist mul-
[13] Y. Liu and M. Hansen, “Incorporating Predictability Into Cost Opti- tiobjective genetic algorithm: Nsga-ii,” IEEE transactions on evolutionary
mization for Ground Delay Programs,” Transportation Science, vol. 50, computation, vol. 6, no. 2, pp. 182–197, 2002.
pp. 132–149, February 2016. [34] R. M. Harris, “Models for runway capacity analysis.,” tech. rep., MITRE
[14] Y. Wang and D. Kulkarni, “Modeling weather impact on ground delay CORP MCLEAN VA, 1972.
programs,” SAE International Journal of Aerospace, vol. 4, no. 2011-01- [35] M. Janić and V. Tošić, “Terminal airspace capacity model,” Transportation
2680, pp. 1207–1215, 2011. Research Part A: General, vol. 16, no. 4, pp. 253–260, 1982.

VOLUME xx, 2021 15

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

[36] M. Hansen, “Micro-level analysis of airport delay externalities using [61] T. Tatsukawa, S. Nagaoka, K. Anzai, and K. Fujii, “A study of air traffic
deterministic queuing models: a case study,” Journal of Air Transport simulation around haneda airport using step back cellular automaton,”
Management, vol. 8, no. 2, pp. 73–87, 2002. in 2018 AIAA Guidance, Navigation, and Control Conference, p. 1597,
[37] N. Pyrgiotis, K. M. Malone, and A. Odoni, “Modelling delay propagation 2018.
within an airport network,” Transportation Research Part C: Emerging [62] K. Sekine, T. Tatsukawa, S. Nagaoka, and K. Fujii, “Preliminary study
Technologies, vol. 27, pp. 60–75, 2013. of multi-objective air traffic optimization by using step back cellular
[38] P. K. Menon, G. D. Sweriduk, and K. D. Bilimoria, “New approach for automaton,” in AIAA Aviation 2019 Forum, p. 3407, 2019.
modeling, analysis, and control of air traffic flow,” Journal of guidance, [63] EUROCONTROL, “Base of aircraft data (bada).” https://www.
control, and dynamics, vol. 27, no. 5, pp. 737–744, 2004. eurocontrol.int/sites/default/files/publication/files/bada-factsheet.pdf,
[39] P. Menon, G. Sweriduk, T. Lam, G. Diaz, and K. D. Bilimoria, “Computer- 2015.
aided eulerian air traffic flow modeling and predictive control,” Journal of [64] “Passenger traffic 2017 final (annual),” Airports Council International,
Guidance, Control, and Dynamics, vol. 29, no. 1, pp. 12–19, 2006. 2019.
[40] D. B. Work and A. M. Bayen, “Convex formulations of air traffic flow op- [65] Air Navigation Services Department Civil Aviation Bureau, MLIT, “Data
timization problems,” Proceedings of the IEEE, vol. 96, no. 12, pp. 2096– extracted from the air navigation services department’s database becomes
2112, 2008. available.” https://www.mlit.go.jp/common/001082594.pdf, 2015.
[41] D. Sun and A. M. Bayen, “Multicommodity eulerian-lagrangian large-
capacity cell transmission model for en route traffic,” Journal of guidance,
control, and dynamics, vol. 31, no. 3, pp. 616–628, 2008.
[42] P. Wei, Y. Cao, and D. Sun, “Total unimodularity and decomposition
method for large-scale air traffic cell transmission model,” Transportation
Research Part B: Methodological, vol. 53, pp. 1–16, 2013.
[43] H. Zhang, Y. Xu, L. Yang, and H. Liu, “Macroscopic model and simulation
analysis of air traffic flow in airport terminal area,” Discrete Dynamics in
Nature and society, vol. 2014, 2014.
[44] T. Ezaki and K. Nishinari, “Potential global jamming transition in aviation
networks,” Physical Review E, vol. 90, no. 2, p. 022807, 2014.
[45] D. Chen, M. Hu, Y. Ma, and J. Yin, “A network-based dynamic air traffic
flow model for short-term en route traffic prediction,” Journal of Advanced
Transportation, vol. 50, no. 8, pp. 2174–2192, 2016. KATSUHIRO SEKINE received his B.S. degree
[46] K. Ng, C. Lee, F. T. Chan, C.-H. Chen, and Y. Qin, “A two-stage robust in Industrial Management and Engineering from
optimisation for terminal traffic flow problem,” Applied Soft Computing, the Tokyo University of Science, Tokyo, Japan,
vol. 89, p. 106048, 2020. in 2019. During his career, he visited German
[47] K. K. Ng, C.-H. Chen, and C. Lee, “Mathematical programming for- Aerospace Center (DLR) as a research trainee and
mulations for robust airside terminal traffic flow optimisation problem,” carried out research on extended arrival manage-
Computers Industrial Engineering, vol. 154, p. 107119, 2021. ment. He is currently taking a master’s at Tokyo
[48] S. Alam, H. A. Abbass, and M. Barlow, “Atoms: Air traffic operations and University of Science. He is also a research trainee
management simulator,” IEEE Transactions on intelligent transportation at Air Traffic Management Department of Elec-
systems, vol. 9, no. 2, pp. 209–225, 2008. tronic Navigation Research Institute (ENRI) and
[49] D. Šišlák, P. Volf, and M. Pěchouček, “Agent-based cooperative decen- National Institute of Maritime, Port and Aviation Technology. His research
tralized airplane-collision avoidance,” IEEE Transactions on Intelligent interests include multi-objective optimization, statistical and machine learn-
Transportation Systems, vol. 12, no. 1, pp. 36–46, 2010. ing method, modeling and simulation, and their applications in air traffic
[50] A. K. Agogino and K. Tumer, “A multiagent approach to managing air
management.
traffic flow,” Autonomous Agents and Multi-Agent Systems, vol. 24, no. 1,
pp. 1–25, 2012.
[51] Y. Guleria, Q. Cai, S. Alam, and L. Li, “A multi-agent approach for
reactionary delay prediction of flights,” IEEE Access, vol. 7, pp. 181565–
181579, 2019.
[52] D. Sun, I. S. Strub, and A. M. Bayen, “Comparison of the performance of
four eulerian network flow models for strategic air traffic management,”
Networks & Heterogeneous Media, vol. 2, no. 4, p. 569.
[53] P. Sarkar, “A brief history of cellular automata,” Acm computing surveys
(csur), vol. 32, no. 1, pp. 80–107, 2000.
[54] C. Kim, K. Abubaker, and O. Obah, “Cellular automata modeling of en
route and arrival self-spacing for autonomous aircrafts,” in 50th Annual
Meeting of Air Traffic Controllers Association, pp. 127–134, 2005.
[55] S.-P. Yu, X.-B. Cao, and J. Zhang, “A real-time schedule method for air- TOMOAKI TATSUKAWA received his B.S. de-
craft landing scheduling problem based on cellular automation,” Applied gree in 2002 and M.S. degree in engineering in
Soft Computing, vol. 11, no. 4, pp. 3485–3493, 2011. 2004 from Tokyo Institute of Technology, Tokyo,
[56] Y. He, K. Cai, Y. Li, and M. Xiao, “An improved cellular-automaton-based Japan. He was a software engineer at SGI Japan
algorithm for real-time aircraft landing scheduling,” in 2014 Seventh In- Ltd., from 2004 to 2010. He received his Ph.D. in
ternational Symposium on Computational Intelligence and Design, vol. 1, engineering in 2012 from the University of Tokyo,
pp. 284–288, IEEE, 2014. Tokyo, Japan. In 2012–2015, he was a Post-
[57] S. jin Wang and Y. hui Gong, “Research on air route network nodes doctoral Research Fellow in Institute of Space and
optimization with avoiding the three areas,” Safety Science, vol. 66, pp. 9–
Astronautical Science, Japan Aerospace Explo-
18, 2014.
ration Agency (ISAS/JAXA), Sagamihara, Japan.
[58] W.-X. Lim and Z.-W. Zhong, “Re-planning of flight routes avoiding con-
vective weather and the “three areas”,” IEEE Transactions on Intelligent
In 2015–2019, he was a junior Associate Professor at Tokyo University
Transportation Systems, vol. 19, no. 3, pp. 868–877, 2017. of Science, Tokyo, Japan. Since April 2020, he has been an Associate
[59] F. Enayatollahi and M. A. Atashgah, “Wind effect analysis on air traffic Professor with Department of Information and Computer Technology, Tokyo
congestion in terminal area via cellular automata,” Aviation, vol. 22, no. 3, University of Science. His areas of research interest are multiobjective evolu-
pp. 102–114, 2018. tionary computation, aerodynamic design optimization, and multi-objective
[60] F. Enayatollahi, M. A. Atashgah, S. Malaek, and P. Thulasiraman, “Pbn- design exploration. He is a member of IEEE and JSME. He received the
based time-optimal terminal air traffic control using cellular automata,” Young Researcher Award of IEEE Computational Intelligence Society Japan
IEEE Transactions on Aerospace and Electronic Systems, 2021. Chapter in 2011.

16 VOLUME xx, 2021

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3084215, IEEE Access

K. Sekine et al.: Multi-objective takeoff time optimization using cellular automaton-based simulator

ERI ITOH received her PhD degree in Aero-

nautics and Astronautics from the University of
Tokyo, Japan, in 2017. After her research experi-
ence at Euroconctol Experimental Centre, Nether-
lands Aerospace Centre NLR, NASA Ames Re-
search Center, and Nanyang Technological Uni-
versity, she currently holds positions as an As-
sociate Professor at Aeronautics and Astronautics
Department of the University of Tokyo and Chief
Researcher at Air Traffic Management Depart-
ment of Electronic Navigation Research Institute and National Institute
of Maritime, Port and Aviation Technology. Her research interests cover
automation system design, which collaborates with human operators, in the
air traffic management including airspace and airport operation. Combining
data-driven analysis, mathematical models, and simulation studies, she
works to realize even more resilient air traffic operation. She joins several
academic and administrative communities as committee and/or advisory
members, including International Council of the Aeronautical Sciences
(ICAS), ICAO Surveillance Panel, Airborne Surveillance Working Group,
and Science Council of Japan. Prof. Itoh is a recipient of McCarthy Award
in 2006, John J. Green Award in 2010 from ICAS, and the Distinguished
Service Award from Air Traffic Control Association in Japan, in 2019.

KOZO FUJII received his Ph. D in engineering

in 1980 from the University of Tokyo, Tokyo,
Japan. He was a NRC Research Associate at
NASA Ames Research Center, US, from 1981
to 1983. He was a Senior Research Scientist at
the National Aerospace Laboratory, Chofu, Japan,
from 1984 to 1988. He was an Associate Pro-
fessor in the Institute of Space and Astronauti-
cal Science, Japan Aerospace Exploration Agency
(ISAS/JAXA), Sagamihara, Japan, from 1988 to
1997. He was a Professor in ISAS/JAXA from 1997-2015. Since April 2015,
he has been a Professor at Tokyo University of Science, Tokyo, Japan. He has
been a Fellow of AIAA since 2004 and served as editorial board members
of many journals. He received many awards, such as Daniel Florence
Guggenheim Award from International Council of the Aeronautical Sciences
(ICAS) in 2004. He is one of the 210 members of “Science Council of
Japan.” His research interests are aero-acoustics and flow control.

VOLUME xx, 2021 17

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/