A Dynamic Control Method For Extended Arrival Management Using Enroute Speed Adjustment and Route Change Strategy

Transportation Research Part C 149 (2023) 104064
Contents lists available at ScienceDirect
Transportation Research Part C

journal homepage: www.elsevier.com/locate/trc
A dynamic control method for extended arrival management using

enroute speed adjustment and route change strategy
Ying Huo a,b ,∗, Daniel Delahaye a , Mohammed Sbihi a
a ENAC, Université de Toulouse, 7 Avenue Edouard Belin, Toulouse 31055, France
b Civil Aviation Management Institute of China, Department of General Aviation, Beijing 10012, China
ARTICLE INFO ABSTRACT
Keywords: The continuous growth of air traffic leads to congestion in the surrounding area of airport.
Extended arrival management The optimization of the air traffic in this area requires the improvement of management
Dynamic control strategies, especially for the arrival air traffic. SESAR has proposed the concept of Extended-
Simulated annealing
Arrival MANagement (E-AMAN), which aims to plan the arrival streams from an earlier
Air traffic management
stage in order to achieve delay absorption and earlier planning in the en-route phases of
flights. Based on this concept, we address a dynamic/on-line air traffic control problem to
provide control decisions for flights in order to achieve safety and efficiency during arrival.
The dynamic feature is realized by periodically updating the flight information for aircraft
trajectory predictions. Sub-problems corresponding to each information update are established.
The objective of this problem is to minimize the weighted sum of route congestion and the
conflict resolution workload in the enroute segment, and the number of conflicts in the TMA
(Terminal Maneuvering Area). A dynamic weight assignment approach is applied for ensuring
that as an aircraft get closer to the TMA, its weight for TMA metrics increases accordingly. A
case study based on the arrival traffic of the Paris Charles de Gaulle (CDG) airport is investigated
by using a meta-heuristic simulated annealing algorithm combined with the rolling horizon
approach. Final results are analyzed in terms of the reduction of concerned safety issues, final
control decision distribution, and the time transferred through enroute decision change.
1. Introduction
According to the long-term forecasts done by International Civil Aviation Organization (ICAO) in 2018, global passenger traffic
is expected to grow at 4.2% annually from 2018 to 2038 using the traffic data of 2018 as a baseline. Even though the outbreak
of COVID-19 has caused a huge impact on air transportation, a regular comprehensive assessment of the latest traffic situation in
Europe shows that now 80% of air traffic has regained compared to the level in 2019 (SESAR, 2021a). In a forecast conducted by
EUROCONTROL, air traffic in the transportation network of Europe could recover to the level of 2019 by the end of 2023 (SESAR,
2021b). In the post-pandemic, the growth of air traffic will still lead to congestion in the network, especially for the traffic around the
hub airports. Considering the traffic in the TMA, arrival management has more challenges since it is more likely to experience various
disturbances which further yield accumulated congestion in this small airspace area. In this case, safety risks and the workload of
traffic controllers increase. Considering the fact that the airspace and the relevant infrastructures, as well as the air traffic controllers
are limited resources, the efficiency of arrival management to mitigate the congestion depends on not only making full use of those
resources, but also exploring advanced techniques and management strategies.
∗ Corresponding author at: Civil Aviation Management Institute of China, Department of General Aviation, Beijing 10012, China.
E-mail addresses: huoying@camic.cn (Y. Huo), delahaye@recherche.enac.fr (D. Delahaye), mohammed.sbihi@enac.fr (M. Sbihi).
https://doi.org/10.1016/j.trc.2023.104064
Received 29 December 2021; Received in revised form 13 January 2023; Accepted 15 February 2023
Available online 27 February 2023
0968-090X/© 2023 Elsevier Ltd. All rights reserved.
Y. Huo et al. Transportation Research Part C 149 (2023) 104064
Nowadays, airspace management and arrival management collaborate loosely. The arrival traffic is managed and sequenced
in the airspace around the airport by using the Arrival MANager (AMAN) system. The current AMAN system mainly focuses on
arrival flights that are 100–200 NM away from the airport. Despite the implementation of Ground Delay Program (GDP) or other
measures, common procedures such as holding and stacking are frequently used. Though these procedures are useful, they might
lead to undesirable results such as delay, fuel cost, noise effect, and extra workload for both controllers and pilots. Considering this
fact, an collaborative decision making process incorporating the enroute operations and arrival planning for a specific airport can
be very promising.
Collaborative Decision Making (CDM) is one of the most important concept conceived by sharing information and data between
airport operators, aircraft operators, ground handlers and air traffic controllers. Through broad information awareness, both long-
range planning and tactical decisions can be made. CDM has been successfully implemented for GDP to manage the rate of incoming
flights to an airport. The application of GDP aims to absorb the potential delay on the ground in departure airports rather than in
the air. The capacity of the destination airport, more specifically the arrival profile affects the take-off times of flights for arriving
at this airport to mitigate the congestion around the airport (Ball and Lulli, 2004). Further, CDM is considered to be applied in
the enroute airspace to help with the issues such as lack of analytical support for rerouting and spatial or temporal modifications
along the route (Ball et al., 2001). With the support of CDM, long-range arrival management can be realized by considering the
flight arrival estimates on waypoints shared from the air carriers as well as the service ability at the airport and its surrounding
areas (Hao et al., 2020).
Based on the potential benefits of information sharing, significant works are currently ongoing worldwide. Single European Sky
ATM Research (SESAR) has proposed the concept of Extended-AMAN (E-AMAN) to make arrival management information available
earlier in the flight, allowing any required sequencing or scheduling actions to be taken earlier (SESAR, 2015b). In addition, a similar
concept named Long Range Air Traffic Flow Management (LR-ATFM) conducted trials in the Asia-Pacific (APAC) region based on
the already established ATFM elements (CANSO, 2022). The underlying mechanism in these concepts is the enroute trajectory
adjustment, with which the time and sequence of an aircraft to arrive at the terminal area are well planned so as to alleviate
congestion and improve operational efficiency. The deployment of this process requires high-quality collaboration from multiple
parties such as Air Navigation Service Provider (ANSP), controllers in the enroute sectors, pilots and airline information centers.
The concept can enable a powerful arrival management support tool that takes into account the arrival management information
for a large horizon. In order to verify this solution, validation exercises have been carried out at the London-Heathrow airport.
The results highlighted the efficiency gained in reducing holding times, and environmental benefits in noise emissions and fuel
saving (SESAR, 2019).
Previous studies related to E-AMAN concept include exploring the potential benefits of control strategies such as enroute speed
control or imposing time constraints. Delgado has conducted a series of research aiming at discovering the benefits of speed reduction
in the cruising phase of flights for fuel saving (Delgado and Prats, 2009) and delay absorption (Delgado and Prats, 2012). An extended
study combining the GDP with the cruising speed regulation is proposed to recover the delay that is supposed to be imposed by
GDP. This strategy provides alternative or backup options to balance the capacity in the airport and in the air (Delgado et al.,
2013). The results of their works imply that the core idea of enroute speed reduction shows great potential and benefits in arrival
management. Nikoleris et al. (2012) investigated the distance required for flights to absorb a certain amount of delay by analyzing
flight data that have been implemented with enroute speed control. By comparing with a simulation based delay-free schedule, the
speed control strategy with a speed margin of 10% from the nominal speeds can absorb a considerable amount of delay. Matsuno
and Andreeva-Mori (2020) analyzed the achievable airborne delays by speed control. The results showed that 2–4 min delays per
30 min flight time can be achieved by applying the fuel-equivalent speed, and 2%–3% fuel savings are averagely expected.
Some works provided promising methods for applying E-AMAN concept in order to achieve multiple kinds of interests. Jones
et al. (2013) presented a dynamic multi-objective integer programming model, in which the objective function includes fuel saving
and delay minimization. In this paper, practical control manners are considered for a horizon of 500NM from the airport. Aircraft are
controlled by the initial Expected Time of Arrival (ETA) and the updated Control Time of Arrival (CTA) which are issued depending
on their time distances from the concerned horizon. For determining the priority of specific control movements, two successive
rolling horizon periods (scheduling windows) were introduced. The objective of the most recent window is given a higher weight
than that of the other window. In this way, aircraft near the airport were urged to be compliant with the CTA while the later aircraft
were preferred to be delayed in the enroute segment. Then, a series of trial studies were made, which proved the persuasiveness of
this model. Peters (2017) introduced a practical control tool based on the E-AMAN concept. Using the flight information provided by
ANSP, they apply enroute speed adjustment and alternative route assignment to meet the Required Time of Arrival (RTA) at specific
waypoints. Solution has been validated through live trials, and benefits in terms of delay reduction and emissions are achieved.
Moreover, simulation based model is widely used for the validation of the E-AMAN conception. Swenson et al. (2011) proposed an
advanced air traffic control decision support tool that integrates precise time and trajectory prediction. The vectoring at the meter
fixes, merge points and runways is conducted based on the sequence determined from an extended further upstream management.
Delay is pre-considered as it is expected to be absorbed in the cruising phase. This approach is implemented in the real operational
environment and compared with current Air Traffic Control (ATC) operations under several levels of demand. Schultz et al. (2021)
proposed a mixed-integer linear programming optimization model to minimize the capacity exceeded at the approach sector of
Singapore Changi Airport considering different time shifting abilities in the enroute phase of arrival flights. The inputs of the model
are derived from a statistical analysis of real arrival traffic flying time distribution and the clustered arrival network. In this work,
the optimization is done at a macroscopic level where the operation details and weather effects are not included. Considering the
same concept, Rosenow et al. (2022) established an optimization model for minimizing fuel consumption and the capacity violation
2
in the approach sector by considering the flight decisions of speed, alternate route or both. This work considered three critical time
points before flights arrive at the airport and analyzed the optimal time for conducting this optimization and taking related measures.
Dhief et al. (2020) established an optimization model to absorb arrival delay through enroute speed adjustment. The objective of
this model is to minimize the weighted sum of deviations between assigned delays and achieved delay, speed rate changes and
conflicts of all flights. In this work, the assigned delay times of flights are predicted by using a holding prediction model and a
delay prediction model with the real ADS-B data and flight plans from 200NM to the destination airport. Based on the framework
of Dhief et al. (2020), Jun et al. (2022) provided some improvements including using more abundant arrival data for abstracting
delay information at different distances from the destination airport, incorporating environmental benefits in the objective function
and applying speed control dynamically every 100 NM for flights. Results show that 65% of delay can be absorbed through the
dynamic speed control strategy and great benefits are achieved in fuel saving and emissions reduction. Jones et al. (2018) proposed
three integer programming models: one deterministic baseline model and two stochastic models accommodating uncertainties, to
evaluate the time transferred capability from TMA to enroute phase by assigning CTA. By comparing to the baseline model, results
show that the stochastic models can achieve promising results of delay transfer as well as fuel savings.
Several existing works apply the time constraints for flights arriving at a specific point or using enroute speed adjustment strategy
to evaluate the potential benefits that can be achieved. However, none of them consider the resource planning and operational
situations in the TMA. To this end, based on the E-AMAN concept, we consider the complete flight operation process from 500 NM
to the destination airport until landing. Comment control advisories such as speed regulation (enroute and TMA), route change,
runway assignment are taken as decision variables to achieve coordinated management. In order to trace the constant situation
change while aircraft fly, dynamic features are added to the problem. A key enabler of this process is collaborative information
sharing where information on flights, routes, weather, etc. is provided or updated periodically such that decisions can be modified
according to the circumstances at the time. It is known that 500NM from the airport involves many FIRs. The coordination of
arrival flights at different locations while flying to the destination airport could be a challenge to the application of the proposed
model. While the broadly applied CDM indicates the possibility of information collection and distribution among participants for
collaborative decision-making. Therefore, in this work, we assume seamless information dissemination in the system is guaranteed.
The main contributions of this paper are:
(1) Developing a dynamic/on-line optimization model for arrival traffic to constantly provide updated instructions for flights
from 500NM of the airport until landing so as to achieve a smooth arrival without holding or stacking over the airport.
(2) Introducing a dynamic weighted approach for assigning different weights to the metrics in the objective function considering
the flying process of each aircraft such that the near real-time safety issues can be dealt with priority.
(3) Combining the Simulated Annealing (SA) algorithm with the rolling horizon approach to solve the problem with dynamic
features so that solutions are optimized regularly according to the updated status of flights and environment.
The remainder of this paper is organized as follows: Section 2 describes the problem and introduces the details of the E-AMAN
concept. Section 3 presents the mathematical model for solving the dynamic problem. Then, in Section 4, the optimization solution
for this problem is presented. The case study and the computational results are given in Section 5. Section 6 provides the discussion
of our work. Finally, the conclusion is provided in Section 7.
2. Problem description
The deployment of E-AMAN relies on the current available operational models and systems. The arrival sequencing and
scheduling manager for the TMA helps controllers to optimize capacity by giving associated instructions to pilots for conducting
appropriate maneuvers taking into account safety, fuel consumption and efficiency in the terminal area and at runways. The enroute
decision-making tool optimizes the service provision by assisting with detection and monitoring tasks. As an integration of the
above-mentioned tools, E-AMAN requires all parties to share information using a System-Wide Information Management (SWIM)
service (SESAR, 2019). Based on the SWIM, the systems could provide useful advisories for controllers that are in charge of higher
(cruise) flight levels. Instructions are then delivered to the pilots who make adjustments to aircraft speed such that they meet Time To
Gain (TTG)/Time To Lose (TTL) before reaching the TMA. Fig. 1 shows an application of E-AMAN with geographical information.
The CTA is computed by the AMAN system in accordance with the predicted information, aircraft operation limitations, safety
requirements etc. Then, it is proposed to the pilots by the controllers, based on which flight on-board computer is used to achieve
required accuracy (SESAR, 2015a). Through enroute speed control flight sequences are preplanned and then perform STAR in the
TMA. Note that flights with low operational ranges are limited in adjusting arrival times through speed regulation. Therefore,
according to whether the flight horizon exceeds the limit of 500NM, flights are classified as long-haul flights and short-haul flights.
Adapted to the E-AMAN concept, this study focuses on the arrival traffic within the boundary of 500NM from the destination
airport. The arrival routes are abstracted as a node-link network. Aircraft that enter the boundary are operated on the enroute
network, then they follow the route in the TMA, and further execute the approach procedures. In this study, we use IAF (Initial
Approach Fix) in the TMA which are the final merging points of traffic flows to identify the two segments (enroute and TMA),
of which the safety assessment methods are differentiated. We propose a mechanism that focuses on the real time controls
including speed regulation, alternative route assignments, and runway assignments. In order to create an ideal traffic management
environment, holding and stacking are not considered in our system. It is known that information sharing is important for controllers
to deliver proper commands. Therefore, the integrated problem aiming at minimizing the safety issues in both enroute phase and
TMA is addressed. Moreover, in real operations, the position and operating parameters of an aircraft keep changing with time, which
3
Fig. 1. Application of Extending AMAN from 500NM to the destination airport by Assigning a CTA (Peters, 2017). Arrival traffic flying through enroute phase
and then perform STAR in the TMA.
increases the difficulty in solving the dynamic/on-line problem. In this context, the proposed algorithm makes periodic updates based
on the previous schedule to take into account the newly upcoming flights and keep track of the aircraft trajectories that are already
in the network. The air traffic is then expected to be smoothly operated on the concerned horizon without generating extra holding
time while accomplishing the tasks of minimizing potential safety issues.
3. Mathematical model
3.1. Network abstraction
It is known from the concept of E-AMAN, the flights that arrive at the same destination within a horizon of approximately 500NM.
Taking the Paris CDG airport as an example, the trajectories of the arrival flights within 500NM from the airport are displayed in
Fig. 2. The 𝑥-axis and 𝑦-axis display the longitude and latitude. In this figure, flights that come from the east side are gradually
merged into the main streams through several waypoints, while flights that come from the west side follow a free routing strategy
can be directed from one waypoint directly to another. That is why the trajectories from the east side of the Paris CDG airport are
more dispersed.
In order to simplify the network, routes that represent the main streams of traffic are abstracted according to the real flight
trajectories inside the E-AMAN horizon. The network abstraction is represented as a graph  = ( , ) with the node set  and link
set . In the network,  is constituted of several subsets, and we have  = 𝑏 ∪ 𝑐 ∪ 𝑒 ∪ 𝑡 ∪ 𝑟 , where 𝑏 denotes the node set
of boundary waypoints, 𝑐 represents the node set in the enroute segment, 𝑒 stands for the node set that defines the boundary of
the enroute segment and the TMA, 𝑡 represents the nodes extracted from the arrival route in the approach phase, and 𝑟 denotes
the set of runway thresholds. We assume that the network is divided into two main regions, the enroute network and the terminal
network, taking the nodes in 𝑒 as the transition points. The nodes are connected by links. In line with the predefined network
regions, the link set is defined as  = 𝑐 ∪ 𝑡 , where 𝑐 and 𝑡 denote the link set in the enroute segment and the link set in the
TMA, respectively.
Each route 𝑢 ∈  consists of two sub-routes 𝑢𝑒 and 𝑢𝑡 which refer to the route in the enroute segment and in the TMA respectively.
In the enroute segment, 𝑢𝑒 is composed of a set of nodes (𝑛𝑢1 , 𝑛𝑢2 , … , 𝜈𝑒 ), and a set of links (𝑙1𝑢 , 𝑙2𝑢 , …) that connect those nodes from the
boundary until the TMA entry node. In the node set, 𝑛𝑏 ∈ 𝑏 is the first node that an aircraft passes to enter the system, therefore
we can also replace 𝑛𝑢1 to 𝑛𝑢𝑏 . We denote 𝑛𝑢𝑖 as the 𝑖th node that an aircraft would pass by following the route 𝑢 in the enroute phase.
Moreover, we have 𝜈𝑒 ∈ 𝑒 . For links, we denote that 𝑙𝑖𝑢 ∈ 𝑐 is an enroute link with 𝑖 indicating the sequence order of this link
on route 𝑢. Similarly, a route 𝑢𝑡 in the TMA consists of a set of nodes (𝜈𝑒 , 𝜈1 , 𝜈2 , … , 𝜈𝑟 ) and a set of links (𝜇1 , 𝜇2 , …), where 𝜈𝑟 ∈ 𝑟
and 𝜇𝑖 ∈ 𝑡 . The TMA route is uniquely defined by specifying the TMA entry node (𝜈𝑒 ) and a landing runway (𝜈𝑟 ). Each aircraft
follows exactly one route to arrive at the airport. An example of the notations representing different resources on a random route
is illustrated in Fig. 3. In addition, the abstracted network of the whole system is shown in Fig. 4, including the networks of the
4
Fig. 2. Trajectories of all flights arriving at the Paris CDG airport on Feb 10th, 2017. Only the trajectories within 500NM of the airport are displayed.
Fig. 3. Example of the resource notations of a random route from a boundary entry point until a runway threshold. The whole route is divided into two parts,
enroute and TMA.
enroute segment and the TMA. In order to better simulate the aircraft operation process, the 𝑥-axis and 𝑦-axis represent the distance
with a unit of nautical mile with respect to a specific reference point. In this study, we assume that the speeds of aircraft keep
constant on each link in the enroute phase, and once aircraft enter the TMA, a constant deceleration operation is performed.
3.2. Given information of flights
For a flight 𝑓 in the arrival flight set  , the following data considered as input are given:
5
Fig. 4. Arrival routes network abstraction for the enroute phase and the TMA within 500NM from the destination airport. Enroute network and TMA network
are indicated in blue and red, respectively.
• 𝑐𝑓 : aircraft categories (heavy, medium, light).

• 𝑛𝑏𝑓 : boundary node, from which the flight 𝑓 enters the E-AMAN horizon. For short-haul flights, this represents the departure
airport.
• 𝜅𝑓 : flight type for identifying whether flight 𝑓 is a long haul or short haul flight.
• 𝑓 : arrival route set of flight 𝑓 in the cruising phase. Note that flight can be assigned to an alternate route from one of the
waypoints in the enroute phase such that path stretching can be applied. Hence, multiple route options could be included.
• ℎ𝑓 : the sequence order of a specific waypoint on the nominal route that aircraft 𝑓 operated on. At this waypoint, an alternative
route starts and deviates from the nominal route. If there is no such waypoint on this route, ℎ𝑓 is set to be −1.
• 𝑡𝑏𝑓 : boundary entry time of flight 𝑓 .
• 𝑣𝑜𝑝𝑡
𝑓
: nominal cruising speed obtained based on the general aircraft performance and environment condition, this speed is
associated with the aircraft type.
• 𝜈𝑓𝑒 : TMA entry node of flight 𝑓 .
• 𝑟𝑓 : initial landing runway of flight 𝑓 .
3.3. Decision variables and constraints
Due to the differences in geographical structure and operational constraints of the cruise phase (enroute) and the approach phase
(TMA), decision variables are specified accordingly.
3.3.1. Decision variables for enroute segment

Considering the instructions that the air traffic controllers usually give to the pilots, two kinds of decision variables are applied
in the enroute segment, which are:
• Alternative route 𝑢𝑓 :
𝑢𝑓 ∈ 𝑓 (1)
6
An alternative route usually diverts from a waypoint which can increase or decrease aircraft flying distance in the cruising
phase, leading to a gain or loss of time when arriving at the TMA. If an aircraft has not passed the waypoint that leads to an
alternative route, the aircraft can be assigned to any of the possible routes in the enroute segment. Note that not all aircraft
have an option of changing route, as for some aircraft, only a single route is available to go to the TMA.
𝑢
• Enroute link speed 𝑣𝑙 𝑓 , where 𝑖 is a sequential indicator representing the 𝑖th link on route 𝑢𝑓 that aircraft 𝑓 passes through:
𝑖
𝑢
𝑣𝑙 𝑓 ∈ 𝑉𝑓 ∶= {𝑣𝑜𝑝𝑡
𝑓
+ 𝑗𝛥𝑣𝑓 |𝑗 ∈ Z and − 10 ⩽ 𝑗 ⩽ 10} (2)
𝑖
where 𝛥𝑣𝑓 is a speed adjustment step that is set to be 0.01𝑣𝑜𝑝𝑡𝑓

. The speed range is determined by referring to some literatures
that takes into account the aircraft structural limit and the common speed range used for aircraft. In Delgado and Prats (2012),
a speed reduction concept that aims to absorb the delay in the cruising phase without increasing fuel consumption is proposed.
Through a comprehensive study of parameters such as flight level, cost index, and aircraft category, they stated that a speed
reduction up to −12% can be achieved with respect to the nominal speed for certain types of flights. Moreover, Nikoleris et al.
(2012) has stated that a speed reduction up to 10% of nominal cruising speed is the common expected range of speed reduction
in the cruising phase. Nikoleris et al. (2012) also applied a speed reduction up to 10% in the cruising phase to investigate the
ability of delay absorption of aircraft. Further, in Huang and Tomlin (2009), a speed variation between 12% to −10% of the
nominal speed is applied to solve the conflicts in the sectors. Considering all these possible choices, in our problem, the speed
constraints are set with a range indicated in Eq. (2).
3.3.2. Decision variables for the TMA

Decision variables for the TMA can refer to our previous study that addressed the scheduling problem in the TMA (Huo et al.,
2021). One key difference in decision variables selection from the previous study is that the margin of TMA entry time is not
considered. We assume that through enroute adjustment, required time transfer can be achieved in this model, and a smooth
operation from the enroute phase to the approach is permitted without assigning holding and stacking. Therefore, two decision
variables are implemented to eliminate conflicts in the TMA:
• TMA entry speed 𝑣𝑡𝑓 :
𝑣𝑡𝑓 ∈ 𝑉𝑓𝑡 ∶= {𝑣𝑜𝑝𝑡𝑅

𝑓 𝑓 |𝑗 ∈ Z and − 10 ⩽ 𝑗 ⩽ 10}
+ 𝑗𝛥𝑣𝑅 (3)
where 𝑣𝑜𝑝𝑡𝑅
𝑓
denotes the nominal speed for flights when arriving at the TMA entry point and 𝛥𝑣𝑅
𝑓
= 0.01𝛥𝑜𝑝𝑡𝑅
𝑓
.
• Landing runway 𝑟𝑓 :
𝑟𝑓 ∈  𝑟 (4)
During the approach phase, aircraft are assigned different runways to ensure sufficient usage of runway capacity.
The decision vector for all flights is represented as 𝐱. The decisions vector for a flight 𝑓 is denoted as 𝐱𝑓 , which contains all the
𝑢 𝑢
decision variables of this flight: 𝐱𝑓 = (𝑢𝑓 , 𝐯𝑙 𝑓 , 𝑣𝑡𝑓 , 𝑟𝑓 ), where 𝐯𝑙 𝑓 is a set of speeds of 𝑓 associated with the enroute links. We denote
𝑖 𝑖
𝑢
the 𝑖th link of aircraft 𝑓 in the enroute phase when following the route 𝑢𝑓 as 𝑙𝑖 𝑓 , which implies that route decision 𝑢𝑓 of 𝑓 should
be specified first.
3.4. Metrics evaluation
The evaluation metrics proposed for the enroute phase and the TMA are detailed. The essential consideration is to avoid the
violation of airspace capacity and avoid separation losses. In this study, considering the operational characteristics in different
phases, enroute metrics are set at a macroscopic level, while the metrics for the TMA are more tactical.
3.4.1. Traffic flow-based evaluation in the enroute segment

For our problem, only the arrival flights to a specific destination are concerned. These flights belong to a part of the total enroute
traffic. In this case, the maximum capacity of the air traffic control sector is not suitable for the safety evaluation. Therefore, we
propose a traffic flow-based approach to evaluate the capacity violation of the arrival traffic in the route network by counting the
number of flights passing a specific link within a fixed time interval. In the process of traffic flow-based evaluation, 𝛥 (e.g. 1 min)
is set as a time unit that constitutes the evaluation time interval which is set as 𝐿 = 𝑞 ⋅ 𝛥, where 𝑞 is a coefficient to define the
length of the evaluation time interval. For each time interval, the concerned events such as aircraft arriving at a specific waypoint
are focused. As shown in Fig. 5, each horizontal line indicates an evaluation time interval 𝐿 that starts from a time 𝑡𝑖 . The number
of the concerned events in a time period [𝑡𝑖 , 𝑡𝑖 + 𝐿] are counted and denoted as 𝐸(𝑡𝑖 ). The events are represented by the black arrows
located on the time horizon with respect to their occurring time. A continuous moving of the evaluation time interval is proceeded
by shifting one time step 𝛥 afterward.
7
Fig. 5. Traffic flow-based evaluation with the time period rolling scheme to evaluate enroute congestion metric. The concerned events are represented by the
black arrows, the number of the events corresponding to each time window are counted and used for the congestion evaluation.
Fig. 6. Resolution workload evaluation on node using traffic flows coming from the parent links and the angle constructed by these two parent links.
3.4.2. Congestion evaluation

According to the above-mentioned flow-based evaluation method, the congestion of the links associated with the enroute segment
is evaluated. The concerned events refer to aircraft passing through the link entry nodes. Through this way, the air traffic density on
a link can be estimated. In this approach, one important thing is to decide the length of the evaluation time interval. A proper time
length not only can capture the traffic density variation but also avoid information loss caused by long time averaging. Since traffic
gradually converges to the destination airport, the arrival rate can be small on links that are far away from the airport and can be
high on links near the airport. We think 10 minutes seems to be a suitable time interval to be applied for detecting the number of
flights entering a link. Therefore, we set 𝑞 equals to 10, and the number of flights that enter link 𝑙 from 𝑡 to 𝑡 + 𝐿 can be expressed
as follows:
∑
𝑠=(𝑞−1) ∑
𝐸𝑙 (𝑡) = 𝜔𝑡+𝑠𝛥
𝑓 ,𝑙
𝑙 ∈ 𝑐 , (5)
𝑠=0 𝑓 ∈
where 𝜔𝑡+𝑠𝛥
𝑓 ,𝑙
is a binary parameter that equals to 1 if flight 𝑓 is predicted to enter (or to exit) link 𝑙 at the time slot of 𝑡+𝑠𝛥, 𝑠 ∈ [0, 𝑞−1];
otherwise equals to 0. Taking Fig. 5 as an example, the black arrows in this figure can be interpreted as the times when aircraft pass
a specific link entry node, then the number of aircraft that enter the link during time period 𝐿 starting from time 𝑡𝑖 are counted, in
this example, 𝐸𝑙 (𝑡𝑖 ) = 𝐸(𝑡𝑖 ).
The horizontal separation standard between aircraft flying at the same altitude is 5NM in the controlled enroute airspace
prescribed in ICAO (2016). Based on which, the link capacity for a time period of 𝐿 can be determined by transferring the distance
based requirement to a time based one. The capacity is then defined as:
𝑀 ⋅ 𝑣𝑙
𝑅𝑙 = (6)
𝐾
where 𝑀 refers to the separation constraint of one flight per 5NM. 𝐾 is a parameter that is determined by computing the proportion
of the time period 𝐿 to one hour. As 𝐿 = 10 min, in this case, 𝐾 = 6. 𝑣𝑙 represents the average speed of all aircraft operated on link
𝑙. Let us define a congestion indicator 𝜙𝑙 (𝑡), which is the excess number of aircraft over the capacity 𝑅𝑙 for a time period from 𝑡 to
𝑡 + 𝐿.
𝜙𝑙 (𝑡) = max[(𝐸𝑙 (𝑡) − 𝑅𝑙 ), 0], 𝑙 ∈ 𝑐 (7)
If 𝐸𝑙 (𝑡)−𝑅𝑙 is bigger than 0, congestion is registered for the time period [𝑡, 𝑡+𝐿]. Otherwise, there is no congestion. The evaluation
will be carried out on all enroute links for the rolling time periods on the time horizon.
3.4.3. Resolution workload evaluation

In the arrival route network, apart from capacity management, another task of the air traffic controllers is to coordinate the
traffic merging. Considering the workload, a limited number of aircraft arriving at the same waypoint from different directions can
8
Table 1
Wake turbulence separation minima for different categories of aircraft 𝑠𝑓 ,𝑔 , with associated
aircraft categories, in NM, (ICAO, 2016).
Categories Leading Aircraft 𝑓
Heavy Medium Light
Heavy 4 3 3
Trailing
Medium 5 3 3
Aircraft, 𝑔
Light 6 5 3
Table 2
Single runway separation requirements, 𝜏𝑓𝑟𝑐,𝑔 , with associated aircraft categories, in seconds,
(Frankovich, 2012).
Categories Leading Aircraft 𝑓
Heavy Medium Light
Heavy 96 60 60
Trailing
Medium 157 69 69
Aircraft, 𝑔
Light 207 123 82
be managed. We evaluate the associated workload based on the metric proposed in Delahaye and Puechmorel (2013). For a node
where two or more links converge, the conflict resolution workload is evaluated based on the air traffic coming from the preceding
links. The links that are incident on a node can be called parent links of the node. Suppose that two links 𝑙𝑖 and 𝑙𝑗 are parent links of
node 𝑛 which belong to the parent link set 𝑛 of node 𝑛. The average aircraft speeds on the links are given by 𝑣𝑙𝑖 and 𝑣𝑙𝑗 respectively.
𝐸𝑙𝑂𝑢𝑡 (𝑡) and 𝐸𝑙𝑂𝑢𝑡 (𝑡) are the numbers of flights that exit the parent links during the time interval [𝑡, 𝑡 + 𝐿]. The conflicts resolution
𝑖 𝑗
workload 𝐸𝑛 (𝑡) from time [𝑡, 𝑡 + 𝐿] can be computed as follows:
√
2 2
∑ 2𝑁𝑠 𝑣𝑙𝑖 − 2𝑣𝑙𝑖 𝑣𝑙𝑗 cos 𝜃𝑖,𝑗 + 𝑣𝑙𝑗
𝐸𝑛 (𝑡) = ⋅ 𝐸𝑙𝑐,𝑂𝑢𝑡 (𝑡)𝐸𝑙𝑐,𝑂𝑢𝑡 (𝑡) 𝑛 ∈ 𝑐 , (8)
𝑙 ,𝑙 ∈
𝑣𝑙𝑖 𝑣𝑙𝑗 sin 𝜃𝑖,𝑗 𝐾 𝑖 𝑗
𝑖 𝑗 𝑛
𝑙𝑖 ≠‴𝑗
where 𝑁𝑠 is the standard separation norm, 𝜃𝑖,𝑗 denotes the angle formed by two convergent flows associated with link 𝑙𝑖 and 𝑙𝑗 . In
this equation, we assume that the unit of the speed is NM/H and 𝑁𝑠 = 5NM. Moreover, 𝐸𝑙𝑐,𝑂𝑢𝑡 (𝑡) and 𝐸𝑙𝑐,𝑂𝑢𝑡 (𝑡) count the number of
𝑖 𝑗
aircraft for a time period of 𝐿, therefore, 𝐾 is used to convert the unit from hour to 𝐿 (10 min). 𝐸𝑛 (𝑡) represents the average number
of conflicts at the crossing node 𝑛 for the time [𝑡, 𝑡 + 𝐿]. Fig. 6 displays the situation corresponding to the pre-mentioned description,
where the conflict resolution workload is evaluated. If more than two parent links exist for a node, the conflict resolution workload
is just the sum of the workloads induced by the possible pairs of the parent links. The indicator of the excess of the resolution
workload 𝜙𝑛 (𝑡) is given as:
𝜙𝑛 (𝑡) = max[(𝐸𝑛 (𝑡) − 𝑅𝑛 ), 0], 𝑛 ∈ 𝑐 , (9)
where 𝑅𝑛 is the maximum conflict resolution workload. It is known that the workload is hard to be evaluated. In order to estimate
a suitable value for 𝑅𝑛 , we set up an upper limit by taking into account the aircraft speeds, the maximum number of aircraft that
can exit from a link during 𝐿 and a set of angles that appears in the network for evaluation. Finally, 𝑅𝑛 is set to 1.34, namely, the
potential conflicts of air traffic flow flying to the same point need to be lower than 8 for each hour.
3.4.4. Conflict detection in the TMA

To ensure flight safety, three separation requirements are considered: wake turbulence constraints, horizontal separation, and
runway separation.
• Wake turbulence separation: the separation between two successively operated aircraft 𝑓 and 𝑔 should satisfy the wake
turbulence separation minima 𝑠𝑓 ,𝑔 associated with their aircraft categories 𝑐𝑓 , 𝑐𝑔 . The minimum separation standards are
given in Table 1.
• Horizontal separation: Aircraft must satisfy a minimum horizontal separation ℎ = 3𝑁𝑀 based on the radar separation in the
TMA when they are not flying on the same link.
• Runway separation: The runway separation requirements specified for two aircraft landing at the same runway with respect
to their associated aircraft categories are given in Table 2 (Frankovich, 2012). These requirements are also related to the
associated occupancy times of aircraft on the runway.
A violation of the aforementioned separation rules gives rise to three types of conflicts. A link conflict refers to a violation of
minimum wake turbulence separation between two successive aircraft flying on the same link. Catch-up conflicts and overtake
conflicts are examples of such conflicts. A node conflict refers to the loss of horizontal separation between two aircraft on two
different links adjacent to the same node. Finally, a runway conflict refers to the loss of runway separation between two aircraft
landing on the same runway.
9
Fig. 7. Node conflict detection scenarios regarding different trajectory interaction patterns of the flight pair.
Fig. 8. Link conflict detection configuration for two aircraft entering the same link consecutively.
In order to adapt to a time-based metering system, which is convenient for conflict detection and metering the flows, the following
lemmas transfer the distance-based separation to a time-based separation. Their proofs can be found in our previous work (Huo et al.,
2021). Suppose that (𝑓 , 𝑔) is a pair of distinct aircraft operated consecutively.
Lemma 1. Let 𝜈 be a node belonging to the routes of 𝑓 and 𝑔. Assume that aircraft 𝑓 passes node 𝜈 before 𝑔. Then, there is no horizontal
conflict between 𝑓 and 𝑔 at node 𝜈 if and only if:
{
𝑛𝑐 𝑎, if 𝑓 and 𝑔 enter 𝜈 through the same link (see Fig. 7(a)),
𝑡𝑔,𝜈 − 𝑡𝑓 ,𝜈 ≥ 𝜏𝑓 ,𝑔,𝜈 ∶= (10)
max(𝑎, 𝑏), if 𝑓 and 𝑔 enter 𝜈 from different links (see Fig. 7(b)).
where
⎧ ℎ
⎪ 𝑣𝑔,𝜈 , if 𝑣𝑔,𝜈 ≤ 𝑣𝑓 ,𝜈 cos 𝜃,
⎪
⎪ ℎ if 𝑣𝑓 ,𝜈 ≤ 𝑣𝑔,𝜈 cos 𝜃,
𝑎 = ⎨ 𝑣𝑓 ,𝜈 ,
⎪ √ 2 2
⎪ ℎ (𝑣𝑓 ,𝜈 +𝑣𝑔,𝜈 −2𝑣𝑓 ,𝜈 𝑣𝑔,𝜈 cos 𝜃) , otherwise
⎪ 𝑣𝑔,𝜈 𝑣𝑓 ,𝜈 sin 𝜃
(11)
⎩
⎧ ℎ , if 𝑣𝑓 ,𝜈 cos 𝜑 ≤ 𝑣𝑔,𝜈 ,
⎪ 𝑣𝑔,𝜈
𝑏 =⎨ √ 2
ℎ (𝑣𝑓 ,𝜈 +𝑣2𝑔,𝜈 −2𝑣𝑓 ,𝜈 𝑣𝑔,𝜈 cos 𝜑)
⎪ , otherwise.
⎩ 𝑣𝑔,𝜈 𝑣𝑓 ,𝜈 sin 𝜑
The angles 𝜃 and 𝜑 are shown in Fig. 7. Moreover, 𝑣𝑓 ,𝜈 , 𝑣𝑔,𝜈 denote the speeds of aircraft 𝑓 and 𝑔, respectively, at node 𝜈. We remind
that ℎ represents the horizontal separation requirement which equals to 3NM.
Lemma 2. Let 𝜇 = (𝜈, 𝜔) be a link belonging to the routes of 𝑓 and 𝑔. Assume that aircraft 𝑓 enters the link 𝜇 before 𝑔 as shown in Fig. 8.
The average speeds on 𝜇 for both aircraft are computed and denoted as 𝑣𝑓 ,𝜇 and 𝑣𝑔,𝜇 . Then, there is no conflict between 𝑓 and 𝑔 on link 𝜇
if and only if:
( )
𝑠𝑓 ,𝑔 𝑠𝑓 ,𝑔 𝑧(𝑣𝑔,𝜇 − 𝑣𝑓 ,𝜇 )
𝑡𝑔,𝜈 − 𝑡𝑓 ,𝜈 ≥ 𝜏𝑓𝑙𝑐,𝑔,𝜇 ∶= max , + (12)
𝑣𝑓 ,𝜇 𝑣𝑔,𝜇 𝑣𝑔,𝜇 𝑣𝑓 ,𝜇
where 𝑧 represents the length of the link 𝜇. Remind that 𝑡𝑓 ,𝜈 and 𝑡𝑔,𝜈 represent the times that aircraft 𝑓 and 𝑔 enter the link 𝜇, respectively.
𝑠𝑓 ,𝑔 is the wake turbulence separation requirement associated with the categories of aircraft 𝑓 and 𝑔 (see Table 1).
10
Fig. 9. The relations of the ground speed, true airspeed and the wind speed, in which the wind speed direction is uncertain. The ground speed of the aircraft
can be obtained by adding the wind speed vector projected to the same direction of the true airspeed.
Lemma 3. If 𝑓 and 𝑔 land at the same runway 𝑟 and 𝑓 lands before 𝑔, then there is no runway conflict between 𝑓 and 𝑔 if and only if:
𝑡𝑔,𝑟 − 𝑡𝑓 ,𝑟 ≥ 𝜏𝑓𝑟𝑐,𝑔 (13)

where 𝜏𝑓𝑟𝑐,𝑔 is given in Table 2 determined by the associated aircraft categories of aircraft 𝑓 and 𝑔. 𝑡𝑔,𝑟 and 𝑡𝑓 ,𝑟 represent the arrival times of
aircraft 𝑔 and 𝑓 at the runway threshold.
Based on the above-mentioned lemmas, we introduce the following conflict indicators for each pair of aircraft (𝑓 , 𝑔) when they
are in the TMA.
• Node conflict indicator

{
1, if 0 ≤ 𝑡𝑔,𝑣 − 𝑡𝑓 ,𝜈 ≤ 𝜏𝑓𝑛𝑐,𝑔,𝜈 or 0 ≤ 𝑡𝑓 ,𝜈 − 𝑡𝑔,𝜈 ≤ 𝜏𝑔,𝑓
𝑛𝑐 , ∀𝜈 ∈ 𝑢𝑡𝑓 (𝐱) ∩ 𝑢𝑡𝑔 (𝐱) ∩ {𝑡 ∪ 𝑒 }
𝜙nc
𝑓 ,𝑔,𝜈
= ,𝜈 (14)
0, otherwise
• Link conflict indicator,

{
lc 1, if 0 ≤ 𝑡𝑔,𝜈 − 𝑡𝑓 ,𝜈 ≤ 𝜏𝑓𝑙𝑐,𝑔,𝜇 or 0 ≤ 𝑡𝑓 ,𝜈 − 𝑡𝑔,𝜈 ≤ 𝜏𝑔,𝑓
𝑙𝑐
,𝜇
, ∀𝜇 ∈ 𝑢𝑡𝑓 (𝐱) ∩ 𝑢𝑡𝑔 (𝐱) ∩ 𝑡 ,
𝜙𝑓 ,𝑔,𝜇 = (15)
0, otherwise 𝜇 = (𝜈, 𝜔),
• Runway conflict indicator

{
1, if 0 ≤ 𝑡𝑔,𝑟 − 𝑡𝑓 ,𝑟 ≤ 𝜏𝑓𝑟𝑐,𝑔 or 0 ≤ 𝑡𝑓 ,𝑟 − 𝑡𝑔,𝑟 ≤ 𝜏𝑔,𝑓
𝑟𝑐 , if 𝑟𝑓 = 𝑟𝑔 = 𝑟 ∈ 𝑟 ,
𝜙rc
𝑓 ,𝑔
= (16)
0, otherwise
The abusing notation 𝑢𝑡𝑓 (𝐱) ∩ 𝑢𝑡𝑔 (𝐱) ∩ {𝑡 ∪ 𝑒 } (resp. 𝑢𝑡𝑓 (𝐱) ∩ 𝑢𝑡𝑔 (𝐱) ∩ 𝑡 ) designates the set of common nodes (resp. links) between
the routes of 𝑓 and 𝑔 under current decisions 𝐱.
3.5. Wind effects
Flights are always subject to uncertainties coming from themselves and the environment. In this study, we have considered the
influence of wind, which is the most common factor that affects the aircraft trajectory in the enroute phase by affecting the speed
of the aircraft. As a matter of fact, the wind is a very dynamic factor, even if there are accurate predictions. In order to simulate
the wind, it is characterized by randomly changing in a range of 𝑣𝑊 ∈ [𝑣min𝑊
, 𝑣max
𝑊
]. The ground speed of the flights on each link is
therefore a vector composition of the true airspeed vector of aircraft and wind vector on the route direction:
𝑣⃖⃖⃖⃖⃖
𝐺⃗
𝑓 ,𝑙 = 𝑣 𝑓 ,𝑙⃗ + 𝑣
⃖⃖⃖⃖⃖ 𝑊⃗.
⃖⃖⃖⃖⃖ (17)
where 𝑣⃖⃖⃖⃖⃖
𝐺 ⃗ is the ground speed vector on the route direction, 𝑣
𝑓 ,𝑙 𝑓 ,𝑙⃗ denotes the aircraft true airspeed on the route direction, and 𝑣
⃖⃖⃖⃖⃖ ⃖⃖⃖⃖⃖
𝑊⃗
represents the wind component on the route direction. The ground speed is considered as the actual operating speed of each aircraft
for the cruising phase and is used for related computations. Fig. 9 shows the relations of wind speed, true aircraft speed and ground
speed.
3.6. Objective function
The objective function is set as a weighted sum of the previously introduced metrics, which is given as:
{∑
𝑇 ∑ ∑ }
𝐺(𝐱) = 𝛼 ( 𝜙𝑙 (𝑡) + 𝜙𝑛 (𝑡))
𝑡 𝑙∈𝑡 𝑛∈𝑐
{ ∑ ∑ ∑ ∑ } (18)
+𝛽 ( 𝜙nc
𝑓 ,𝑔,𝜈
+ 𝜙lc
𝑓 ,𝑔,𝜇
)+ 𝜙rc
𝑓 ,𝑔
𝑓 ,𝑔∈ , 𝜈∈𝑢𝑓 ∩𝑢𝑔 ∩{𝑡 ∪𝑒 } 𝜇∈𝑢𝑓 ∩𝑢𝑔 ∩𝑡 𝑓 ,𝑔∈ ,
𝑔>𝑓 𝑔>𝑓 |𝑟𝑓 =𝑟𝑔
where 𝛼 and 𝛽 are the weights for the metrics associated with the enroute phase and the TMA, respectively. More specifically,
the first term represents the weighted value of enroute metric including the sum of congestion of all links and the sum of conflict
resolution workload detected at enroute nodes for all time steps 𝑡 ∈ 𝑇 . The second term shows the sum of node conflict, link, conflict
and runway conflict detected in the TMA.
11
Fig. 10. Dynamic framework with rolling horizon approach. Rolling horizon starts from time 𝑡𝑠 with a length of 𝜏. Each horizon rolling can be interpreted
as an update of the current time, for which information update and aircraft trajectories prediction are carried out. The prediction horizon is an unfixed value
computed based on the operational parameters and current locations of the aircraft in the system.
3.7. Dynamic arrival management
Flight operation is an ongoing and dynamic process subject to uncertainty filling around. Air traffic controllers usually give
instructions to the pilots according to the revised information both from the flights and the environment. Based on this fact, the
newly available information needs to be involved in the assessment of future situations in order to achieve collaborative control.
To realize this concern, we propose a dynamic control model that combines static sub-problem solving and a dynamic weighted
approach. Through these designs, reliable decisions can be provided to deal with recent problems while considering future situations.
3.7.1. Dynamic control framework

In a dynamic environment, the positions, speeds and directions of the aircraft that need to be managed are constantly changing.
In order to have a high level of accuracy in traffic prediction, it is necessary to constantly update the information required by the
controllers to achieve situational awareness. The dynamic features of this problem are captured by two actions. First, the list of
flights that are scheduled to arrive in the system within a certain amount of time is updated. Second, we keep track of the trajectory
information and operational parameters of the flights that are already in the system. A typical approach for such dynamic problem
is the rolling horizon (sometimes it is referred to as receding horizon) approach. The horizon is defined as a short time period to
ensure that air traffic controllers have accurate knowledge of aircraft operations. Considering that the ANSP would update the list
of flights that have to be scheduled every 15–30 min (Jones et al., 2013), we choose a time period of 𝜏 = 20 min as the length of
the rolling horizon. After each 𝜏 units of time, the considered horizon is rolled, and the aircraft information would be updated. The
time, at which the information update should be performed is referred to as update time. Suppose that 𝑡𝑠 and 𝑡𝑒 are the start time
and the end time of the overall time horizon, then the 𝑘th update time (𝑘th information update) corresponds to a current time of
𝑡𝑐 = 𝑡𝑠 +(𝑘−1)𝜏. Based on this framework, the algorithm is established by solving the sub-problems associated with each information
update. Two key steps are involved to establish each sub-problem:
• Information update: this process includes specifying the list of flights that have to be scheduled into the E-AMAN bound within
𝜏 units of time, and updating trajectory information of aircraft already in the system. The trajectory information of the aircraft
already in the network is tracked from their beginning locations associated with the previous update time to the locations of
the current time based on the previously assigned decisions. Moreover, any aircraft that has landed before the current time is
removed from the network.
• Aircraft trajectory prediction: this is carried out to achieve a comprehensive awareness of the air traffic situation based on the
updated information. All aircraft in the network are simulated from their current locations until they land on the runway. The
sub-problem is then established based on the current and predicted information aiming at generating suitable control decisions
to ensure safe and efficient air traffic operation. We denote a prediction horizon 𝑡𝑟 as the longest time required for all aircraft
in the network to reach the runway threshold from their current positions.
Fig. 10 illustrates the rolling horizon approach. The vertical dash lines on the time horizon indicate the update time. For each
update time, the prediction horizon 𝑡𝑟 is varied and computed based on the aircraft operational parameters including the decisions
obtained from the optimization of the previous time window and the current aircraft locations.
Another critical concern is the determination of the freezing zone. In some literature, the freezing zone is usually set with
time (Murça and Müller, 2015; Bennell et al., 2017). Considering the context of our problem, the freezing zone is chosen as a
geographical area, in which the decisions of aircraft are fixed. In our model, the operation speed profile of each aircraft in the TMA
only depends on its two decision variables: TMA entry speed and the runway. These two decision variables are determined before
an aircraft gets in the TMA, therefore the terminal area can be naturally considered as the freezing zone. Note that at any update
time, some aircraft are operated in the TMA, these aircraft are equally simulated and involved in the conflict detection process in
order to guarantee overall safety.
12
In the rolling horizon framework, it is necessary to define a parameter for each aircraft that can explicitly give rise to its flying
progress in the network. It is worth mentioning that, we assume that the control speeds of an aircraft are given on each link,
therefore, it is more practical to associate the aircraft position with a link instead of obtaining an exact distance or coordinate.
Considering that we are now at 𝑘th update time, we denote 𝑝𝑘𝑓 the progress parameter of an aircraft 𝑓 , which can be interpreted as
the sequence order of the link that 𝑓 is currently on, with respect to the chosen route. In summary, at the current time 𝑡𝑠 + (𝑘 − 1)𝜏,
aircraft can be classified according to the following states:
• Pre-processing: aircraft that are scheduled to pass the boundary entry node within 𝜏 units of time. All aircraft in the pre-
processing state constitute the flight set 𝑝 . Since aircraft in this state have not entered any link, for an arbitrary aircraft 𝑓
belonging to 𝑝 , its progress parameter at the 𝑘th update time is:
𝑝𝑘𝑓 = 0, 𝑓 ∈ 𝑝 , (19)
• Enroute: aircraft that have passed the boundary entry node but did not enter the TMA yet belong to the enroute flight set 𝑒 .
In this state, aircraft trajectories are simulated from the current positions based on the control inputs obtained from previous
optimizations. We then consider a model that can explicitly express the progress transition between time 𝑡𝑠 + 𝑘𝜏 to 𝑡𝑠 + (𝑘 + 1)𝜏
of aircraft 𝑓 :
𝑝𝑘+1
𝑓
= 𝑦(𝑝𝑘𝑓 , 𝐱𝑓𝑘 ), 𝑓 ∈ 𝑒 , (20)
where 𝑦(.) represents the real operation process during the period of 𝜏 units of time, and 𝐱𝑓𝑘 is the control input vector. Since
the lengths of the links are different, it is possible that 𝑝𝑘+1
𝑓
equals to 𝑝𝑘𝑓 .
• Approach: aircraft that have passed the TMA entry node and operated in the TMA are put in the approach flight set 𝑡 . We
consider the operation in the TMA as the last state. Since the decision variables related to the operation in the TMA are
determined before the aircraft gets in the TMA, from a macroscopic point of view, the route from the TMA entry node to the
runway can be seen as one last link. Once an aircraft gets out of the runway, its progress parameter is set to −1, which means
it is out of the network and is no longer concerned.
In order to clearly explain the dynamic control frame, a simplified example is given with one arrival route and several
consecutively operated aircraft. Fig. 11 shows the situations of two consecutive rolling horizons. We suppose that aircraft 𝑓 , 𝑔
and 𝑚 enter the E-AMAN horizon from the same boundary entry node. As shown in Fig. 11(a), at time 𝑡𝑠 + 𝑘𝜏, aircraft 𝑓 is already in
the system and aircraft 𝑔 will enter the boundary within 𝜏 units of time, thereby, 𝑔 ∈ 𝑝 . Since 𝑓 is on the first link of its route and
𝑔 has not entered the system yet, we have 𝑝𝑘𝑓 = 1 and 𝑝𝑘𝑔 = 0. Based on the current positions of aircraft and the decision inputs, the
predicted trajectories of all aircraft in the system that start from their current positions until landing are obtained. The sub-problem
associated with the current update time is established. By minimizing the objective, control decisions such as alternative route,
enroute speed for the next link, terminal entry speed and runway are generated for all aircraft. For 𝑓 and 𝑔, speeds on their next
links (with associated sequence order of 𝑝𝑘+1 𝑓
and 𝑝𝑘+1
𝑔 ) are the control decisions obtained by optimizing the sub-problem of 𝑘th
𝑢
optimization and will be delivered to pilots before 𝑓 arrives at link 𝑙𝑝𝑘+1 and 𝑔 arrives at 𝑙𝑝𝑘+1 . The speeds are denoted as 𝑣𝑙 𝑓 and
𝑓 𝑔 𝑝𝑘+1
𝑓
𝑢
𝑣𝑙 𝑔 , in this case 𝑢𝑓 = 𝑢𝑔 .
𝑝𝑘+1
𝑔
After 𝜏 units of time passed, the locations of aircraft 𝑓 and 𝑔 are updated, as shown in Fig. 11(b), from which we know that
𝑝𝑘+1
𝑓
= 2 and 𝑝𝑘+1
𝑔 = 1. The newly scheduled aircraft 𝑚 ∈ 𝑝 enters the boundary at the required entry time, based on which its
trajectory prediction is made. The sub-problem associated with the current time is then established based on the predicted trajectory
information of all aircraft in the network. The decisions that interactively minimize safety issues will be assigned to all aircraft.
Taking 𝑓 as an example, its new available control decisions (speed or if possible the route decision) at 𝑝𝑘+2
𝑓
th link along the route
𝑢𝑓 is specified and will be delivered to the pilots.
3.7.2. Dynamic weighted approach

During the flight operation, aircraft in the system could be located anywhere in the route network. The aircraft that are far
from the TMA have potential ability to impact further situations by changing their decisions, whereas aircraft that are sufficiently
close to the TMA can only affect the future situation inside the TMA. In addition, the decisions assigned to aircraft should always
give priority to solving short-term problems rather than those in the long future. Therefore, unified weights for the metrics of the
enroute segment and the TMA are not realistic and not effective for a dynamic problem. We then proposed a dynamic weighted
approach that weights the metrics associated with the enroute segment and the metrics associated with the TMA for each aircraft in
accordance with its location. For simplicity, these two weights are referred to as enroute weight and TMA weight. For an aircraft,
the enroute weight and TMA weight are determined based on the distance that needs to be proceed from the starting point of next
link. The sum of the enroute weight and TMA weight of an aircraft is 1. The weight assigned to the metrics related to enroute phase
at the 𝑘th update time for aircraft 𝑓 is denoted as 𝛼𝑓𝑘 (𝐱𝑓𝑘 ), and it is computed as follows:
⎧ 𝑑𝑓𝑒 (𝐱𝑓𝑘 )
⎪ 𝑘 , if 𝑑𝑓𝑒 (𝐱𝑓𝑘 ) > 0,
𝛼𝑓𝑘 (𝐱𝑓𝑘 ) = ⎨ 𝐷𝑓 (𝐱𝑓 ) (21)
⎪0, otherwise
⎩
13
Fig. 11. Dynamic simulation process. At each update time, aircraft trajectories are simulated from their starting location until landing.
Fig. 12. The valid distances for dynamic weight assignment approach. For the enroute phase, the distance from the next waypoint until TMA entry point is
considered. The length used for the TMA weight starts from the TMA entry point until the runway threshold.
where 𝐷𝑓 (𝐱𝑓𝑘 ) and 𝑑𝑓𝑒 (𝐱𝑓𝑘 ) are the total distance and the enroute distance that need to be proceeded by aircraft 𝑓 . In order to be
consistent with decision making process, for an aircraft that is currently at the enroute phase, the valid distance starts from the next
waypoint that the aircraft will pass until the runway threshold. Fig. 12 illustrates the distance for computing the weights allocation
for aircraft 𝑓 , in which 𝑑𝑓𝑡𝑚𝑎 is the route length in the TMA based on its originally assigned runway. 𝑑𝑓𝑒 (𝐱𝑓𝑘 ) is the distance that needs
to proceed in the enroute phase under the already issued control decision 𝐱𝑓𝑘 . The total distance is the sum of the enroute distance
𝑑𝑓𝑒 (𝐱𝑓𝑘 ) and the distance in the TMA, which is 𝑑𝑓𝑡𝑚𝑎 . In this figure, aircraft 𝑓 is at the second link on the route. The enroute distance
account for the weight assignment is the total distance of the third and the fourth link.
Considering that the metrics evaluations usually involve multiple aircraft, the weights of a single aircraft should be transferred
to the weight for a metric and incorporated into the objective function. Notations related to this process are given as follows:
• Weight of congestion evaluation in enroute segment

We denote 𝐴𝑘𝑙 (𝑡) the weight for the congestion evaluation of link 𝑙 at time interval [𝑡, 𝑡 + 𝐿]. This weight is simply an average
of enroute weights of aircraft involved in a specific congestion evaluation. The value of the weight is given as:
∑𝑠=𝑞−1 ∑ 𝑡+𝑠𝛥 𝑘 𝑘 𝑘
𝑠=0 𝑓 ∈𝑝 ∪𝑒 𝜔
𝑓 ,𝑙
(𝐱𝑓 )𝛼𝑓 (𝐱𝑓 )
𝑙∈𝑢𝑓
𝐴𝑘𝑙 (𝑡) = 𝑙 ∈ 𝑐 , (22)
𝐸𝑙 (𝑡)
∑ ∑
We remind here that 𝐸𝑙 (𝑡) = 𝑠=(𝑞−1) 𝑠=0
𝑡+𝑠𝛥
𝑓 ∈𝑝 ∪𝑒 𝜔𝑓 ,𝑙 is the number of aircraft that pass the entry node of link 𝑙 during time
period [𝑡, 𝑡 + 𝐿]. 𝛼𝑓 (𝐱𝑓 ) is the enroute weight of aircraft 𝑓 . The decision vector (𝐱𝑓𝑘 ) can be omitted when there is no ambiguity.
𝑘 𝑘
14
• Weight of conflict resolution workload in enroute segment

The weight of the conflict resolution workload is computed by averaging the enroute weights of aircraft involved in the specific
evaluation. Considering an evaluation on node 𝑛 at a specific time interval [𝑡, 𝑡 + 𝐿], the aircraft getting out of its parent links
are involved. Mention that if node 𝑛 has more than two parent links, then the sum of the combination of each two parent links
is concerned for both the metrics evaluation and the weight assignment. For this case, the weight is then given as:
∑𝑠=𝑞−1 ∑ ∑
( 𝑓 ∈𝑝 ∪𝑒 𝜔𝑡+𝑠𝛥 𝛼 𝑘 + 𝑔∈𝑝 ∪𝑒 𝜔𝑡+𝑠𝛥 𝛼𝑘 )
𝑔,𝑙𝑗 𝑔
∑ 𝑠=0
𝑙𝑖 ∈𝑢𝑓
𝑓 ,𝑙𝑖 𝑓
𝑙𝑗 ∈𝑢𝑔
𝑘
𝐴𝑛 (𝑡) = 𝑛 ∈ 𝑐 , (23)
𝑙𝑖 ,𝑙𝑗 ∈𝑛 𝐸𝑙𝑂𝑢𝑡 (𝑡) + 𝐸𝑙𝑂𝑢𝑡 (𝑡)
𝑖 𝑗
𝑙𝑖 ≠‴𝑗
where 𝜔𝑡+𝑠𝛥
𝑓 ,𝑙𝑖
(resp. 𝜔𝑡+𝑠𝛥
𝑔,𝑙𝑗
) represents the event that aircraft 𝑓 gets out of the link 𝑙𝑖 at time slot 𝑡 + 𝑠𝛥. Correspondingly, we
∑𝑠=𝑞−1 ∑ 𝑡+𝑠𝛥 𝑘
have 𝐸𝑙𝑂𝑢𝑡 (𝑡) = 𝑠=0 𝑓 ∈𝑝 ∪𝑒 𝜔
𝑓 ,𝑙
𝛼𝑓 (resp. 𝐸𝑙𝑂𝑢𝑡 (𝑡)) represents the number of aircraft that get out from link 𝑙𝑖 (resp. link 𝑙𝑗 )
𝑖 𝑙𝑖 ∈𝑢𝑓 𝑖 𝑗
at time interval [𝑡, 𝑡 + 𝐿].

• Weight of conflicts in the TMA
In the TMA, conflicts between two consecutively operated aircraft 𝑓 and 𝑔 passing the same waypoint or runway can be
modeled by the following way. For each conflict, two aircraft are involved, and the detected weight of the conflict is the
average weight of the TMA weights associated with these two aircraft.
2 − 𝛼𝑓𝑘 − 𝛼𝑔𝑘
𝑘𝑓 ,𝑔 = , 𝑢𝑡𝑓 ∩ 𝑢𝑡𝑔 ≠ ∅ (24)
2
We remind that 𝛼𝑓𝑘
(resp. 𝛼𝑔𝑘 ) is the enroute weight of aircraft 𝑓 (resp. 𝑔). Correspondingly, 1 − 𝛼𝑓𝑘 (resp. 1 − 𝛼𝑔𝑘 ) is the TMA
weight of aircraft 𝑓 (resp. 𝑔).
Fig. 13 provides a diagram of the dynamic optimization process. By applying the rolling horizon approach, the algorithm
periodically executes several procedures including current time and information updating, aircraft trajectory simulations (prediction)
to establish sub-problems. The control decisions can be obtained through optimizing the dynamically weighted objective function.
3.8. Objective function of the dynamic model
The objective that we want to minimize for the 𝑘th sub-problem is given as:
𝑡𝑟
∑ ⎛∑ ∑ ⎞
𝐺(𝐱𝑘 ) = ⎜ 𝐴𝑘𝑙 (𝑡)𝜙𝑘𝑙 (𝑡) + 𝐴𝑘𝑛 (𝑡)𝜙𝑘𝑛 (𝑡)⎟
⎜
𝑡=𝑡𝑠 +(𝑘−1)𝜏 ⎝𝑙∈𝑐
⎟
𝑛∈𝑐 ⎠
(25)
∑ ⎛ ∑ ∑ ∑ ⎞
+ 𝐴𝑘𝑓 ,𝑔 ⎜ 𝜙nc + 𝜙lc + 𝜙rc ⎟
⎜ 𝑓 ,𝑔,𝜈 𝑓 ,𝑔,𝜇 𝑓 ,𝑔 ⎟
𝑓 ,𝑔∈
𝑔>𝑓
⎝𝜈∈𝑢𝑓 ∩𝑢𝑔 ∩∩{𝑡 ∪𝑒 } 𝜇∈𝑢𝑓 ∩𝑢𝑔 ∩𝑡 𝑟𝑓 =𝑟𝑔 ⎠
where the first term represents the weighted metrics associated with the enroute phase, and the second term is the weighted number
of conflicts in the TMA. The objective is evaluated based on the information of each sub-problem. As predictions of aircraft vary in
time and distance, decisions from the former planning (previous time windows) are included in the objective as well.
4. Solution algorithms
4.1. Simulated annealing (SA)
Simulated annealing is a meta-heuristic algorithm. It is used to find an approximate global optimum other than a precise local
one in a fixed amount of time. This approach is interpreted as a slow decrease in the probability of accepting worse solutions as
the solution space is explored. The probability is related to temperature, which is a continuously decreasing parameter. At each
time step, the algorithm selects a solution close to the current one noted as its neighborhood solution, the quality of this solution is
measured, and the decision is made to move to it or to stay with the current solution. The neighborhood solution can be accepted
when the new solution is better than the current one. Moreover, when the neighborhood solution is worse, it can be accepted with
a probability that is related to the current temperature. With this procedure repeating, the approximate global optimum solution
can be achieved (Aarts and Korst, 1988). Typically this step is repeated until the system reaches a state that is good enough for the
application, or until a given computation budget has been exhausted. Several important factors in SA that need to be specified for
the current problem are given:
• Cooling process and the number of transitions: we applied a geometrical law 𝑇𝑖 = 𝛽𝑇𝑖−1 to execute the temperature decreasing.
For each temperature, the number of transitions 𝑀 is fixed. In a transition, three steps are involved which are neighborhood
generation, neighborhood solution evaluation, and solution acceptance or rejection. Both 𝛽 and 𝑀 are user-defined parameters
considering the size and the features of the problem.
15
Fig. 13. Dynamic optimization framework based on the rolling horizon approach. Based on this approach, relevant information is updated regularly to ensure
accurate trajectory predictions, then the sub-problems are established and optimized to generate suitable control decisions for aircraft considering the long-term
situation and the overall safety. The output of the previous sub-problem is part of the input of the current problem.
• Neighborhood generation: this procedure plays a critical role in finding the potential better solutions. The neighborhood
generation can be done randomly or with some specific designs according to the problem.
• Solution acceptance: the acceptance of the neighborhood solution is based on the Metropolis algorithm (Metropolis et al.,
1953). We suppose that the objective values of the current solution 𝐱 and neighborhood solution 𝐱′ are represented as 𝐺(𝐱)
and 𝐺(𝐱′ ). If the difference 𝐺(𝐱) − 𝐺(𝐱′ ) is positive, the neighborhood solution 𝐱′ becomes the new current state. Otherwise,
if the difference is less than or equal to 0, then the transition from solution 𝐱 to solution 𝐱′ is based on a random probability
𝑃 𝑟 which is expressed by:
⎧
⎪1, if 𝐺(𝐱) < 𝐺(𝐱′ )
𝑃 𝑟 = ⎨ 𝐺(𝐱)−𝐺(𝐱′ ) (26)
⎪𝑒 𝑇 , otherwise.
⎩
• Stopping criterion: the algorithm terminates when the temperature reaches the lowest acceptable temperature, which is set
to 10−4 𝑇𝑜 , where 𝑇𝑜 is the initial temperature for the cooling process. Moreover, we stop the algorithm if a solution with the
objective function equals to 0 is found.
Algorithm 1 displays the pseudo-code of SA, in which the mentioned procedures are displayed, an approximate global minimum
will be returned.
16
Algorithm 1 Simulated annealing algorithm.

1: procedure SAOptimization(𝐺(x), x,  , 𝑇0 )
2: 𝑇𝑐 ← 𝑇0 ;
3: while 𝑇𝑐 > 0.0001 ⋅ 𝑇0 , 𝐺(x) is not optimum solution do
4: for 𝑖 = 1 to 𝑀 do
5: x′ ← NeighborhoodFunction(x);
6: Calculate the new objective 𝐺(x′ );
7: if 𝐺(x) > 𝐺(x′ ) then
8: 𝐺(x) ← 𝐺(x′ );
9: 𝜉 ← random(0, 1);
𝑘 ′ ))
10: else if 𝜉 < exp( 𝐺(x )−𝐺(x
𝑇
) then
𝑐
11: 𝐺(x) ← 𝐺(x′ );
12: end if
13: end for
14: 𝑇𝑐 = 𝑇𝑐 ⋅ 𝛽;
15: end while
16: return 𝐺(x)
17: end procedure
Algorithm 2 Neighborhood selection design for 𝑘th sub-problem. For a chosen flight, the decision selected to be changed is biased
by its weight. Aircraft positions need to be specified to determine whether the alternative route can be chosen.
1: procedure NeighborhoodFunction(x𝑘 )
2: Generate random number, 𝜉 ∈ (0 − 1);
3: Randomly choose one aircraft 𝑓 in the flight set corresponding to x𝑘 ;
4: if 𝜉 <= 𝛼𝑓𝑘 (𝐱𝑓𝑘 ) then
5: if 𝑝𝑘𝑓 < ℎ𝑓 then
6: Randomly choose a decision variable from 𝑢𝑓 and 𝑣𝑙 , change its value in accordance with the possible range.
𝑝𝑘+1
𝑓
7: else
8: Randomly choose one value for 𝑣𝑙 with in the defined range.
𝑝𝑘+1
𝑓
9: end if
10: else
11: Randomly choose either 𝑣𝑓 or 𝑟𝑓 , change its value in accordance to the possible range.
12: end if
13: end procedure
4.2. Adaptation of SA to the dynamic optimization problem
Considering the complexity of the E-AMAN system, SA is adapted to solve this problem under the proposed dynamic control
environment. It is mentioned that the dynamic feature is captured through periodical information update, based on which sub-
problems are established accordingly. In this process, the neighborhood function need to be specified in accordance with the
corresponding sub-problem. Neighborhood selection refers to finding suitable control decisions for aircraft operation. It is known that
the selection of an alternative route is also an option to adjust enroute trajectory. One limitation is that only when an aircraft does
not pass the node that separates the main route and the alternative route, the decision of changing route is available. Consequently,
in each sub-problem, the neighborhood selection process for an aircraft needs to first specify whether this aircraft can choose the
alternative route at its current position or not. Fig. 14 provides an example to explain the condition of alternative route selection. In
this figure, a route 𝑢 and its alternative route 𝑢′ are displayed. The node that separates the alternative route and the nominal route
is marked in red. Aircraft 𝑔 is now at a position before the red node and it can be directed to any of the available routes. Instead,
aircraft 𝑓 has passed the red node, therefore, it has to follow the assigned route.
Apart from the decision of alternative route, other decision variables also need to be considered. A neighborhood solution is
generated by randomly choosing one aircraft and changing one of its decision variables within the defined range. For each aircraft,
the values of its enroute weights and TMA weights bias the chosen of decision variables. For example, if an aircraft has a high
enroute weight, which means it is far away from the TMA, changing the decisions such as enroute speed and assigning alternative
route are likely to be more reasonable. Whereas, for an aircraft near TMA, a higher probability is allocated to change the decision
variables such as TMA entry speed and runway. Algorithm 2 describes the neighborhood solution selection method for a sub-problem
based on a random selecting strategy. We remind that ℎ𝑓 denotes the sequence order of a node, at which the alternative route is
diverted from the main route. If there is no alternative route for this aircraft, then ℎ𝑓 = −1. The pseudo-code of SA for solving
17
Fig. 14. Example of alternative route selection condition. Aircraft located before the node that separates the alternative route and the nominal route can be
assigned to either route. Aircraft that have passed that node need to follow the assigned route, and the decision of route is fixed.
Algorithm 3 Dynamic optimization framework based on the rolling horizon approach. Updated information is used to establish the
sub-problems. By solving the sub-problem, control decisions for aircraft in the system are obtained.
Require: get 𝑡𝑠 , 𝑡𝑒 ;
1: 𝑘 ⟵ 1;
2: while 𝑡𝑠 + (𝑘 − 1)𝜏 <= 𝑡𝑒 do
3: 𝑡𝑐 ⟵ 𝑡𝑠 + (𝑘 − 1)𝜏;
4: Get pre-processing flight set 𝑝 based on 𝑡𝑐 and assign 𝑝𝑘𝑓 = 0, 𝑓 ∈ 𝑝 ;
5: for 𝑓 ∈ 𝑝 ∪ 𝑒 ∪ 𝑡 do
6: Update aircraft progress parameter based on decision inputs, 𝑝𝑘𝑓 = 𝑦(𝑝𝑘−1
𝑓
, 𝐱𝑓𝑘−1 );
7: Compute the weight 𝛼𝑓𝑘 ;
8: Predict the trajectories of 𝑓 based on the decisions 𝐱𝑓𝑘−1 obtained from the previous sub-problem .
9: end for
10: for 𝑙 ∈ 𝑐 ; 𝑛 ∈ 𝑐 do
11: Compute the weights for the enroute metrics 𝐴𝑘𝑙 (𝑡) and 𝐴𝑘𝑛 (𝑡) at each 𝑡 ∈ (𝑡𝑐 + 𝑖𝛥|𝑖 ∈ [0, 𝑡𝑟 ∕𝛥])
12: end for
13: Compute the objective function 𝐺(𝐱𝑘 );
14: Execute the optimization procedure: 𝑆𝑢𝑏𝑃 𝑟𝑜𝑏𝑙𝑒𝑚𝑂𝑝𝑡𝑖𝑚𝑖𝑧𝑎𝑡𝑖𝑜𝑛(𝐺(𝐱𝑘 ), 𝐱𝑘−1 );
15: for 𝑓 ∈ 𝑝 ∪ 𝑒 do
𝑢 𝑢 𝑢 𝑢
16: Obtain new decisions: 𝐱𝑓𝑘 = (𝑢𝑓 , 𝐯𝑓𝑓 , 𝑣𝑡𝑓 , 𝑟𝑓 ), 𝐯𝑓𝑓 = (𝑣𝑙 𝑓 , ..., 𝑣𝑙 𝑓 );
1 𝑝𝑘+1
𝑓
17: end for
18: Obtain new decision inputs 𝐱𝑘 ;
19: 𝑘 ⟵ 𝑘 + 1;
20: end while
the dynamic optimization problem is given in Algorithm 3. The parameters are mostly set through empirical analysis to ensure a
suitable adaptation of the algorithm to our problem, these values are given in Table 3.
5. Results
5.1. Case study
The algorithm has been tested on the flights arriving at the Paris CDG airport on 10th February 2017. Fig. 15 displays the number
of arrival aircraft in the system with respect to the time based on the real traffic data, from which we select the period from 5.AM
to 1.PM to investigate the performance of our proposed model. The chosen period is indicated with a gray background. From Fig. 4,
we know that arrival traffic comes from various directions and finally converges to four TMA entry points. The enroute network
information associated with the four points is given in Table 4, including the number of boundary entry nodes, number of routes,
number of waypoints and number of links that are abstracted to establish the network.
In addition, the information regarding the arrival flights in the chosen period associated with the four TMA entry points is given
in Table 5. All flights involved in this period are heavy and medium with an overall mix ratio of heavy and medium equals to 0.27.
Considering the distribution of the haul type of flights, short-haul flights account for the majority of the air traffic that converges
18
Fig. 15. The number of aircraft in the network for arriving at the Paris CDG airport on 10th February 2017 are shown with respect to time. The time period
from 5.AM to 1.PM is marked by gray background with the highest traffic density is chosen for case study.
Table 3
Empirically-set parameter values of the simulated annealing algorithm for dynamic optimization.
Parameters Values
Geometrical temperature reduction coefficient 0.996
Number of transitions of each temperature step 50
Final temperature 10−4 𝑇0
Table 4
The structure information of the enroute arrival network of the Paris CDG airport within 500 NM. The number of boundary
entry nodes, routes, links and waypoints is given with respect to four TMA entry directions represented by four TMA entry nodes
MOPAR, LORNI, OKIPA and BANOX.
# of boundary entry nodes # of routes # of waypoints # of links
MOPAR 38 43 78 83
LORNI 43 56 103 116
OKIPA 41 55 96 110
BANOX 14 25 42 52
Total 136 179 319 361
Table 5
Arrival traffic information associated with the four TMA entry points.
# of flights Medium Heavy Long haul Short haul
MOPAR 41 40 42 39
LORNI 75 12 44 43
OKIPA 84 7 34 57
BANOX 43 7 13 37
Total 243 66 133 176
to the TMA entry point BANOX. This is because BANOX is located in the north-west of the airport, and there are lots of domestic
flights coming from this direction. As for the routes in the TMA, the west-flow runway configuration (26L, 27R | 26R, 27L) is applied
in this problem and the network abstracted for the TMA is the same as in the previous chapter and it is integrated in the network
shown in Fig. 4.
5.2. Results analysis
In our problem, the results are analyzed from two points of view. First, we investigate the metrics involved in the objective
for each sub-problem during the optimization process. Second, the numerical analysis of time transfer at the TMA entry point is
displayed. In our study, the enroute control adjustment is essentially reflected in the arrival time at the TMA, which is bounded
due to the possible ranges of decisions. The achieved time margins at the TMA entry point can be interpreted as the time transfer
between the TMA and the enroute phase for each aircraft, based on which, the safety issues can be resolved. Therefore, the time
transfer also represents the ability of this approach in achieving the desired goals.
19
Fig. 16. The variations of the objective function, and non-weighted metrics of link conflicts, node conflicts and runway conflicts for all sub-problems. Each
sub-problem corresponds to a continuous reduction. Sequence of the sub-problem is marked in the 𝑥-axis corresponding to its optimization process.
Table 6
The final results of some sub-problems whose objective values did not achieve 0. The related non-weighted metrics are given.
In each sub-problem, the conflicts that occurred within the prediction range are all counted.
𝑘th sub-problem Congestion Workload Link Conflict Node Conflict Runway conflict Total
k = 6 0 0 1 4 4 9
k = 7 0 0 0 8 4 12
k = 17 0 0 0 0 1 1
Table 7
For each sub-problem, the safety issues that occurred within 𝜏 units of time from the associated update time are counted. Recent
safety issues that are not eliminated are given in the form of non-weighted metrics with the associated sub-problems are indicated.
𝑘th information update Congestion Workload Link conflict Node conflict Runway conflict Total
k = 6 0 0 1 0 0 1
k = 7 0 0 0 2 0 2
k = 17 0 0 0 0 1 1
5.2.1. Objective function minimization performance

The safety issues related to the enroute phase and the TMA are targeted by including them in the objective function. One possible
way to evaluate the performance of the algorithm is to investigate the variation of metric values during optimization. Fig. 16 shows
the evolution of the non-weighted metrics including the total value of all metrics and the three types of conflicts in the TMA.
The congestion and the conflict resolution workload are not included in this figure, since compared to the value of conflicts, the
congestion and conflict resolution workload are small. This is because normally, a route can be provided to aircraft with different
destinations, while in this study, we only focus on the air traffic destined to one airport. In Fig. 16, the optimization processes of all
sub-problems are shown, and we can observe that the metrics can be effectively minimized even with high initial values. However,
for some sub-problems, the final objective values do not achieve 0, which means potential safety issues still exist. The associated
results are given in Table 6, where the TMA conflicts in 6th, 7th and 17th sub-problem are not eliminated with conflict values equal
to 9, 12, 1 respectively.
It is worth mentioning that, in each sub-problem, aircraft in every location, no matter whether they just enter the network or
they are at the TMA, perform trajectory predictions, and all predicted information is used for objective function evaluation. For some
aircraft, the predicted range can be up to several hours but the decisions for later operations may be updated due to new sub-problem
optimization. Considering that sub-problem optimization provides decisions for flight operations in the next sub-problem, real world
operational situation is actually the trajectory prediction or simulation of the first 𝜏 units of time in each sub-problem. Considering
this fact, the enroute congestion, conflict resolution workload, and conflicts in the TMA that arise in the first 𝜏 units of time of each
sub-problems should be analyzed from the perspective of the real situation. Fig. 17 illustrates the variation of the non-weighted
metrics derive from trajectory simulations of 𝜏 units of time from each update time, and the corresponding results with metrics do
not equal to 0 are given in Table 7. The difference between these two results indicates that most of the predicted conflicts arising
after 𝜏 units time are solved in the latter sub-problems. For the conflicts shown in Table 7, specific measures can be applied by
controllers at the tactical level, such as route stretching, holding, etc. (to add adjustment margins).
5.2.2. Computation time

The computation time of each sub-problem varies in accordance with the problem size. From the aspect of problem solving, the
number of aircraft involved in the sub-problem usually plays an important role in affecting the size of the problem. Therefore, in
20
Fig. 17. The variation of the total objective value and the metrics associated with the conflicts in the TMA of all sub-problems within 𝜏 units of time from
the update time. The metrics are reduced nearly to 0 for each sub-problem. The final values of the metrics for the sub-problems show the real safety issues
appearing in the system.
Fig. 18. The computation time of each sub-problem and the number of aircraft involves in each sub-problem.
Fig. 18, the information of sub-problems in terms of the number of aircraft and the computation times are given. The number of
aircraft involved in each sub-problem can correspond to the 𝑦-axis on the left, while the computation times are indicated on the
right. From this figure, we can see that in most cases, the number of aircraft involved in the problem is positively correlated with
the computation time. Since we consider the time period with the highest traffic density in the Paris CDG airport, we can assume
that the computation time for each sub-problem is no more than the maximum computation time among all sub-problems, which is
174.2 s. It is known that the Paris CDG airport is a hub airport with a high traffic density, the case study is not the most extreme
case in the world, but it shows the ability of our algorithm to solve a complex problem in an acceptable time. In our problem, the
rolling horizon for periodic updates is set to 𝜏 = 20 mins, which is much bigger than the maximum sub-problem solving time. This
implies the potential usability of this algorithm to be adapted to a practical situation. One exception regarding the problem size and
the computation time is shown in the 15th sub-problem, where the solution is found very fast, this may be due to the randomness
of the solution approach, since the algorithm randomly explores the state space.
5.2.3. Trajectory control

The trajectory control is analyzed in terms of the variation of the decision variables, since the chosen decision variables would
finally determine the trajectory of flights. Fig. 19 displays the enroute speed deviations from the nominal enroute speed of each
aircraft on its passed links. The speed deviations are shown in 𝑥-axis by percentages, the height of the column corresponds to the
frequency that a specific deviation occurring during the operations of all flights. The enroute speed deviation distributions of long-
haul flights and short-haul flights are shown separately in Figs. 19(a) and 19(b). Aircraft that conduct speed reduction in the enroute
phase absorb the delay that is supposed to be dealt with by holding at the TMA. When an aircraft speed increase is conducted, less
transition time on the network resources is used so as to rapidly spare the limited resources for later flights. In these two figures, no
clear features on the distributions of both long-haul flights and short-haul flights can be detected. While, when we consider aircraft
maneuvering, we can find that speed reduction comes more often than speed increase, especially for long-haul flights. Moreover,
the speed deviations are not crowded at high values, which implies the benefit of early planning of flights while predicting their
21
Fig. 19. Frequency distribution of enroute speed deviation of each aircraft on each link taking the enroute optimal speed as a reference.
Table 8
Runway assignment comparison before and after optimization.
Runway Number of landings Number of landings
before optimization after optimization
26L 146 (47.2%) 155 (50.1%)
27R 163 (52.7%) 154 (49.9%)
whole trajectories. It is worth mentioning that wind plays an important role in influencing the speed of aircraft routes, and the
instructions provided by the controllers usually depend on an accurate forecast of the weather conditions. This algorithm would
also rely on weather information to make adjustments for the decisions chosen for the aircraft. As for the route change, among 309
arrival flights, 142 of them only have one route to be followed, therefore are not considered in this analysis. Among the flights that
have alternative routes, 76 flights are required to follow their alternative route, and 81 flights stay on their initial arrival route.
Both the enroute speed and the chosen of alternative route affect the aircraft arrival time at the TMA. In our case, most of the flights
follow their initial routes, which means that the system can rely more on earlier speed regulations to adjust the flight trajectories
so as to achieve safety in the enroute phase and provide an efficient environment for flights that are operated in the TMA.
Further, the runway assignment is shown in Table 8, in which the numbers of flights that are assigned to the two runways before
and after optimization are given. It is shown that after optimization, flights are averagely assigned to the two runways. Additionally,
in Fig. 20, the speed deviations at four TMA entry points are computed taking real TMA entry speeds from input data as references.
The results are also displayed separately for long-haul flights (Fig. 20(a)) and short-haul flights (Fig. 20(b)). Our algorithm chooses
suitable values for flights to achieve smooth and safe approaches. Since holding time is not considered in this problem. We can
conclude that early control under overall situation awareness considering the future alleviates the control burden by only adjusting
the aircraft speed at a specific range in the TMA.
One thing that needs to be emphasized is that the results presented here are the parameters obtained after all flights are
optimized. In the dynamic optimization model, sub-optimization problem solving produces speed on the next link, alternative route
(if available), TMA entry speed, and runway assignment for each flight considering the overall situation. However, some decisions
may not be required immediately and frequent change of decisions creates extra workload and confusion for controllers. Therefore,
we assume that only the decisions that need to be executed within 𝜏 units of time are provided to the controllers and pilots.
5.2.4. Time transfer analysis

In some cases, CTA is assigned when applying the concept of E-AMAN to realize control in terms of time constraints for aircraft
arriving at a waypoint. Therefore, it is necessary to conduct an analysis of the time margin achieved in our study to figure out
the time transfer capability of enroute speed changing and alternative route selection in the dynamic optimization model. In this
study, the time margins need to be achieved are not pre-defined or pre-assigned for each aircraft before operations, the results of
time transfer (either delay or flight time reduction) in the enroute phase are driven by the optimization model to ensure safe and
smooth operation in the network. Since the TMA entry points are assumed to be the boundary points of the enroute phase and the
TMA, time transfer due to early and coordinated enroute adjustment can be obtained by comparing the aircraft arrival times at TMA
entry points derived from the dynamic optimization model and a proper baseline model. Without applying the E-AMAN concept,
enroute air traffic is managed separately without coordinating with the operational situation in the TMA, and flights arriving at
the TMA can be controlled with stacking and holding to wait for landing slots. Therefore, we perform an optimization focusing on
22
Fig. 20. Frequency distribution of TMA entry speed deviation of arrival flights taking the initial TMA entry speed as a reference.
Fig. 21. The gain or loss of time in the cruising phase by applying early trajectory adjustment based on the results of the dynamic optimization model and the
baseline model.
the enroute phase with an aim of eliminating congestion, resolution workload, and reducing the speed deviation. In the baseline
model, the original time information of flights entering the boundary is used as input. The enroute network is also used. Enroute
decision variables, which are aircraft speeds on the links and the alternative route assignment in the dynamic optimization model are
identically implemented including the constraints related to their ranges. The objective of the baseline model is to minimize the sum
of the congestion, conflict resolution workload, and speed deviation from the nominal speed of each aircraft. The speed deviation
is considered since speed directly affects the flying time. Minimizing speed deviations encourages aircraft to operate near nominal
speed, which leads to a result without unnecessary extreme speed deviation and potentially to a reduction of fuel consumption. For
𝑢 𝑢
an aircraft, the speed deviation is defined as 𝑣𝑂𝑝𝑡
𝑓
− 𝑣𝑙 𝑓 , where 𝑣𝑂𝑝𝑡
𝑓
and 𝑣𝑙 𝑓 represent the nominal enroute speed and the assigned
𝑖 𝑖
control speed on link 𝑙𝑖 of aircraft 𝑓 , respectively.
The transfer time of an aircraft is interpreted as the enroute flying time difference between the dynamic optimization model
and the baseline model. Fig. 21 shows the transfer time of each aircraft, in which the negative values indicate earliness and the
positive values indicate tardiness. Numerical details are given in Table 9, where the results are investigated with respect to the four
approach directions specified by TMA entry points. The early arrival of aircraft at the TMA reveals the resources in the TMA are also
exploited. The maximum time of earliness and tardiness associated with the four TMA entry points are given, where the biggest
values are shown in bold. Results show that the maximum transfer time among all aircraft is 355.8 s for earliness and 335.1 s for
tardiness. Moreover, the average transferred times for the four approach directions are listed as well.
The result for time transfer shows that the times to gain or to lose for each aircraft in the enroute phase under a long-range
collaborative arrival management is less than 6 min compared to the normal procedure of enroute management. An average time
margin of fewer than 2 min is achieved to enable a smooth arrival without imposing holding or stacking over the destination airport
while preserving or enhancing the existing level of safety. In addition, by flying at cruise level and avoiding holding and stacking,
our approach tends to promote efficiency.
23
Table 9
Time difference at TMA entry point between the schedule provided by the dynamic optimization model and the baseline model.
Earliness (absolute value) Tardiness (absolute value)
Nb of aircraft Max (s) Ave (s) Nb of aircraft Max (s) Ave (s)
MOPAR 46 344.9 78.63 35 177.6 70.7
LORNI 49 252.3 102 38 335.06 79.8
OKIPA 51 355.8 92.8 40 189.65 72.1
BANOX 22 217.7 81.2 28 243.23 84.16
6. Discussion
The E-AMAN concept emphasizes a smooth traffic management process by considering not only airspace around airport but
also further upstream airspace. Compared to other studies that are mainly focused on delay absorption by enroute speed reduction,
this work involves operational details in enroute and TMA to conduct collaborative decision making with a dynamic framework.
In the model establishment, we initially assumed that the holding procedure over the airport does not exist, then smooth arrival
in the TMA is forced, which leads to less controller workload and a more economical flying pattern. In dynamic problem solving,
information update is critical for the sub-problem establishment, based on which flight trajectories can be predicted for obtaining
situational awareness. Further, sub-problems solving provides decisions for the next period. The dynamic optimization process is
actually similar to the real operational situation. However, several limits exist, First, information sharing requires in-time responses
from all parties such as ANSPs, airlines, controllers, etc. which is hard to coordinate. Second, in this model, we establish a network
abstraction for aircraft trajectory prediction, but flight inherent disturbances may cause errors and flights may not exactly follow the
routes. Third, from the results, we can find that when flight density is high, some conflicts still exist. Although differences between
conflicts from the sub-problems (Table 6) and the real conflicts (Table 7) indicate that the dynamic weight assignment approach
functions well, the remaining conflicts show the possibility that the solution may not achieve optimal and extra tactical strategy
still needed to deal with unexpected safety issues. Considering all these limitations, some of them can be addressed. For example,
we believe that a proper uncertainty management model can help with the errors in trajectory prediction; and some other solution
algorithm can be implemented to see whether conflict-free solutions can be achieved.
7. Conclusion
In this paper, we investigate a dynamic control optimization problem based on the E-AMAN concept for arrival management
considering the main safety issues in the enroute phase and the TMA. The main features of E-AMAN concept lie in two aspects,
the extended horizon which is 500NM from the destination airport, and trajectory adjustment under good situational awareness of
the overall situation. In line with this concept, we established a dynamic optimization model, aiming at adapting the concept of
E-AMAN to problem formulation and provide potential value for future development.
The dynamic nature is addressed through the techniques of the rolling horizon and the freezing horizon. The rolling horizon
is defined according to a 20-minutes information update cycle, while the frozen horizon refers to a geographical area in which
the decisions of aircraft are fixed. With the rolling horizon approach, the dynamic problem is tackled by periodically updating the
aircraft trajectory information and updating the aircraft set according to whether the aircraft is in the network. The sub-problem
associated with each information update takes the congestion, conflict resolution workload in the enroute phase and the conflicts
in the TMA into the objective function. In order to enable a long-term awareness of the air traffic situation, predictions of aircraft
trajectories are made for the sub-problems. Additionally, considering the fact that the recent safety issues are always critical, we
introduced a dynamic weight assignment approach. In this approach, different weights are assigned for the enroute metrics and the
TMA metrics with a sum value of 1 according to the position of the aircraft with respect to its route. Through this design, aircraft
that are far away from the TMA are preferably controlled for en-route safety problems, while when aircraft approaches the TMA,
more weights tend to be assigned to the metrics of the TMA.
Results of the static sub-problems are analyzed from the aspects of optimization performance, control decision distributions and
time transfer. The evolution of the metrics in the optimization process, especially the conflicts in TMA implies that the proposed
algorithm can effectively reduce safety issues through achievable time transfer. Using the data of the Paris CDG airport at peak
hours, the optimized solution required an average time transfer of around 2 min.
For future work, the model will be improved to be more realistic by increasing the flexibility of decision assignments for aircraft
during the cruising phase and analyzing the attributes of uncertainties on enroute trajectory prediction to add robustness to the
model.
CRediT authorship contribution statement
Ying Huo: Methodology, Software, Formal analysis, Investigation, Writing – original draft, Visualization. Daniel Delahaye:
Conceptualization, Writing – review & editing, Supervision. Mohammed Sbihi: Methodology, Writing – review & editing.
24
Acknowledgments
This research is supported by grants from the Funds of China Scholarship Council (201708070018).
References
Aarts, Emile, Korst, Jan, 1988. Simulated annealing and Boltzmann machines.
Ball, Michael O., Chen, Chien-Yu, Hoffman, Robert, Vossen, Thomas, 2001. Collaborative decision making in air traffic management: Current and future research
directions. In: Bianco, Lucio, Dell’Olmo, Paolo, Odoni, Amedeo R. (Eds.), New Concepts and Methods in Air Traffic Management. In: Transportation Analysis,
Springer Berlin Heidelberg, pp. 17–30. http://dx.doi.org/10.1007/978-3-662-04632-6_2, ISBN: 978-3-642-07491-2 978-3-662-04632-6.
Ball, Michael O., Lulli, Guglielmo, 2004. Ground delay programs: Optimizing over the included flight set based on distance. Air Traffic Control Q. 12 (1), 1–25.
http://dx.doi.org/10.2514/atcq.12.1.1, (ISSN: 1064-3818, 2472-5757).
Bennell, Julia A., Mesgarpour, Mohammad, Potts, Chris N., 2017. Dynamic scheduling of aircraft landings. European J. Oper. Res. 258 (1), 315–327.
CANSO, 2022. Long-range air traffic flow management concept. https://canso.org/publication/long-range-air-traffic-flow-management-concept/ (Accessed:
20-Dec-2022).
Delahaye, Daniel, Puechmorel, Stéphane, 2013. Modeling and Optimization of Air Traffic. Wiley-ISTE, Hoboken, NJ.
Delgado, Luis, Prats, Xavier, 2009. Fuel consumption assessment for speed variation concepts during the cruise phase. In: ATM Economics Conference.
Delgado, Luis, Prats, Xavier, 2012. En route speed reduction concept for absorbing air traffic flow management delays. J. Aircr. 49 (1), 214–224.
Delgado, Luis, Prats, Xavier, Sridhar, Banavar, 2013. Cruise speed reduction for ground delay programs: A case study for San Francisco International Airport
arrivals. Transp. Res. C 36, 83–96.
Dhief, Imen, Lim, Zhi Jun, Goh, Sim Kuan, Pham, Duc-Thinh, Alam, Sameer, Schultz, Michael, 2020. Speed control strategies for E-AMAN using holding
detection-delay prediction model. In: 10th SESAR Innovation Days, SID2020.
Frankovich, Michael Joseph, 2012. Air traffic flow management at airports: A unified optimization approach (Ph.D. thesis). Massachusetts Institute of Technology.
Hao, Yeming, Lovell, David J, Ball, Michael O, Torres, Sergio, Nagle, Gaurav M, 2020. An arrival scheduling model for incorporating collaborative decision-making
concepts into time- based flow management. In: International Conference on Research in Air Transportation, ICRAT2020.
Huang, Haomiao, Tomlin, Claire, 2009. A network-based approach to en-route sector aircraft trajectory planning. In: AIAA Guidance, Navigation, and Control
Conference. p. 6169.
Huo, Ying, Delahaye, Daniel, Sbihi, Mohammed, 2021. A probabilistic model based optimization for aircraft scheduling in terminal area under uncertainty.
Transp. Res. C 132, 103374.
ICAO, 2016. Air traffic management,doc4444. Technical report.
Jones, James C., Lovell, David J., Ball, Michael O., 2013. En route speed control methods for transferring terminal delay. In: 10th USA/Europe Air Traffic
Management Research and Development Seminar (ATM2013), Chicago.
Jones, James C., Lovell, David J., Ball, Michael O., 2018. Stochastic optimization models for transferring delay along flight trajectories to reduce fuel usage.
Transp. Sci. 52 (1), 134–149. http://dx.doi.org/10.1287/trsc.2016.0689, (ISSN: 0041-1655, 1526-5447).
Jun, Lim Zhi, Alam, Sameer, Dhief, Imen, Schultz, Michael, 2022. Towards a greener extended-arrival manager in air traffic control: A heuristic approach for
dynamic speed control using machine-learned delay prediction model. J. Air Transport Manag. (ISSN: 0969-6997) 103, 102250.
Matsuno, Yoshinori, Andreeva-Mori, Adriana, 2020. Analysis of achievable airborne delay and compliance rate by speed control: A case study of international
arrivals at Tokyo International Airport. IEEE Access 8, 1. http://dx.doi.org/10.1109/ACCESS.2020.2994109.
Metropolis, Nicholas, Rosenbluth, Arianna W, Rosenbluth, Marshall N, Teller, Augusta H, Teller, Edward, 1953. Equation of state calculations by fast computing
machines. J. Chem. Phys. 21 (6), 1087–1092.
Murça, Mayara Condé Rocha, Müller, Carlos, 2015. Control-based optimization approach for aircraft scheduling in a terminal area with alternative arrival routes.
Transp. Res. Part E: Logist. Transp. Rev. 73, 96–113.
Nikoleris, Anastasios, Chatterji, Gano, Almog, Noam, Palopo, Kee, 2012. Arrival delay absorption using extended metering with speed control. In: 12th AIAA
Aviation Technology, Integration, and Operations (ATIO) Conference and 14th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference. p. 5640.
Peters, Arjen, 2017. Applicability of extended arrival manager to assist ANSPs in determining an arrival sequence that accounts for en-route delays. Preprint
Submitted To Lecturers Aviation.
Rosenow, Judith, Asadi, Ehsan, Lubig, Daniel, Schultz, Michael, Fricke, Hartmut, 2022. Long range air traffic flow management with flight-specific flight
performance. Future Transp. 2, 310–327. http://dx.doi.org/10.3390/futuretransp2020017.
Schultz, Michael, Lubig, Daniel, Asadi, Ehsan, Rosenow, Judith, Itoh, Eri, Athota, Srinivas, Duong, Vu N., 2021. Implementation of a long-range air traffic flow
management for the Asia-Pacific region. IEEE Access (ISSN: 2169-3536) 9, 124640–124659. http://dx.doi.org/10.1109/ACCESS.2021.3110371.
SESAR, 2015a. Controlled time of arrival (CTA) in medium density and complexity environments. https://www.sesarju.eu/node/2199.
SESAR, 2015b. SESAR extended AMAN presented at world ATM congress. https://www.sesarju.eu/newsroom/all-news/sesar-extended-aman-presented-world-
atm-congress (Accessed: 13-Mars-2015).
SESAR, 2019. SESAR solution catalogue 2019. Technical report.
SESAR, 2021a. COVID-19 impact on the European air traffic network. https://www.eurocontrol.int/covid19.
SESAR, 2021b. Eurocontrol forecast update 2021–2027. https://www.eurocontrol.int/publication/eurocontrol-forecast-update-2021-2027 (Accessed 15-Oct-2021).
Swenson, Harry N, Thipphavong, Jane, Sadovsky, Alex, Chen, Liang, Sullivan, Chris, Martin, Lynne, 2011. Design and evaluation of the terminal area precision
scheduling and spacing system.
25

A Dynamic Control Method For Extended Arrival Management Using Enroute Speed Adjustment and Route Change Strategy

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

A Dynamic Control Method For Extended Arrival Management Using Enroute Speed Adjustment and Route Change Strategy

Uploaded by

Copyright:

Available Formats

Transportation Research Part C 149 (2023) 104064

Contents lists available at ScienceDirect

Transportation Research Part C

A dynamic control method for extended arrival management using

ARTICLE INFO ABSTRACT

3.1. Network abstraction

3.2. Given information of flights

• 𝑐𝑓 : aircraft categories (heavy, medium, light).

3.3. Decision variables and constraints

3.3.1. Decision variables for enroute segment

where 𝛥𝑣𝑓 is a speed adjustment step that is set to be 0.01𝑣𝑜𝑝𝑡𝑓

3.3.2. Decision variables for the TMA

• TMA entry speed 𝑣𝑡𝑓 :

𝑣𝑡𝑓 ∈ 𝑉𝑓𝑡 ∶= {𝑣𝑜𝑝𝑡𝑅

3.4. Metrics evaluation

3.4.1. Traffic flow-based evaluation in the enroute segment

3.4.2. Congestion evaluation

𝜙𝑙 (𝑡) = max[(𝐸𝑙 (𝑡) − 𝑅𝑙 ), 0], 𝑙 ∈ 𝑐 (7)

3.4.3. Resolution workload evaluation

𝜙𝑛 (𝑡) = max[(𝐸𝑛 (𝑡) − 𝑅𝑛 ), 0], 𝑛 ∈ 𝑐 , (9)

3.4.4. Conflict detection in the TMA

𝑡𝑔,𝑟 − 𝑡𝑓 ,𝑟 ≥ 𝜏𝑓𝑟𝑐,𝑔 (13)

• Node conflict indicator

• Link conflict indicator,

• Runway conflict indicator

3.5. Wind effects

3.6. Objective function

3.7. Dynamic arrival management

3.7.1. Dynamic control framework

3.7.2. Dynamic weighted approach

• Weight of congestion evaluation in enroute segment

• Weight of conflict resolution workload in enroute segment

at time interval [𝑡, 𝑡 + 𝐿].

3.8. Objective function of the dynamic model

4.1. Simulated annealing (SA)

Algorithm 1 Simulated annealing algorithm.

4.2. Adaptation of SA to the dynamic optimization problem

5.1. Case study

5.2. Results analysis

5.2.1. Objective function minimization performance

5.2.2. Computation time

5.2.3. Trajectory control

5.2.4. Time transfer analysis

CRediT authorship contribution statement

You might also like