Models for Aircraft Landing Optimization

Mohammad Mesgarpour Chris N. Potts Julia A. Bennell

Abstract—Due to an anticipated increase in air traffic during the reasons for imposing separations. The minimum required
the next decade, air traffic control in busy airports is one of separation governs the minimum permissible distance interval
the main challenges confronting controllers in the near future. between aircraft lined up in sequence on the approach to
Since the runway is often a bottleneck in an airport system, there
is great interest in optimizing usage of the runway. Our study landing on the runway. Generally, the Wake Vortex (WV)
first presents a brief review of the aircraft landing problem. A separation required between consecutive aircraft depends on
model for the problem is then introduced, and possible solution the type of the aircraft, and therefore it is sequence dependent.
approaches are discussed. During peak capacity operations, the WV is often a major
concern. It effectively determines runway throughput and thus
I. I NTRODUCTION
limits airport capacity in the terminal airspace. Significant
Airport capacity (and hence runways and gates) is increas- asymmetries in minimum required separations can offer an
ingly becoming a limiting factor in meeting the rising demand opportunity to reduce airborne delays by shifting aircraft
for more flights. One of the main factors that determines positions in the landing sequence, although finding the best
the runway throughput at airports is the required separation way to achieve this presents a challenge.
between aircraft during take-off and landing. Dependency of There are well-known procedures for making an aircraft
separation on the leading and trailing aircraft and the type wait to land, such as using Vector-for-Space (VFS) or Holding
of aircraft add to the challenge of solving the sequencing Pattern (HP) [3]. Nevertheless, reassigning an aircraft to
and scheduling problem. Due to its complexity, it is hard landing time far from its initial place in FCFS sequence is
to find the optimal solution to the problem in most cases. not always feasible because of the operational constraints in
Thus, it draws significant attention from different scientific practice. The Constrained Position Shifting (CPS) concept
communities with numerous research studies carried out on is introduced by Dear [4] for the ALP to prevent the final
modelling and developing algorithms to increase capacity at positions of aircraft in landing sequence from differing from
an airport. the FCFS order by more than a pre-specified number, called
Since the appearance of a paper of Blumstein [1] on the Maximum Position Shift (MPS). Furthermore, when the
estimating the capacity of an arrivals runway, there have been a MPS is small, it maintains fairness among the aircraft by not
variety of studies on airport runway optimization. Although the deviating too far from the FCFS sequence. Relative Position
literature during the last three decades contains more than sixty Shifting (RPS) is a variant of MPS that takes into account the
publications on aircraft landing optimization, the majority of closeness of the aircraft to the runway when specifying the
the proposed methods have never been implemented. MPS.
The Aircraft Landing Problem (ALP), which is the focus This paper focuses on the techniques and tools of opera-
of our work, is to sequence landing aircraft onto the available tional research and management science (OR/MS) for solving
runways at an airport and to assign each aircraft a landing the ALP. Section II is a brief review of the literature on the use
time, subject to variety of operational constraints. The simple of OR/MS techniques for tackling the ALP. A mixed integer
way of sequencing and scheduling of landing aircraft on a programming (MIP) model is proposed in Section III. Finally,
single runway is using the First-Come-First-Served (FCFS) future research directions are discussed in Section IV.
approach which assigns scheduled landing time to each aircraft
based upon the sequence implied by the earliest time that the II. L ITERATURE R EVIEW
aircraft can land. It has been found that FCFS is rarely the Most previous research on the ALP considers a static (off-
best sequencing order in terms of capacity, average delay or line) environment based on a given set of aircraft operating
even average passenger delay [2]. over a predefined time horizon. However, a more realistic
The prime responsibility of the air traffic controllers is model considers a dynamic (on-line or real-time) environment,
safety of the flights. Standard vertical and horizontal sep- where solutions are revised as aircraft arrive over time and new
arations which keep flights apart provide one of the main information becomes available.
Air Traffic Control (ATC) safety devices. Vortices generated A variety of OR/MS techniques have been used to model
by the aircraft as a consequence of their lift are one of the problem, such as mixed integer programming (MIP)
and queueing theory, while commonly used approaches for rithms for the ALP. For example, in addition to an extensive
machine scheduling and travelling salesman problems (TSP) literature overview on the ALP, Beasley et al. [13] develop
are also useful in the development of solution algorithms. algorithms for both single and multiple runways. The model
Various search techniques have been also applied to solve the is based on an earlier MIP formulation presented in [14].
problem such as Dynamic Programming (DP), Branch-and- Different metaheuristic approaches have been examined for
Bound, Branch-and-Price, Genetic Algorithms/Programming, scheduling landing aircraft. One of the first and the simplest
Ant Colony Optimization, etc. application of a GA in minimizing the earliness/lateness for the
Based on a DP approach for solving the TSP, Psaraftis [5] ALP is investigated by Stevens [15]. Based on his work on the
develops three algorithm for the static case of the ALP to permutation-based approach, Ciesielski and Scerri [16], [17]
examine two alternative objectives, the landing time of the compare two GAs for a real-time dynamic ALP in terms of
last aircraft, and the total passenger delay with respect to the percentage of valid solutions and best fitness by specifying
FCFS discipline. The CPS concept is also incorporated within a 30-second time slot and variable times between landings.
the algorithm. Bianco et al. [6] point out the relationship Cheng et al. [18] design four different GAs for solving the
between the ALP,P a machine scheduling problem (denoted by multiple runway ALP.
1|rj , seq-dep | Cj ) with n jobs and the cumulative TSP with A Population Heuristic (PH) is developed by Beasley et
ready times. A dynamic programming formulation and three al. [19] to improve the utilization of a single runway. The
lower bounds are proposed for the scheduling problem. algorithm aims to minimize the squared deviations from esti-
The scheduling of aircraft landing is formulated as a single mated landing time in the presence of five separation criteria.
machine scheduling problem in Bayen et al. [7]. A DP Later, Hansen [20] examines the efficiency and effectiveness
approach and a linear programming relaxation and rounding of various genetic approaches for the ALP. Regarding the
are used in the main algorithm. Two approximation algorithms objective function, three different formulations are presented
for the sum of arrival times of all aircraft and the arrival time by Capri and Ignaccolo [2] with respect to minimizing the
of the last aircraft have performance bounds of 5 and 3, respec- delays which depend on the aircraft classes, maximizing the
tively. Because different classes of aircraft are not considered, system capacity, and minimizing the sum of landing times.
the required separation between landings is independent of the The Receding Horizon Control (RHC)-based GA that is
aircraft type. introduced by Hu and Chen [21] minimizes the airborne delay,
Recently, Balakrishnan and Chandran [8] present a DP- which is the deviation of actual landing time from estimated
based approach to maximize the runway throughput (equiv- landing time in a dynamic environment. Hu and Paolo [22],
alently, minimizing the landing time of the last aircraft). The [23] experiment with alternative solution representations in a
problem of scheduling landing aircraft in a CPS environment is single and multiple runway ALP. Two different population
considered subject to various operational constraints imposed heuristics, Scatter Search and the Bionomic Algorithm, are
by arrival time windows, minimum aircraft separation require- applied by Pinol and Beasley [24] to the multiple runway ALP.
ments, and precedence relations. The problem is formulated Both linear and non-linear objective functions are considered.
as a modified shortest path problem on a network with Ant Colony Optimization (ACO), which is a constructive
O(n(2k + 1)2k+2 ) arcs, where n is the number of aircraft metaheuristic technique with biological foundation, has been
and k is the maximum position shift. applied to the ALP by Randall [25]. His algorithm aims to
Chandran and Balakrishnan [9] also propose a DP algorithm minimize the difference between an estimated landing time and
to compute the tradeoff curve between the robustness (relia- the actual landing time for each aircraft, subject to a specified
bility of a schedule) and throughput based on their earlier time window and separation criteria.
work [8]. More recently, Lee and Balakrishnan [10] extend Dynamic programming exhibits the best performance
the previous framework [8], [9] and present a DP algorithm among exact methods because of its enumerative nature. How-
for minimizing the total delay costs of an arrival schedule. ever, the ability to control run time in local search methods
Also, the problem of minimizing the fuel costs of the arrival such as GAs makes them serious candidates for use in decision
schedule as the main operating cost for most airline, has been support tools for air traffic controllers.
studied using the proposed algorithm by allowing the earliest
III. P ROBLEM D EFINITION
landing time to be less than the estimated landing times which
is known as Time Advance (TA). Several polynomial-time The majority of research on the ALP considers sequencing
DP algorithms for the ALP based on machine scheduling the aircraft in the Terminal Manoeuvering Area (TMA). How-
concepts are presented in [11], [12]. Six sequencing algorithms ever, sequencing aircraft further away from the airport (such
which include three DPs, two FCFS rules, and a heuristic that as Extended TMA) may produce better results. The problem
represents a potential algorithm for an operational AMAN is divided into three time stages.
(Arrival Manager) system are implemented. Moreover, four • Stage 1 (Sequencing Stage): The first stage starts by
delay sharing strategies include all delay in hold, delay as entering the aircraft into the airport landing planner’s
late as possible, delay as early as possible, and delay evenly radar range about 40 minutes before touchdown.
throughout the route strategies are implemented. • Stage 2 (Modifying Schedule Stage): The second stage
The literature also contains several branch-and-bound algo- starts 11 minutes before landing and takes 8 minutes and
 includes the final approach step. B. Objective Function
• Stage 3 (Freezing Stage): The last stage consists of the Choosing an appropriate objective function for the ALP is
runway controller’s range of operation which is 3 minutes controversial and not all stakeholders (ATC, airport, airlines,
long. and government) agree on the selection process. However,
Sequencing and scheduling of arrival aircraft are performed the following multi-criteria objective function can potentially
in stage 1. As time progresses and new aircraft enter the satisfy the interests of all the parties.
sequencing stage, the sequence and schedule have to be • Minimizing the average delay which includes the lateness
updated, which is done every five minutes. In stage 2, the and earliness.
sequence is not usually changed, with only minor adjustments
n
to the schedules being made. As the aircraft is so close to the X
runway in stage 3, neither the sequence nor the schedule can (Li max{(SLTi − TLTi − Al+
i ), 0}
i=1
be modified.
+ Ei max{(TLTi − SLTi − Al−
i ), 0}). (1)
A. Notation
• Maximizing the runway throughput (or runway utiliza-
Decision Variables
tion). Equivalently, the average of the landing times can
SLTi The scheduled landing time of each aircraft i. be minimized rather than maximizing the number of
Xij Defined to be 1 if aircraft i lands before (not neces- aircraft landing on the runway (throughput or lead time).
sarily immediately) aircraft j, and 0 otherwise. n
Parameters 1X
SLTi . (2)
n The number of aircraft to be scheduled. n i=1
A The set of available aircraft for landing, A = • Minimizing the fuel burn cost (and hence minimizing
{1, . . . , n}, which is updated every 5 minutes. carbon dioxide emissions). Fuel cost is almost 50% of the
The parameters below are defined for each aircraft i, for i = operating cost. The fuel burn depends on different factors
1, . . . , n. such as pilot flying techniques, altitude, air speed, aircraft
TLTi The target (or expected) landing time of aircraft i model, aircraft weight (including passengers and cargo),
based on the assigned time slot which is normally and fuel in the tank. Consequently, the extra fuel burn cost
specified in flight plan. caused by lateness and earliness has to be considered.
Esti The estimated (or predicted) landing time of aircraft n
i calculated by trajectory synchronizer equipment
X
(FB+
i max{(SLTi − TLTi ), 0}
after entering the aircraft into the radar range, and i=1
is normally based on the FCFS sequence. + FB−
− i max{(TLTi − SLTi ), 0}). (3)
Ali The allowed earliness for aircraft i to land.
Al+ The allowed lateness for aircraft i to land. Since the ALP may involve the simultaneous optimization
i
of various correlated dependent objectives that are not nec-
ELTi The earliest possible landing time for aircraft i,
essarily aligned, a trade-off among the objectives is required.
subject to technical and operational restrictions.
Therefore, they need to be optimized in the form of a weighted
LLTi The latest possible landing time for aircraft i which multi-criteria objective function.
is usually determined from fuel limitation, maximum
allowed delay, or meeting a connecting flight. C. Constraints
Ei The earliness penalty cost per unit for aircraft i to A variety of operational constraints can be imposed for the
be advanced more than Al− i . ALP, the most typical of which are the following.
Li The lateness penalty cost per minute for aircraft i to • Runway Usage: Each runway can be used by at most one
be delayed more than Al+ i . aircraft at a time so either aircraft i lands before j or vice
Pij Defined to be 1 if aircraft i must land before (not versa.
necessarily immediately) aircraft j, and 0 otherwise. Xij + Xji = 1 ∀i, j ∈ A, i 6= j. (4)
Sij The minimum time separation between aircraft i and
j, if aircraft i lands before aircraft j. • Wake Vortex (WV) Separation: Aircraft have to observe
a separation distance to avoid turbulence caused by
TSi The time shifting of aircraft i, which is the maximum
preceding aircraft.
time deviation of this aircraft from/to Esti in the
landing sequence.
SLTi + Sij ≤ SLTj + M (1 − Xij )
FB+ i Average required fuel burn cost per minute for air-
∀i, j ∈ A, i 6= j, M  0. (5)
craft i to be delayed.
FB− i Average required fuel burn cost per minute for air- • Time Constraint: Based on operational and technical
craft i to be advanced. considerations such as limited fuel, airspeed, etc., each
 aircraft has a maximum and a minimum allowable air- [4] R. Dear, “The Dynamic Scheduling of Aircraft in the Near Terminal
borne time which have to be treated as hard constraints. Area,” R76-9, Flight Transportation Laboratory, MIT, USA, Tech. Rep.,
1976.
ELTi ≤ SLTi ≤ LLTi ∀i ∈ A. (6) [5] H. N. Psaraftis, “A Dynamic Programming Approach to the Aircraft
Sequencing Problem,” R78-4, Flight Transportation Laboratory, MIT,
A time slot (or time window) assigned to each landing USA, Tech. Rep., 1978.
[6] L. Bianco, P. Dell’Olmo, and S. Giordani, “Minimizing Total Comple-
aircraft which typically starts 5 minutes before TLTi and tion Time Subject to Release Dates and Sequence-Dependent Processing
ends 10 minutes after TLTi does not necessarily coincide Times,” Annals of Operations Research, vol. 86, pp. 393–415, 1999.
with the time constraint. [7] A. M. Bayen, C. J. Tomlin, Y. Ye, and J. Zhang, “An Approxima-
tion Algorithm for Scheduling Aircraft with Holding Time,” in 43rd
• Time Shifting: There is limited flexibility in moving the IEEE Conference on Decision and Control, Atlantis, Paradise Island,
aircraft’s landing time either forward or backward in time Bahamas, December 14-17, 2004.
relative to its estimated landing time. Time shifting is [8] H. Balakrishnan and B. Chandran, “Scheduling Aircraft Landings under
Constrained Position Shifting,” in AIAA Guidance, Navigation and
considered rather than position shifting in re-sequencing Control Conference and Exhibit, Keystone, Colorado, USA, August 21-
the aircraft since it can be dependent on aircraft type. 24, 2006.
[9] B. Chandran and H. Balakrishnan, “A Dynamic Programming Algorithm
(Esti − TSi ) ≤ SLTi ≤ (Esti + TSi ) ∀i ∈ A. (7) for Robust Runway Scheduling,” in Proceedings of the American Control
Conference, New York, NY, USA, July 11-13, 2007.
• Precedence Constraint: Airline preferences may dictate [10] H. Lee and H. Balakrishnan, “Fuel Cost, Delay and Throughput
Tradeoffs in Runway Scheduling,” in Proceeding of American Control
that one aircraft should land before another. Conference (ACC 08), Seattle, Washington, USA, June 11-13, 2008.
[11] A. R. Brentnall, “Aircraft Arrival Management,” Ph.D. dissertation,
SLTi Pij < SLTj ∀i, j ∈ A, i 6= j. (8) University of Southampton, UK, 2006.
[12] A. R. Brentnall and R. C. H. Cheng, “Some Effects of Aircraft Arrival
IV. F UTURE R ESEARCH D IRECTIONS Sequence Algorithms,” Journal of Operational Research Society, pp.
Although many research papers on the ALP have been 1–11, 2008.
[13] J. E. Beasley, M. Krishnamoorthy, Y. M. Sharaiha, and D. Abramson,
published during the last three decades, the majority have not “Scheduling Aircraft Landings - The Static Case,” Transportation Sci-
developed methods that have been implemented. The reason ence, vol. 34(2), pp. 180–197, 2000.
could be because the methods may relax or dismiss hard [14] J. Abela, D. Abramson, M. Krishnamoorthy, and A. D. Silva, “Com-
puting Optimal Schedules for Landing Aircraft,” in Proceeding of 12th
(critical) operational constraints, have unreasonable algorithm National Conference of the Australian Society for Operations Research,
runtime, study a static rather than dynamic environment, Adelaide, Australia, July 7-9, pp. 71 - 90, 1993.
ignore the requirements of the various stakeholders, or depend [15] G. Stevens, “An Approach to Scheduling Aircraft Landing Times Using
Genetic Algorithms,” Honours Thesis, Department of Computer Science,
on features of a specific airport. Existing research generally RMIT University, Melbourne, Australia, 1995.
considers some of the common and obvious constraints. This [16] V. Ciesielski and P. Scerri, “An Anytime Algorithm for Scheduling of
research aims to capture vital operational constraints that have Aircraft Landing Times Using Genetic Algorithms,” Australian Journal
of Intelligent Information Processing Systems, vol. 4(3/4), pp. 206–213,
been observed from the daily work of controllers in our model 1997.
building. [17] ——, “Real Time Genetic Scheduling of Aircraft Landing Times,” in D.
Solution approaches for the model remain to be developed. Fogel (Editor), Proceedings of The 1998 IEEE International Conference
on Evolutionary Computation (ICEC’98), Anchorage, Alaska, May 4-9,
The solutions must be obtained quickly to be of use to air traf- 1998.
fic controllers. Since the problem is complex (it is NP-hard), [18] V. H. L. Cheng, L. S. Crawford, and P. K. Menon, “Air Traffic
heuristic methods including local search algorithms may be Control Using Genetic Search Techniques,” in Proceedings of the IEEE
International Conference on Control Applications, Hawaii, HA, USA,
more appropriate than enumerative methods such as dynamic August 22-27, 1999.
programming which can be computationally demanding. [19] J. E. Beasley, J. Sonander, and P. Havelock, “Scheduling Aircraft
Landing at London Heathrow using a Population Heuristic,” Journal
ACKNOWLEDGMENT of Operational Research Society, vol. 52, pp. 483–493, 2001.
[20] J. V. Hansen, “Genetic Search Methods in Air Traffic Control,” Com-
This work has been co-financed by the European Organ- puters & Operations Research, vol. 3, pp. 445–459, 2004.
isation for the Safety of Air Navigation (EUROCONTROL) [21] X.-B. Hu and W.-H. Chen, “Genetic Algorithm Based on Receding
under its Research Grant scheme. Horizon Control for Arrival Sequencing and Scheduling,” Engineering
Applications of Artificial Intelligence, vol. 18(5), pp. 633–642, 2005.
The content of the work does not necessarily reflect the [22] X.-B. Hu and E. D. Paolo, “Binary-Representation-Based Genetic Al-
official position of EUROCONTROL on the matter. gorithm for Aircraft Arrival Sequencing and Scheduling,” IEEE Trans-
C 2010, EUROCONTROL and the University of Southamp-
actions on Intelligent Transportation Systems, vol. 9(2), pp. 301–310,
2008.
ton. All Rights reserved. [23] ——, “An Efficient Genetic Algorithm with Uniform Crossover for Air
Traffic Control,” Computer & Operations Research, vol. 36, pp. 245–
R EFERENCES 259, 2009.
[1] A. Blumstein, “The landing Capacity of a Runway,” Operations Re- [24] H. Pinol and J. E. Beasley, “Scatter Search and Bionomic Algorithms
search, vol. 7(6), pp. 752–763, 1959. for the Aircraft Landing Problem,” European Journal of operational
[2] S. Capri and M. Ignaccolo, “Genetic Algorithms for Solving the Research, vol. 171, pp. 439–462, 2006.
Aircraft-Sequencing Problem: The Introduction of Departures into the [25] M. C. Randall, “Scheduling Aircraft Landings Using Ant Colony Op-
Dynamic Model,” Journal of Air Transport Management, vol. 10(5), pp. timisation,” in Proceedings of The IASTED International Conference
345–351, 2004. Artificial Intelligence and Soft Computing, Banff, Canada, July 17-19,
[3] K. Artiouchine, P. Baptiste, and C. Durr, “Runway Sequencing with 2002.
Holding Patterns,” European Journal of Operational Research, vol.
189(3), pp. 1254–1266, 2008.