You are on page 1of 8

Available online

Available
Available at www.sciencedirect.com
online
online at
at www.sciencedirect.com
www.sciencedirect.com

ScienceDirect
Transportation Research Procedia 47 (2020) 299–306
Transportation
Transportation Research
Research Procedia
Procedia 00
00 (2019)
(2019) 000–000
000–000 www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia

22nd EURO Working Group on Transportation Meeting, EWGT 2019, 18-20 September 2019,
Barcelona, Spain

Integration of airport terminal arrival route selection, runway


assignment and aircraft trajectory optimisation
Adrian Bareaa,∗
a,∗, Raul de Celisaa , Luis Cadarsoaa
aa European Institute
European Institute for
for Aviation
Aviation Training
Training and
and Accreditation,
Accreditation, Rey
Rey Juan
Juan Carlos
Carlos University,
University, Fuenlabrada
Fuenlabrada (Madrid)
(Madrid) 28943,
28943, Spain
Spain

Abstract
Abstract
Incoming
Incoming air air traffic
traffic in
in aa given
given airport
airport can
can be
be provided
provided byby aa great
great diversity
diversity of
of air
air routes.
routes. However,
However, airports
airports comprise
comprise aa limited
limited
number
number of of runways.
runways. The The reduction
reduction inin the
the number
number of of paths
paths that
that aircraft
aircraft can
can transit
transit takes
takes place
place in
in terminal
terminal arrival
arrival routes,
routes, which
which actact as
as
an
an interface
interface between
between incoming
incoming routes
routes and
and approach
approach trajectories.
trajectories. This
This occurrence
occurrence entails
entails that
that air
air traffic
traffic has
has to
to be
be carefully
carefully managed
managed
in
in terminal
terminal arrival
arrival routes
routes inin order
order to
to prevent
prevent possible
possible bottlenecks.
bottlenecks. This
This work
work presents
presents an
an optimisation
optimisation model
model that
that manages
manages notnot only
only
approach
approach and and landing
landing operations,
operations, but
but also
also terminal
terminal arrival
arrival routes,
routes, deciding
deciding on
on runway
runway assignment,
assignment, terminal
terminal arrival
arrival route
route selection
selection
and
and aircraft
aircraft trajectory.
trajectory. The
The proposed
proposed integrated
integrated model
model leads
leads to
to aa mixed
mixed integer
integer non-linear
non-linear problem.
problem. ForFor its
its resolution,
resolution, aa Benders
Benders
decomposition
decomposition is proposed. On the one hand, the master model deals with runway assignment and terminal arrival route
is proposed. On the one hand, the master model deals with runway assignment and terminal arrival route selection,
selection,
making
making use use of
of aa set
set of
of binary
binary variables.
variables. On
On the
the other
other hand,
hand, the
the sub-model
sub-model deals
deals with
with the
the trajectory
trajectory calculation
calculation problem,
problem, managing
managing
aa set
set of
of continuous
continuous variables
variables and
and minimising
minimising aa combination
combination of of fuel
fuel consumption
consumption andand delay.
delay.
© 2020 The Authors. Published by Elsevier B.V.
cc 2020

 2020
This The
The
is an Authors.
Authors.
open Published
accessPublished by Elsevier
by
article under Elsevier B.V.
B.V.
the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review under
Peer-review under responsibility
responsibility of
of the
the scientific
scientific committee
committee of
committee of the
of the 22nd
the 22nd EURO
22nd EURO WorkingGroup
Working
Euro Working Grouponon
Group onTransportation
TransportationMeeting
Transportation Meeting..
Meeting..
Keywords: Air
Keywords: Air Traffic
Traffic Management;
Management; Optimal
Optimal Control;
Control; Decomposition
Decomposition Methods
Methods

1.
1. Introduction
Introduction

The
The current
current growth
growth in in air
air traffic
traffic demand
demand motivates
motivates thethe investment
investment in in the
the development
development of of technologies
technologies in
in order
order to
to
comply with
comply with the
the future
future evolution
evolution ofof air
air transport
transport industry
industry by by means
means ofof optimising
optimising the
the operational
operational performance
performance andand
automatising
automatising thethe processes
processes for
for air
air traffic
traffic management.
management. Terminal
Terminal area
area and
and airport
airport runway
runway management
management are are ongoing
ongoing
challenges
challenges for
for air
air traffic
traffic controllers.
controllers. Bennell
Bennell et
et al.
al. (2013)
(2013) review
review the
the techniques
techniques and
and tools
tools of
of operational
operational research
research and
and
management
management science
science that
that are
are used
used for
for scheduling
scheduling aircraft
aircraft landings
landings and
and take-offs.
take-offs. Kim
Kim etet al.
al. (2014)
(2014) present
present anan opti-
opti-
misation
misation model
model for
for simultaneously
simultaneously assigning
assigning aircraft
aircraft to
to runways
runways andand scheduling
scheduling the
the arrival
arrival and
and departure
departure operations
operations
on
on these
these runways.
runways. Chandrasekar
Chandrasekar and and Hwang
Hwang (2014)
(2014) propose
propose aa framework
framework to to compute,
compute, with
with computational
computational efficiency,
efficiency,
the
the optimal
optimal runway
runway assignment
assignment andand sequencing
sequencing of of arrival
arrival and
and departure
departure operations
operations at
at an
an airport.
airport. The
The performance
performance of of
continuous
continuous descent
descent approach
approach procedure,
procedure, via via multi-phase
multi-phase optimal
optimal vertical
vertical trajectory
trajectory generation
generation problems
problems with
with respect
respect

∗∗ Corresponding
Corresponding author.
author.
E-mail address:
E-mail address: adrian.barea@urjc.es
adrian.barea@urjc.es

2352-1465 
2352-1465 cc 2020
2020 The
The Authors.
Authors. Published
Published byby Elsevier
Elsevier B.V.
B.V.
Peer-review©under
2352-1465
Peer-review
2020responsibility
under The Authors. of Published
responsibility of the scientificbycommittee
the scientific Elsevier of
committee B.V.
the 22nd
of the 22nd EURO
EURO Working
Working Group
Group on
on Transportation
Transportation Meeting..
Meeting..
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review under responsibility of the scientific committee of the 22nd Euro Working Group on Transportation Meeting.
10.1016/j.trpro.2020.03.102
300 Adrian Barea et al. / Transportation Research Procedia 47 (2020) 299–306
2 Adrian Barea et al. / Transportation Research Procedia 00 (2019) 000–000

Fig. 1. a) Route network b) Aircraft forces

to flight time and fuel consumption, is investigated in Park and Clarke (2015). A cooperative genetic algorithm to find
a feasible runway configuration for a given traffic demand at an airport is presented in Ahmed et al. (2018). Murata
et al. (2018) focus on how to reduce the cognitive loads of air traffic controllers while optimising aircraft landing
sequences.
Standard terminal arrival routes (STAR) link the last point of aircraft routes with initial approach fix (IAF) points,
which represent the beginning of approach trajectories. There is a considerable number of routes that connect a given
airport with incoming air traffic. Those routes converge in a common set of final points. In turn, STARs connect such
set of final points with an even smaller set of IAF points. These sequential reductions in the number of paths that
aircraft can transit are prone to produce a bottleneck effect, in particular during high-demand time periods, in which a
number of aircraft have to perform holding manoeuvres while others are performing their approach operations. This
fact results in additional fuel consumption and delays.
This paper aims at integrating terminal arrival route selection, runway assignment and trajectory optimisation.
Height constraints are introduced to comply with aircraft noise and emissions rules, as well as time separation con-
straints to prevent aircraft collisions. A combination of fuel consumption and arrival time delay is optimised. Each
segment between a clearance point and an IAF or between an IAF and a runway is modelled as a 2D trajectory. The
development of this research is motivated by two main premises.
On the one hand, the integration of terminal arrival route selection or runway assignment, topics related to air traffic
controllers, with the selection of aircraft thrust and aerodynamic parameters to achieve the desired trajectory, topics
related to aircraft pilots, which results in a more general approach to aircraft landing operations. On the other hand,
the unification of terminal arrival route selection and runway assignment as a single problem so as to simultaneously
decide the final point of aircraft air route, number and time of holding manoeuvres, IAF points and runway assigned
to each aircraft.
The implementation of these premises achieves a novel scheme that handles a broader problem than those previ-
ously considered in the literature, providing a better solution than the one that would be obtained with the individual
resolution of each of the problems.

2. Problem description

This section presents a qualitative description of the problem to be solved. First, the network in which aircraft
terminal arrival route operations are performed is explained. Second, the needed concepts of aircraft dynamics to
optimise aircraft trajectories are expounded. Third, the specific constraints imposed to aircraft terminal arrival route
operations are presented.

2.1. Aircraft terminal arrival route network

Terminal arrival route structure comprises a set of waypoints linked by route segments. The different types of
waypoints are depicted in Figure 1a):

• Clearance Limit (CL): Points which cannot be traversed without Air Traffic Control (ATC) authorisation.
• Initial Approach Fix (IAF): Points that define the beginning of approach trajectory.
Adrian Barea et al. / Transportation Research Procedia 47 (2020) 299–306 301
Adrian Barea et al. / Transportation Research Procedia 00 (2019) 000–000 3

• Intersection Points (IP): They define the beginning of approach trajectory segments which are common for
every aircraft that lands in a specific runway, regardless of the selected IAF. Such segments are of a critical
importance because of the bottleneck effects that aircraft coming from different IAFs to land in the same runway
may produce.
• Runways (RWY): Ending point of landing operations.

Both CLs and IAFs, unlike IPs, have holding circuits in which aircraft may have to wait for ATC authorisation to
proceed with further operations. Thus, the route followed by an aircraft is defined according to the following sequence:

1. A CL is associated to the aircraft. When the aircraft reaches it, it may have to wait in a holding circuit for a
certain time period.
2. An IAF is associated to the aircraft, which leaves the CL (or the holding circuit associated to it) to proceed to the
IAF. When the IAF is reached, the aircraft may have to wait in another holding circuit.
3. A runway is assigned to the aircraft, which leaves the IAF (or the holding circuit associated to it) and proceeds
to finish the landing operations.

Every aircraft must follow a route which is defined as a succession of the previously introduced points. These
points are linked by means of air segments, where aircraft performance is to be optimised. The selection of the route
to be followed, defined as a sequence of CL, IAF, IP and RWY, is one of the main decisions of the problem.

2.2. Aircraft dynamics

Once the aircraft route is selected, flight performance is optimised, in order to minimise fuel consumption. Note
that fuel consumption is not the only objective of the problem but it is the one which has the greatest impact on flight
dynamics.
Each of the route segments described before and depicted in Figure 1a) are modelled as 2-dimensional motions in
vertical planes. Consequently, flight dynamics equations are simplified and turns in the trajectory are neglected. Note
that this simplification permits to facilitate the problem exposition and the solution approach but it may be got over
with a 3-dimensional model.
Aircraft trajectories refer to the evolution of aircraft position during the considered operations along every segment.
Aircraft position is determined by two variables. Range (x) is the horizontal distance aircraft have travelled with
respect to an arbitrary reference point. Height (h) is the aircraft altitude with respect to sea level. Moreover, aircraft
orientation is defined by the descent angle γ, shown in Figure 1b). Aircraft experiment a series of forces which modify
the aircraft trajectory, which are depicted in Figure 1b):

• Lift force (L) is the aerodynamic force that elevates the aircraft in a direction perpendicular to its displacement
with respect the surrounding air.
• Drag force (D) is the aerodynamic force that acts against the direction of the aircraft displacement with respect
the surrounding air.
• Thrust (T) is the force that engines produce.
• Weight (W) is the attraction of the Earth gravity, perpendicular to the ground.

Pilots can modify aircraft parameters that have influence over such forces in order to control its trajectory, which
is in a vertical plane for the presented problem. On the one hand, engine thrust (T) can be controlled to accelerate
or decelerate the aircraft. On the other hand, aerodynamic control devices allow to modify aircraft geometry in order
to have influence on the descent angle (γ), the lift coefficient (cl ), which is related to the lift force (L), and the drag
coefficient (cd ), which is related to the drag force (D).
302 Adrian Barea et al. / Transportation Research Procedia 47 (2020) 299–306
4 Adrian Barea et al. / Transportation Research Procedia 00 (2019) 000–000

2.3. Operational constraints

Approach and departure trajectories and landing and take-off operations are subject to limitations, which usually
depend on the terminal area, airport and runway where the aircraft flies. Without loss of generality, they may be
summarised as follows:

• Height constraints: airports impose minimum height constraints for each route segment, not only to prevent
collisions with obstacles, but also to comply with sound and emission laws that regulate airport environmental
impact.
• Speed constraints: each aircraft model has a set of specified speeds for each operational phase. Moreover,
airports can impose speed limitations for each route segment in order to guarantee safety of operations and
optimise airport capacity.
• Time separation constraints: aircraft whose intended trajectories share points that are close in space should
reach them at sufficiently different times. That way, several aircraft can maintain similar trajectories without
risk of collision and wake turbulence from an aircraft on a similar track ahead, which can lead to loss of control,
is mitigated.

3. Mathematical model

This section defines a mathematical model that integrates approach trajectory optimisation, terminal arrival route
selection and runway assignment. Before introducing the mathematical formulation, the notations (sets, parameters
and variables) are explained below.

Nomenclature

Sets
F is the set of aircraft indexed by f and f1 .
O is the set of Clearance Limits indexed by o.
Q is the set of Initial Approach Fixes indexed by q.
N is the set of runways indexed by n.
Z is the set of route elements, namely, zCL (Clearance Limit), zIAF (Initial Approach Fix) or zRWY (Runway),
indexed by z.

Parameters
α is the weight of delay with respect to final mass in objective function.
te f is the estimated landing time of aircraft f ∈ F.
σ is the minimum time separation of aircraft.
g is the gravity acceleration.
ρ is the air density.
Sf is the wing surface of aircraft f ∈ F.
Vdes f is the descent speed for Flight Level 100 and below of aircraft f ∈ F.
Vapp f is the approach speed in landing configuration for aircraft f ∈ F.
cd0i f is the constant part of the parasite drag force coefficient of aircraft f ∈ F.
Kf is the induced drag parameter of aircraft f ∈ F.
c f 1 f is a fuel consumption parameter of aircraft f ∈ F.
c f 2 f is a fuel consumption parameter of aircraft f ∈ F.
m s2k f is a fuel consumption parameter of aircraft f ∈ F.
Variables
x f ∈ R+ is the range of aircraft f ∈ F.
Adrian Barea et al. / Transportation Research Procedia 47 (2020) 299–306 303
Adrian Barea et al. / Transportation Research Procedia 00 (2019) 000–000 5

h f ∈ R+ is the height of aircraft f ∈ F.


V f ∈ R+ is the speed of aircraft f ∈ F.
γf ∈ R is the descent angle of aircraft f ∈ F.
m f ∈ R+ is the mass of aircraft f ∈ F.
Lf ∈ R is the aerodynamic lift of aircraft f ∈ F.
cl f ∈ R is the lift coefficient of aircraft f ∈ F.
D f ∈ R+ is the aerodynamic drag of aircraft f ∈ F.
cd f ∈ R+ is the drag (parasite and induced) coefficient of aircraft f ∈ F.
cd0 f ∈ R+ is the variable part of the parasite drag coefficient of aircraft f ∈ F.
Tf ∈ R is the thrust of aircraft f ∈ F.
t f ∈ R+ is the time of aircraft f ∈ F.
r+f ∈ R+ is the positive delay of aircraft f ∈ F, if it finishes operations later than the planned time.
r−f ∈ R+ is the negative delay of aircraft f ∈ F, if it finishes operations before than the planned time.
ψ f,o,q,n ∈ {0, 1} is 1 if aircraft f ∈ F selects the route defined by CL o ∈ O, IAF q ∈ Q and runway n ∈ N. It is 0
otherwise.
δzf, f1 ∈ {0, 1} is 1 if routes of aircraft f ∈ F, f1 ∈ F share the same route element z ∈ Z. It is 0 otherwise.

Note that the most central decision variables are ψ f,o,q,n , which define the route the aircraft follow from the initial
point until the runway, and δzf, f1 , which determine if two aircraft share the same route element. Once the route is
decided, aircraft performance is optimised in order to minimise fuel consumption while conflicts are avoided and
delays minimised.
The objective function (1) minimises a combination of delay and final mass, which is a measure of fuel consump-
tion, for every considered aircraft. The relative importance of such quantities is regulated by the α parameter. Only the
positive part of delay has been consideredso that aircraft that arrive early do not  compensate aircraft that arrive late.
   
f inal
min α r+f − (1 − α) m f (t f = t f ) (1)
f ∈F f ∈F

Nonlinear aircraft flight dynamics modelling is defined by equations (2) to (5):

• Kinematic equations (2) define the evolution  of aircraft


 position asa function of speed and descent angle.
ẋ f , ḣ f = V f cos(γ f ) , sin(γ f ) ∀ f ∈ F (2)
• Equations (3) define the aerodynamic lift and drag forces that aircraft experiencealong the trajectories.
  1  
L f , D f , cd f = ρS f (V f )2 cl f , cd f , cd0i f + cd0 f + K f (cl f )2 ∀ f ∈ F (3)
2
• Dynamics equations (4) relate the forces applied to the aircraft with the accelerations they experiment. Such
equations define the trajectories the aircraft follow. 
  Tf Df Lf g
V̇ f , γ̇ f = − − g sin(γ f ), − cos(γ f ) ∀ f ∈ F (4)
mf mf mf Vf Vf
• Equations (5) define the aircraft mass variation as afunction of thrust,  speed and fuel consumption parameters.
m s2k f
ṁ f = −T f c f 1 f 1 + V f ∀f ∈ F (5)
c f 2f
• Equations (6) compute the delay associated to each aircraft. Its value is modelled as the subtraction of two
positive variables that represent a positive delay and a negative delay so that the objective function (1) only
considers delays in which aircraft finish operations behind schedule.
r+f − r−f = t ff inal − te f ∀ f ∈ F (6)

Height, time separation, initial and final conditions, route and runway constraints follow.
304 Adrian Barea et al. / Transportation Research Procedia 47 (2020) 299–306
6 Adrian Barea et al. / Transportation Research Procedia 00 (2019) 000–000

• Constraints (7) stand for height limitations to comply with noise and emission rules. The function in F(x f ) is
defined according to the terminal area or airport where flight operations take place.
h f ≥ F(x f ) ∀ f ∈ F (7)
• Constraints (8) impose that two aircraft that share a route element z ∈ Z must be separated in time by σ seconds
when reaching them. Note that xz places the route element z in space.
t (x = xz ) − t (x = xz ) ≥ σδz ∀ f ∈ F, ∀ f ∈ F : f < f , ∀z ∈ Z (8)
f1 f1 f f f, f1 1 1
• Constraints (9) impose initial and final speeds defined for each aircraft model so that operations can be per-
formed safely.
V f (t f = tinitial
f ) = Vdes f , V f (t f = t ff inal ) = Vapp f ∀ f ∈ F (9)
• Constraints (10) impose that each aircraft is 
associated
 to exactly one route.
ψ f,o,q,n = 1 ∀ f ∈ F (10)
o∈O q∈Q n∈N
• Constraints (11) - (13) unambiguously define the relation between route variables ψ f,o,q,n and δzf, f1 . Two auxiliary
sets Ωz and A have been used to compactly describe such constraints. A represents all the elements of O, Q and
N sets: A = O ∪ Q ∪ N. For each route element z, Ωz represents the elements of two of the sets O, Q and N
so that the excluded set corresponds to the route element of the selected z, that is, Ωz may be Q ∪ N, O ∪ N or
O ∪ Q, depending on z.
   
ψ f,o,q,n + ψ f1 ,o,q,n ≤ 1 + δzf, f1 ∀ f ∈ F, ∀ f1 ∈ F : f < f1 , ∀z ∈ Z, ∀ (A − Ωz ) (11)

z Ωz
    
ψ f,o,q,n − ψ f1 ,o,q,n ≤ 1 − δzf, f1 ∀ f ∈ F, ∀ f1 ∈ F : f < f1 , ∀z ∈ Z, ∀ (A − Ωz ) (12)
Ωz Ωz
   
ψ f1 ,o,q,n − ψ f,o,q,n ≤ 1 − δzf, f1 ∀ f ∈ F, ∀ f1 ∈ F : f < f1 , ∀z ∈ Z, ∀ (A − Ωz ) (13)
Ωz Ωz

4. Solution approach: time discretisation and Benders decomposition

To formulate a Mixed Integer Nonlinear Programming model, differential equations (2), (4) and (5) are discretised.
That is, a set of discrete time instances j ∈ J is determined, in which the aircraft dynamics state is defined by such
equations. The discretization of such equations is performed by means of an adaptive-step implicit trapezoidal rule
integration scheme, which is given in (14).
∆t  
Ẋ(t) = f (X(t)) −→ X(t j+1 ) − X(t j ) = f (X(t j+1 )) + f (X(t j )) (14)
2
The integration step ∆t, which gives the distance between two time instances, is not a predefined parameter but a
decision variable. As a result, the optimisation model tunes the integration step, refining the mesh or making it larger,
as a function of the behaviour of the rest of variables.
Benders decomposition (Benders (1962), Geoffrion (1972)) is an iterative technique to solve problems with a
structure such that, when some variables are fixed, the resulting problem is significantly easier to solve than the original
one. The approach consists of iteratively solving the master model and the sub-model until the optimal solution is
found, if the problem is convex, or until a local optimum is found otherwise. Here, it allows to split the Mixed Integer
Nonlinear programming model into the following master model and sub-model.
The master model is a Mixed Integer Linear programming problem that performs terminal arrival route selection
and runway assignment. Thus selecting the values of variables ψ f,o,q,n and δzf, f1 . This problem includes constraints
(10) - (13). A new variable θ that defines the value of the master model objective function (15) is introduced, which
constitutes a lower bound of the optimal solution in the solution finding process.
min(θ) (15)
The sub-model is a Nonlinear programming problem that optimises performances of the trajectories. This problem
includes objective function (1) and constraints (2) - (9). New constraints (16) and (17) are introduced to fix the value
of binary variables δzf, f1 and ψ f,o,q,n with the ones obtained in the previous resolution of the master model  δzf, f1 and

ψ f,o,q,n .
δz = 
f, f1 δz (πz ) ∀ f ∈ F, ∀ f1 ∈ F : f < f1 , ∀z ∈ Z
f, f1 A f, f1 (16)
AdrianetBarea
Adrian Barea et al. / Transportation
al. / Transportation ResearchResearch
Procedia Procedia
00 (2019)47 (2020) 299–306
000–000 305
7

Fig. 2. a) Madrid Barajas route network b) Operational case

ψ f,o,q,n =  ψ f,o,q,n (πB f,o,q,n ) ∀ f ∈ F, ∀o ∈ O, ∀q ∈ Q, ∀n ∈ N (17)


z 
Dual variables πA f, f and πB f,o,q,n of constraints (16) and (17), together with the value of the objective function (1), ob j,
1
are used to generate Benders cuts (18), which are added to the master model iteratively. Benders cuts constrain the
value of θ so that at subsequent iterations the master model provides tighter lower bounds to the optimal solution while
generating new values of binary variables δzf, f1 and ψ f,o,q,n to be used in the following sub-model resolution. The value
of the objective function (1) provides an upper bound of the optimal solution.
         
θ ≥ obj + δz − 
f, f1 δz πz
f, f1 A f, f1+ ψ f,o,q,n − 
ψ f,o,q,n πB f,o,q,n (18)
f ∈F f1 ∈F z∈Z f ∈F o∈O q∈Q n∈N
Master model and sub-model are solved iteratively until the difference between upper and lower bounds is less than
a predefined tolerance. Then, the best upper bound obtained is considered the optimal solution. Although Benders
decomposition is an exact method, it cannot guarantee global optimality due to non-linearities.

5. Computational experiments

Real operational data of Madrid Barajas airport have been used for computational experiments. Figure 2a) shows a
schematic representation of Madrid Barajas approach terminal arrival route structure. It comprises a set of CLs, IAFs,
IPs and runways linked by route segments, whose associated height constraint functions F(x f ) from constraints 7,
are defined according to STAR and instrument approach procedure charts. Madrid Barajas can operate according to
two configurations. North configuration makes use of 32L and 32R runways and their associated CLs and IAFs. In
turn, South configuration makes use of 18L and 18R runways. A real operational case involving 175 aircraft has been
considered, shown in figure 2b). It depicts the number of aircraft that intend to land at Madrid Barajas airport as a
function of estimated landing time. In order to test the quality of the solutions generated by this methodology, two
different scenarios have been devised. They are defined as a function of a characteristic time, namely, the time each
aircraft would take to perform the landing trajectory with a constant speed equal to the average of Vdes f and Vapp f .
In Scenario 1, available time for each aircraft to complete the landing operations without delay is the characteristic
time plus 6 minutes. This represents an ideal low-conflict operational case where an intelligent waypoint assignment
and trajectory management should achieve a null total delay while reducing fuel consumption.
In Scenario 2, available time is the characteristic time. This represents a highly congested operational case where
delays are unavoidable. In this case, delay and fuel consumption improvements are in conflict because the faster
aircraft travel to reduce delays, the more fuel is consumed. A good solution should achieve a compromise where fuel
consumption is as similar as possible to the one of scenario 1 while an acceptable delay is accomplished.
The complete test case is solved by means of rolling horizon technique. That is, operations are partitioned in
intervals so that the number of aircraft considered in each interval results in a manageable optimisation problem.
Then, those intervals are sequentially optimised while fixing the solutions of previous ones.
In this particular case, operations have been partitioned in 30 minutes intervals, except the 14:00-15:30 range, that
has been partitioned in 15 minutes intervals to deal with incoming aircraft peaks at 14:35 and 15:00. This way, the
maximum number of aircraft considered in a single optimisation problem is 16, corresponding to 13:30-14:00 interval.
306 Adrian Barea et al. / Transportation Research Procedia 47 (2020) 299–306
8 Adrian Barea et al. / Transportation Research Procedia 00 (2019) 000–000

The model is coded in GAMS 25.1.1 and solved with CONOPT 4.05 and CPLEX 12.8.0 in a PC featuring an
Intel Core i5-7500 and 8GB RAM. Table 1 shows the results of rolling horizon technique for both Madrid Barajas
configurations and both operational scenarios, where the optimality gap is the magnitude of the difference between
upper and lower bounds divided by the lower bound, that is, |(U B − LB)/LB|.

Table 1. Rolling horizon solution


Configuration Scenario Delay (s) Consumed fuel (kg) CPU time (s) Gap
South 1 0 103870.5 13400 1.19 · 10−3
South 2 2217 104141.2 12999 1.84 · 10−3
North 1 0 84967.2 10712 7.58 · 10−4
North 2 685.06 85156.7 10535 9.67 · 10−4

The non-linearity and non-convexity of trajectory dynamics implies that the proposed method cannot distinguish
between local optima and the global optimum. However, the obtained results show that good-quality solutions have
been achieved. Scenario 1 (low-conflict) achieves a null delay while showing a favourable fuel consumption. Scenario
2 (high-conflict) achieves a reasonable delay while showing a fuel consumption similar to the one of scenario 1.
Regarding computational time, given the scale of the problem and the complexity of the model, the times achieved
are satisfactory for pre-tactical planning applications, in which aircraft trajectory optimisation, terminal arrival route
selection and runway assignment will be performed several days prior to the day of operations. However, the average
computational time for solving time-periods in the rolling horizon is 744 seconds. Note that periods that contain many
simultaneous or next in time incoming aircraft (i.e., at 14:35 or at 15:00) feature bigger computational times, ranging
from 1323 to 2347 seconds. Thus, techniques for a better management of those peak time-periods have to be explored
to further accelerate the resolution of the problem.

6. Conclusions

A mathematical formulation of a problem that unifies the computation of approach trajectories, terminal arrival
route selection and runway assignment has been described. Such formulation has been discretised to obtain a mixed
integer nonlinear programming problem, which is solved by means of a Benders decomposition. In order to test
the performance of the proposed method, a large scale real operational case involving 175 aircraft, corresponding to
Madrid Barajas airport is presented. To obtain a solution for the whole operational case, a rolling horizon technique, in
which operations are partitioned in subintervals and sequentially optimised, has been carried out, achieving accurate
solutions in a computational time that is suitable for pre-tactical planning applications.

Acknowledgements

This research was supported by the “Ministerio de Economia y Competitividad, Spain” and “Consejeria de Educa-
cion e Investigacion, Comunidad de Madrid”. [Grants TRA2016-76914-C3-3-P and PEJD-2017-PRE/IND-4470].

References

Ahmed, M., Alam, S., Barlow, M., 2018. A cooperative co-evolutionary optimisation model for best-fit aircraft sequence and feasible runway
configuration in a multi-runway airport. Aerospace 5, 85.
Benders, J.F., 1962. Partitioning procedures for solving mixed-variables programming problems. Numerische mathematik 4, 238–252.
Bennell, J.A., Mesgarpour, M., Potts, C.N., 2013. Airport runway scheduling. Annals of Operations Research 204, 249–270.
Chandrasekar, S., Hwang, I., 2014. Algorithm for optimal arrival and departure sequencing and runway assignment. Journal of Guidance,
Control, and Dynamics 38, 601–613.
Geoffrion, A.M., 1972. Generalized benders decomposition. Journal of optimization theory and applications 10, 237–260.
Kim, B., Li, L., Clarke, J.P., 2014. Runway assignments that minimize terminal airspace and airport surface emissions. Journal of Guidance,
Control, and Dynamics 37, 789–798.
Murata, A., Sato, H., Takadama, K., 2018. Towards adaptive aircraft landing order with aircraft routes partially fixed by air traffic controllers
as human intervention. SICE Journal of Control, Measurement, and System Integration 11, 105–112.
Park, S.G., Clarke, J.P., 2015. Optimal control based vertical trajectory determination for continuous descent arrival procedures. Journal of
Aircraft 52, 1469–1480.

You might also like