Conflict-Free Trajectory Planning Based on a Data-Driven
Conflict-Resolution Model
Esther Calvo-Fernández
∗
and Luis Perez-Sanz
†
Universidad Politécnica de Madrid, 28040 Madrid, Spain
José Manuel Cordero-García
‡
CRIDA, ATM R&D Reference Center, 28022 Madrid, Spain
and
Rosa María Arnaldo-Valdés
§
Universidad Politécnica de Madrid, 28040 Madrid, Spain
DOI: 10.2514/1.G000691
In a trajectory-based operations
’ environment, at some time during the planning phase, shared business
trajectories will become reference business trajectories, and not necessarily conflict free. This paper presents a way of
obtaining a conflict-free solution for all planned trajectories during the strategic phase (before becoming reference
business trajectories). The proposed methodology incorporates 1) a data-driven conflict-resolution model, and 2) a
multiobjective global optimization that considers the interests of a variety of actors in the air traffic management
community: particularly, air navigation service providers and airlines.
Nomenclature
C
=
number of conflicts during a single scenario
d
=
distance, n miles
F
=
conflict-resolution model nonlinear map
f
=
objective of the multiobjective problem
h
=
altitude, ft
k
=
number of clusters
o
=
number of observations to build a histogram
p
n
=
fixed points of the flight plan
S
=
number of scenarios
t
=
time, s
tanα = slope of the Pareto frontier curve
X
i
=
input variables of the conflict-resolution model
x
=
vector of decision variables
Y
i
=
output variables of the conflict-resolution model
ε
i
=
constraint vector of the
ϵ-constraint method
λ
=
longitude, °
Σ
=
solution space of the multiobjective problem
φ
=
latitude, °
I.
Introduction
I
N THE future, air traffic management (ATM) will have to become
increasingly automated in order to handle greater complexity and
larger volumes of traffic.
ATM in Europe is evolving around the trajectory-based operations
(TBOs) concept of operations (SESAR Consortium) [1]. It relies on
individual four-dimensional (4-D) trajectories to pave the way for
new levels of planning, contributing to the objective of dealing with
increasingly complex situations. As such, a TBO offers new
possibilities as regards planning in the strategic phase. This can cover
periods ranging from a few hours to a few days before operation. The
planning stage consists of the strategic and pretactical phases, both of
which occur before the planned trajectory becomes a reference
business trajectory (RBT). If potential conflicts between aircraft
trajectories can be automatically solved at the planning stage, then
controllers will spend less time on tactical conflict detection and
resolution. This will free them up to monitor adherence to the plan
and enable them to deal with more complex traffic situations [2].
However, this shift to reliable deconflicted planning is neither
straightforward nor direct. Nowadays, there are many uncertainties
when executing the plan. Therefore, the information available at the
strategic phase is not reliable for accurate planning purposes,
regardless of how effective the deconfliction models are. TBOs and, in
particular, 4-D trajectory management, are expected to provide the
necessary breakthrough in terms of improved predictability. The
SESAR Program [3] states that the transition to TBOs will
progressively improve the accuracy and availability of information for
both planning and execution purposes. This will be achieved by using
the target time over (TTO) and target time of arrival (TTA) as the main
drivers for improving predictability, and consequently efficiency and
capacity key performance areas (KPAs). Extended planning, ideally
up to the strategic phase, is expected to bring benefits to the overall
ATM system through a set of operational KPAs. This study presents a
method for deconflicted trajectory planning in the strategic phase.
Specifically, the study looks at the business trajectory lifecycle as
part of the TBO concept [4]. It investigates the shared business
trajectory (SBT), the reference business trajectory, and the transition
between them. The SBT is the published business trajectory that is
available for collaborative ATM planning purposes. The RBT is the
trajectory that the user agrees to fly and the air navigation service
providers (ANSPs) and airports agree to facilitate. The SESAR
concept does not specify the time at which the SBT changes to an
RBT, nor does it mention a specific trigger that will cause this to
happen. However, according to the current concept, the RBT does not
necessarily need to be conflict free: it is defined before departure and
is the best balance between the needs of the airspace user and the
constraints of the ATM system. Deconfliction is left for the tactical
phase. The aim of this study is to reconcile the balance, between needs
and constraints, with the requirement for conflict-free planning.
The overall aim of the study is to present a methodology to achieve
conflict-free planning or, at the very least, to minimize the number of
conflicts. It uses 1) a data-driven model with multiple examples of
conflict-resolution actions, and 2) a multiobjective optimization
process that takes the requirements of the different ATM actors into
consideration.
Received 2 June 2016; revision received 16 September 2016; accepted for
publication 19 September 2016; published online 20 December 2016.
Copyright © 2016 by Technical University of Madrid (UPM) and CRIDA.
Published by the American Institute of Aeronautics and Astronautics, Inc.,
with permission. All requests for copying and permission to reprint should be
submitted to CCC at www.copyright.com; employ the ISSN 0731-5090
(print) or 1533-3884 (online) to initiate your request. See also AIAA Rights
and Permissions www.aiaa.org/randp.
*Ph.D. Student, School of Aeronautical and Space Engineering, Ramiro de
Maeztu 7; e.calvo@alumnos.upm.es.
†
Professor, School of Aeronautical and Space Engineering, Ramiro de
Maeztu 7; l.perez@upm.es.
‡
Principal R&D Engineer, Av. de Aragón 402; jmcordero@e-crida.
enaire.es.
§
Professor, School of Aeronautical and Space Engineering, Ramiro de
Maeztu 7; rosamaria.arnaldo@upm.es.
615
J
OURNAL OF
G
UIDANCE
, C
ONTROL
,
AND
D
YNAMICS
Vol. 40, No. 3, March 2017
Downloaded by UNIV. OF ARIZONA on March 14, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.G000691
The data-driven model is not based on a static protocol method.
Rather, it uses a conflict-resolution dataset, thereby dynamically
capturing the behavior of air traffic controllers in conflict situations.
In this way, it deals with the issues raised in [5] regarding ATM
automation.
The methodology demonstrates the feasibility of using the
improved trajectories that the TBO will provide in dynamic
deconflicted flight planning. It presents a data-driven approach that
relies on the use of flight planning and surveillance information only.
This information is available to all air navigation service providers.
The study deals with the transition phase from a SBT to a RBT. The
solution should ideally produce a more refined and collaborative
planning process. To provide a plan that takes the interests of all
stakeholders into consideration, the study includes a multiobjective
optimization process. It balances competing demands such as
reducing fuel consumption (of interest to airlines) and sticking with
traditional air traffic control (ATC) solutions for resolving conflicts.
The two aspects, the data-driven model and the optimization
process, provide a new approach to conflict-free strategic planning as
compared with the present state of the art [6
–18]. This paper 1) sets
out the methodology used to generate the conflict-resolution model,
2) presents a multiobjective optimization to provide the optimal
conflict-free plan, and 3) presents the results.
The research presented here is a proof of concept for strategic
conflict-free planning. The algorithms used are at the script prototype
stage and are developed and tested using MATLAB (matrix laboratory),
which is a multiparadigm numerical computing environment and
fourth-generation programming language.
The approach presented in this paper will be of most interest to the
operational staff in charge of strategic planning before operation. The
window is from days to hours before operation. This approach is not
intended for use by air traffic controllers. The aim is to develop an
operational scenario that is useful for monitoring rather than for
detailed conflict detection and resolution.
The proposed methodology is compatible with the continuous
layered planning scheme defined in the ATM master plan, as it can be
applied to local, subregional, or network environments.
II.
Methodology
This section describes the methodology followed in this study,
specifically focusing on the data and architecture used.
A.
Data
The goal of a data-driven model is to extract information from a
dataset and transform it into an understandable structure for further
use. Therefore, the quality of the data source is the crucial factor.
Marzuoli et al. [19] presented an air traffic flow model based on a
data-driven approach, whereas Salaun et al. [20] presented aircraft
proximity maps based on data-driven flow modeling.
This study uses real operational data provided by ENAIRE, which
is the Spanish air navigation service provider, in coordination with
EUROCONTROL, which is the European network manager. The
specific operational information used in this study is as follows:
1) The study uses flight plan (FP) information from the flight plan
information management system, which is a subsystem of the
Spanish ATC platform. This contains information on every flight plan
currently in flight or scheduled to fly. All the changes and cancellations
that affect flight plans are constantly updated and registered in the
system.
2) Radar tracks (RTs) are information from secondary radars that
are sent to the operational air traffic control platform. They include
the flight call sign, altitude, speed, position, direction, and time. The
information is updated every 5 s.
In the study, we looked at more than 300,000 flights. This covered
a period of 72 days with more than 4200 flights per day. The radar
tracks came from the continental Spain flight information region
(FIR). Figure 1 shows Spain and the surrounding area divided into
FIRs. The borders of the continental Spain FIR are also indicated.
The vast amount of operational data means that any potential bias
due to a particular individual, sector, weather, or traffic pattern is
avoided. The gap between the current flight planning information and
the expected TBO SBT/RBT information will be bridged by having
readily available accurate data [3]. Note that availability is defined as
having the necessary information at the precise moment.
B.
Architecture
The objective is to obtain optimal conflict-free planning that
integrates air traffic controller knowhow. Therefore, the study is
divided into three processes (Fig. 2). The first process generates the
conflict-resolution model using a data-driven approach. In the second
process, the conflict-resolution model is used to resolve any conflicts.
And, finally, the third process looks at the different alternatives to
identify the optimal solution.
The conflict-resolution model, generated in the first phase, has a
large database containing around 18,000 conflicts over the 72-day
period. These conflicts are obtained from all of the updates to the
flight plans in the tactical phase. It also contains the tactical resolution
of the conflicts, calculated by measuring adherence to the active flight
plan by using radar track data. This process is explained in detail in
Sec. II.C.
The model considers vertical and time-based actions to resolve any
conflicts. We have not contemplated horizontal actions because
doing so would require a
“vectoring” strategy, which would entail the
use of new navigation points for rerouting. As such, horizontal
conflict-resolution actions are inherently tactical, and therefore not
suitable in a strategic planning methodology.
In the second process, we detected all the conflicts in the strategic
phase of the flight plans over 48 randomly selected days. The model
from the first process was applied to these conflicts, resulting in a set
of potential resolutions for each conflict. This process is explained in
detail in Sec. II.D.
The alternative resolutions were assessed in the final process of
multiobjective optimization, giving an optimal conflict-free plan.
This process is detailed in Sec. II.E.
C.
Development of a Conflict-Resolution Model Using a Data-Driven
Methodology
Several methods have been proposed to automate air traffic
conflict detection and resolution. Kuchar and Yang [6] compiled a
survey of 68 different modeling methods. Xiangmin et al. [7]
presented a strategic flight conflict-avoidance approach based on a
memetic algorithm. Valenzuela and Rivas [8] offered a method for
conflict detection and resolution based on the use of predefined
trajectory patterns. Matsuno el al. [9] proposed a stochastic near-
optimal control method for determining aircraft conflict-resolution
trajectories in the presence of uncertainty in real time. Durand and
Barnier [10] experimented with an algorithm derived from robotics to
provide arguments for centralized air traffic control.
Optimal conflict-resolution approaches have also been widely
studied, focusing on different objectives. Frazzoli et al. [11] looked at
the resolution of conflicts involving many aircraft and presented a
methodology whereby each aircraft proposed its desired heading
while a centralized authority resolved any conflict that might arise
between aircraft. Clements [12] investigated the problem of optimal
control formulation and incorporated a minimum-time deviation
objective. Raghunathan et al. [13] looked at dynamic optimization
Fig. 1
Continental Spain FIR.
616
CALVO-FERNÁNDEZ ET AL.
Downloaded by UNIV. OF ARIZONA on March 14, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.G000691
strategies for three-dimensional conflict resolution of multiple
aircraft using numerical trajectory optimization methods. Hu et al.
[14] analyzed conflict-free maneuvers to identify the one that
minimized a certain energy cost function. Green and Grace [15]
described the concept for a tool to help air traffic controllers plan
efficient spacing strategies, thereby resulting in reduced workloads
and fuel consumption.
In different studies, the knowledge of air traffic controllers is taken
into consideration. Isaacson and Robinson [2] presented a knowledge-
based conflict-resolution process that allowed predicted conflicts to be
resolved in a way that was consistent with ATC common practice.
Flicker and Fricke [16] used the knowledge of air traffic controllers to
increase acceptance of the conflict-resolution system.
Conflict-free planning is also considered in other studies. Huang
and Tomlin [17] generated conflict-free plans while minimizing
either total aircraft transit time through the sector or deviation from
the nominal flight plan. Yokoyama [18] proposed a decentralized
model predictive control algorithm for planning three-dimensional
conflict-free trajectories for multiple aircraft.
In this study, we analyzed around 18,000 operational conflicts. The
aim was to produce a model based on the actions taken by air traffic
controllers in the tactical phase, specifically by measuring adherence
to the active flight plan by using radar track data.
This model is given by the function
Y FX
(1)
where X is the input vector that characterizes the conflict (in
particular, the relationship between the aircraft trajectories); Y is
the conflict-resolution matrix that contains the solutions proposed by
the model; and FX denotes the nonlinear map that describes the
dependence between the conflict and the resolution.
1.
Data Preparation
The dataset collected during a single scenario is represented as
follows:
fX
i
; Y
i
g
C
i1
(2)
where C is the number of conflicts; X
i
is the input variables; and Y
i
is
the output variables.
The historical database used to develop the model consists of a
collection of conflicts: in other words, a set of sets
ffX
i
; Y
i
g
C
i1
g
S
dt1
(3)
where dt is the time interval of the scenario; and S denotes the total
number of scenarios.
Solomatine and Ostfeld [21] highlighted the importance of data
preparation in any modeling exercise: in particular, the choice of time
intervals used to create the scenarios. A number of studies in other
fields [22
–24] dealt with the selection of optimal data time intervals.
Scenarios were generated at time intervals of 5 min, giving 288
scenarios per day for each day of operation. This time interval
approach was used because the flight plans of a specific flight are
updated frequently; therefore, conflicts must be detected within a
short period of time. The interval was chosen based on a study [25] in
which a conflict-detection model was analyzed using RAMS Plus,
which is an ATM fast-time simulation tool. The study reassessed
conflict samples every 5 min. As shall be seen in the following, in the
Results section (Sec. III), when we looked at the conflicts detected
over a sample period of one day, we found that only 1.10% of the
conflicts detected using 1 min intervals were not detected when 5 min
intervals were used. Therefore, we considered 5 min intervals to be
apt for the study.
The air traffic controller uses flight plan information to predict and
monitor air traffic in the tactical phase, making the necessary changes
to each trajectory in order to ensure aircraft separation. Thus, in each
scenario, the necessary data are 1) the latest updated flight plans, used
to detect and categorize conflicts; and 2) radar tracks, to identify the
actions taken by controllers to resolve the conflicts.
2.
Conflict Detection Based on Historical Data
A conflict is defined as a situation in which two aircraft
simultaneously lose both horizontal and vertical separation minima.
The current en route standard criteria are a minimum horizontal
separation of 5 n miles and a minimum vertical separation of 1000 ft.
It is necessary to project the trajectory of a flight into the future in
order to predict whether or not there will be a conflict. This projection
is based on the available information: in other words, the current
position and the flight plan. As with all predictions, there is a degree
of uncertainty.
There are many studies dealing with trajectory prediction in ATM.
In this study, the position of the aircraft will be projected into the
future along a single trajectory. The position will be extrapolated
using current flight plan information, known to the controller, such as
the waypoint and estimated times. To emulate a TBO environment,
this information must be available at the planning stage as part of the
business trajectory management. This information would typically
be in the form of a set of TTO/TTAs, which can be used to estimate the
speeds if this information has not been included. An important
assumption is that TTO/TTA, or TTA, and speed profile data are
available.
The 4-D trajectory (synthetic flight plan) is generated by sampling
a point every 5 s. This interval was shown in [26] by Durand and
Gotteland to be small enough to detect every conflict. In accordance
Chia sẻ với bạn bè của bạn: |