1
Modeling and Optimization of the Capacity
Allocation Problem with Constraints
Abdellah IDRISSI
Abstract— The Constraint Satisfaction Problem (CSP) is
proven more and more promising to model and to solve a
great number of real problems. A lot of approaches using
constraint reasoning have proposed to solve search problems.
In this paper, we propose a modeling of Capacity Allocation
Problem of an airport (CAP) and of its fixes (F CAP ) in form
of a Constraint Satisfaction Problem. This modeling gives a CSP
problem called F CSP AC. Then, in order to solve the F CSP AC,
we propose an Optimization model called COF CAP . We present
some experimental results through an example followed by
its optimized resolution. Finally, we show how to control the
solutions of an F CSP AC to assist the airport managers to face
unforeseen events and present some future work.
Index Terms— Artificial Intelligence, Constraint Optimization,
Airport Capacity Allocation Problem, Modeling, Solving.
I. I NTRODUCTION
APACITY allocation problems are the core of many realworld scheduling and planning problems. The first steps
to solve a capacity allocation problem consist in formulating
and modeling it. Modeling is one of the central themes of
Artificial Intelligence (AI). A good model generally facilitates
the problem-solving.
However, modeling is difficult and no general guidelines exist on which factors must be taken into account in a modeling
process. Therefore, developing a good model largely remains
an art, depending on experiences and tastes. Nevertheless,
for a given problem, different modeling techniques can be
studied. Among these techniques one can find the constraints
satisfaction problems (CSP) [22].
Formally, a CSP can be defined by the triplet (X, D, C),
where X = {X1 , ..., Xn }, is the set of n variables; D =
{D1 , ..., Dn }, is the set of n domains of values; Di is the
domain of values of the variable Xi and C = {C1 , ..., Ce }, is
the set of e constraints of the problem, specifying compatible
values or excluding incompatible values between variables. To
solve a CSP consists in assigning values to variables in order
to satisfy all the constraints.
Various real problems can be represented in form of a CSP.
In this paper, we model the Capacity Allocation Problem of
an airport and its fixes (F CAP ) in a CSP and apply CSP
techniques to solve it. The CAP of an airport is a key problem
in regulating aerial traffic. This problem will become more and
more important because aerial traffic will greatly increase in
the next years.
C
LaRIA, FRE 2733, Université de Picardie.
33 rue St Leu, 80039 Amiens, France.
Emails: [email protected] , [email protected]
The second author is partially supported by National 973 Program of China
under Grant No. 2005CB321900
and
Chu Min LI
Thus, it is necessary to solve this problem, especially in
case of congestion. The congestion occurs at an airport and
its fixes when the request of the traffic exceeds the available
capacity. The airport managers control the traffic, and solve
these problems of congestion by delaying some flights, so that
the flow on the airport is regulated, and does not exceed the
available capacity. His objective is to minimize the total delay
of the planes.
The prime objective of the present study is to model the
capacity allocation problem of an airport and its fixes in
form of a CSP. For this purpose, it is enough to identify the
variables, the domains of values which these variables can
take, and the constraints of the problem.
The second objective is to solve this problem by using
the resolution algorithms designed specifically for the CSP, in
order to assist the managers of an airport to regulate the arrival
and departure demands, and to use efficiently the available
capacities of a terminal. These CSP resolution algorithms
include the Branch and Bound algorithm and its alternatives,
as presented in [19], [20], [21], [24] and [26].
The third objective is to study the complexity of this
capacity allocation problem and to show that it is an NPcomplete problem.
We will treat the first two objectives in this paper, the third
being the subject of the next studies.
This paper is organized as follows. Section II defines
the capacity allocation problem. In section III we present a
modeling of this problem in form of a CSP which we called
F CSP AC, and propose an optimization model to solve it in
section IV. We give an algorithm in section V and expose in
section VI an application example of an F CSP AC. We show
in section VII how to assist the managers to regulate traffic and
unforeseen events. We conclude in section VIII and discuss
some perspectives of this work.
II. C APACITY A LLOCATION P ROBLEM
A. Presentation
An operational scheme of a single airport system that
reflects the arrival and departure processes at the airport is
shown in figure 1. The system comprises arrival ways called
arrival fixes ‘AF’, departure ways called departure fixes ‘DF’,
and a runway system. There are two separate sets of arrival and
departure fixes located in the near-terminal airspace area (5070 km off the airport), so that the arrival fixes serve only arrival
flow, and the departure fixes serve only departure flow. The
runway system on the ground serves both arrival and departure
flows.
The arrival flights are assigned to specific arrival fixes, and,
before landing, they should go through the fixes. After leaving
Fig. 1.
Example of an airport system
runways, the arriving flights follow the taxiway to the gates
at the terminal. The departing flights are also assigned to the
specific fixes. They are directed to the runways, and go through
the departure fixes after leaving runways.
The arrival queues are formed before the fixes (see figure
1). This means that the flights which are already in the fixes,
must be accepted at the runways. If there is an arrival queue,
a certain number of flights should be delayed in the air.
The departure queue is formed before the runway system,
and flights can be delayed either at their gates or on the
taxiway.
The capacity of arrival and departure fixes is the maximum
number of flights that can cross a fix in a 15 minutes interval
(or other interval). These capacities determine the operational
constraints in the near-terminal airspace. The operational limits
on the ground (runways) are characterized by arrival capacity
and departure capacity. These capacities are generally variable
and interdependent.
The traffic demands for an airport and for the fixes of
this airport are given by the predicted number of arriving
and departing flights per each 15 minutes time interval of a
considered period (see table I).
Definition 1. Airport capacity is defined respectively as the
maximum number of operations (arrivals and departures) that
can be performed within a fixed time interval (e.g., 15 minutes
or other interval) at a given airport under given conditions such
as runway configuration and weather conditions.
Definition 2. Fix capacity is defined as the maximum
number of flights that can cross a fix in a 15 minutes interval
(or other interval) under given conditions such as weather
conditions, etc. These capacities determine the operational
TABLE I
E XAMPLE OF PREDICTED NUMBER OF ARRIVING AND DEPARTING
FLIGHTS PER EACH 15 MINUTES TIME INTERVAL OF A 3 HOURS PERIOD .
constraints in the near-terminal airspace.
Definition 3. Airport capacity allocation problem consists
in determining a balance, between arrivals and departures,
to minimize an adapted objective function for a given time
period.
The capacity of the airports and the capacity of its fixes
become more and more limited compared to the demands and
present real problems for the aerial transportation system. The
optimal allocation of the existing capacity of an airport and
of its fixes during the congestion periods is very important to
maximize the departures and to minimize the delay.
B. Related Work
Aerial traffic is continuously increasing and its regulation
becomes more and more difficult, given the limited capacities
both in ground (airports) and air (aerial sectors). Various
solutions can be used to attack this difficulty: building new airports, physically extending existing runway systems, applying
new technologies to increase airport capacity, and optimizing
utilization of existing airport capacity [4]. See [13] for some
possible measures for increasing airport capacity.
On the other hand, aerial traffic regulation should be optimized and automated, especially when treating unforseen
events. In [15], we proposed a modeling and a resolution of the
aerial conflicts problems by techniques of constraints network
(CSP). This modeling and automatic resolution can be possibly
introduced into project (FREER) [7] developed by Eurocontrol
following the concept “Free Flight” [8].
At an airport, the air traffic managers generally know
the range of arrival/departure capacity tradeoff available for
each runway configuration and use the tradeoff for strategic
planning of arrival and departure traffic. They control and
modify the operational arrival and departure limits during the
periods of congestion to better serve the traffic demand and
to minimize the delay.
However, currently they proceed intuitively because of the
lack of automated decision support to determine the best strategies for allocation of airport arrival and departure capacities.
An experienced specialist can find the best solutions but, in
more complex situations, this task can become impossible. The
theoretical aspects of the problem are addressed in [9], [10].
In [16], we proposed a modeling of the airport capacity
allocation problems by techniques of constraints network
(CSP). We associated an optimization model to it. But, we
had not taken into account the capacities of different arrival
and departure fixes.
In order to optimize and automate the aerial traffic regulation, we propose in this paper a modeling of capacity
allocation problem of an airport and its fixes using constraint
network (CSP). We also propose an optimization procedure
able to give the best allocation of airport (and/or fix) capacity
between arrivals and departures under given operational conditions at the fixes to satisfy the predicted traffic demand for
a time period and to minimize the delay.
We specify this model in terms of a Constraint Satisfaction
Problem (CSP) in the next section.
III. F CSP AC: A CSP MODEL FOR THE A LLOCATION OF
C APACITY AT THE FIXES
A. Constraint Satisfaction Problems
A CSP is defined by the triplet (X, D, C), with X =
{X1 , ..., Xn }, is a set of n variables; D = {D1 , ..., Dn }, is a
set of n domains. Di is the domain of values associated with
variable Xi and C = {C1 , ..., Ce }, is a set of e constraints
of the problem. Solving a CSP consists in assigning values
to variables in order to satisfy all the constraints. In general,
the CSPs are solved by traditional methods combining a
mechanism of search with Backtrack (BT) and a mechanism
of reinforcement of consistencies with each node of search.
We propose to solve the capacity allocation problem of the
airport and of its fixes using CSP resolution algorithms, including the Branch and Bound and its alternatives, as presented in
[19], [20], [21], [24] and [26].
B. Problem Formulation
Figure 2 shows a simplified scheme of an airport constituted
with only one arrival fix AF, one departure fix DF and one
terminal T serving both arrival and departure demands during
a time interval ‘i’ (‘i’ is a period of 15 minutes). The arrival
Fig. 2.
Simplified Scheme of an airport
demands are at the point ‘A’. The departure demands are at
the point ‘D’. The point ‘T’ represents the Terminal.
The problem consists in satisfying the arrival and departure
demands, within the limit of the arrival fix capacity (Cafi ), the
limit of the departure fix capacity (Cdfi ) and the limit of the
total capacity of the terminal (CTi ) during the time interval
‘i’.
In the sequel, we consider that there are ‘j’ arrival fixes
AFjdfi with j = 1, ..., naf and ‘k’ departure fixes DFkdfi
with k = 1, ..., ndf . Each fix ‘j’ or ‘k’ has its own capacity
j
k
noted Caf
or Cdf
during the same interval of time ‘i’.
i
i
It should be noted that the sum of the fixes capacities
Cai + Cdi of an airport is in general lowerPor equal than
naf
j
the capacity of the terminal CTi where Cai = j=1
Caf
and
i
Pndf k
Cdi = k=1 Cdfi .
Cai + Cdi ≤ CTi
(1)
CTi is the sum of Pvi and of Poci .
CTi = Pvi + Poci
(2)
Pvi is the number of empty places at the terminal and Poci is
the number of occupied places at the terminal.
If Xjaf ei represents the number of arrival flights accepted
at fix j (j = 1, ..., naf ) and Ykdf ei the number of departure
flights effectively leaving through fix k (k = 1, ..., ndf ) during
time interval ‘i’, then the number of empty places for the time
interval ‘i+1’ becomes:
naf
ndf
X
X
Ykdf ei −
Xjaf ei
(3)
Pvi+1 = Pvi +
k=1
j=1
In the same way, the number of occupied places for the time
interval ‘i+1’ becomes:
naf
ndf
X
X
Ykdf ei
Poci+1 = Poci +
Xjaf ei −
(4)
j=1
k=1
For the sake of simplification, let us admit in the sequel
that CTi is a constant independent of time. We will note it
CT such as:
CT = Pvi + Poci = Pvi+1 + Poci+1
(5)
We propose hereafter a modeling of this problem in form
of a constraint satisfaction problem (CSP).
C. Problem Modeling
We propose to model this problem of the arrivals/departures
at an airport and at its fixes (figure 2) in form of a CSP
F CSP AC = (Xi , Di , Ci ) during a time interval ‘i’, where:
• Xi = {Xai , Xaei , Ydi , Ydei , Cai , Cdi , Qai , Qdi , X1af ,
i
..., Xnafi , X1af ei , ..., Xnaf ei , Y1dfi , ..., Yndfi , Y1df ei ,
..., Yndf ei , C1afi , ..., Cnafi , C1dfi , ..., Cndfi , Q1afi , ...,
Qnafi , Q1dfi , ..., Qndfi , Xtai , Ytdi , Pvi , Poci } is the set
of variables such as :
– Xai is the number of arrival demands at the point
‘A’ (figure 2).
– Xaei is the number of effective arrivals at ‘A’.
– Ydi is the number of departure demands at the point
‘D’ (figure 2).
– Ydei is the number of effective departures at ‘D’.
– Cai is the arrival capacity at the point ‘A’.
– Cdi is the departure capacity at the point ‘D’. Cai
and Cdi are dictated by the airport managers.
– Qai is the number of arrivals delayed at ‘A’.
– Qdi is the number of departures delayed at ‘D’.
– Xjafi is the number of arrival demands at the fix ‘j’
(figure 1) with j = 1, ..., naf .
– Xjaf ei is the number of effective arrivals at the fix
‘j’ with j = 1, ..., naf .
– Ykdfi is the number of departure demands at the fix
‘k’ (figure 1) with k = 1, ..., ndf .
– Ykdf ei is the number of effective departures at the fix
‘k’ with k = 1, ..., ndf .
– Qjafi is the number of arrivals delayed at the fix ‘j’
with j = 1, ..., naf .
– Qkdfi is the number of departures delayed at the fix
‘k’ with k = 1, ..., ndf .
– Cjafi is the arrival capacity at the fix ‘j’ with j =
1, ..., naf .
– Ckdfi is the departure capacity at the fix ‘k’ with
k = 1, ..., ndf .
– Xtai is the total number of arrival demands for the
time interval ‘i’. Note that for each value of ‘i’, we
add the delay Qai−1 recorded in the previous interval
‘i-1’ to the predicted arrival demands Xai . Then Xai
becomes Xai + Qai−1 which we noted Xtai such as
Qa0 = 0.
– Ytdi is the total number of departure demands for
the time interval ‘i’. For each value of ‘i’, we add
the delay Qdi−1 recorded in the previous interval ‘i1’ to the predicted departure demands Ydi . Then Ydi
becomes Ydi + Qdi−1 which we noted Ytdi such as
Qd0 = 0.
– Pvi and Poci are respectively the set of empty and
occupied places of the terminal at the point ‘T’ of
figure 2 during the same time interval ‘i’.
• Di = {DXa , DXae , DYd , DYde , DCa , DCd , DQa
i
i
i
i
i
i
i
, DQdi , DX1af , ..., DXnaf , DX1af e , ..., DXnaf e ,
i
i
i
i
DY1df , ..., DYndf , DY1df e , ..., DYndf e , DC1af , ...,
i
i
i
i
i
•
DCnaf , DC1df , ..., DCndf , DQ1af , ..., DQnaf ,
i
i
i
i
i
DQ1df , ..., DQndf , DXtd , DYtd , DPvi , DPoci } is the
i
i
i
i
set of variable domains. We consider that all the domains
of these variables are equal and that Di represents a
common domain to all these variables. In other words,
Di = {0, 1, 2, ..., CT }.
Note that the values of arrival and departure capacities are
interdependent and are dictated by the airport managers in
form of couples (Cai , Cdi ). They can recommend to use
for example, at a given period, the set {(18, 29), (24, 24),
(26, 19), (28, 15), (17, 30), (20, 27)}. In this example,
the first couple means that if Cai = 18 then Cdi = 29
and vice versa. The meaning of other couples is similar.
Ci is the set of constraints of the problem. We can
formulate them as follows :
0 ≤ Pvi ; Poci
≤ CT
(6)
0 ≤ Pvi + Poci ≤ CT
(7)
0 ≤ Pvi+1 + Poci+1 ≤ CT
(8)
0≤
naf
X
Xjaf ei ≤ Pvi
(9)
Ykdf ei ≤ Poci
(10)
j=1
0≤
ndf
X
k=1
0 ≤ Xjaf ei ≤ Xjafi , ∀j ∈ {1, ..., naf }
(11)
0 ≤ Ykdf ei ≤ Ykdfi , ∀k ∈ {1, ..., ndf }
(12)
0≤
naf
X
Xjaf ei +
j=1
ndf
X
Ykdf ei ≤ CT
(13)
k=1
Pvi+1 = Pvi +
ndf
X
Ykdf ei −
Poci+1 = Poci +
Xjaf ei −
j=1
Xai =
Xjaf ei
(14)
j=1
k=1
naf
X
naf
X
ndf
X
Ykdf ei
(15)
k=1
naf
X
Xjafi
(16)
Xjaf ei
(17)
Ykdfi
(18)
Ykdf ei
(19)
j=1
Xaei =
naf
X
j=1
Ydi =
ndf
X
k=1
Ydei =
ndf
X
k=1
C ai =
naf
X
Cjafi
(20)
Ckdfi
(21)
j=1
Cd i =
ndf
X
k=1
Xaei ≤ Cai
(22)
Ydei ≤ Cdi
(23)
Xaei + Ydei ≤ Xai + Ydi
(24)
Xjafi ≤ Cjafi
(25)
Ykdfi ≤ Ckdfi
(26)
Qai =
naf
X
Qjafi
(27)
Qkdfi
(28)
j=1
Qdi =
ndf
X
k=1
Xtai =
naf
X
Xtjaf
i
j=1
Xtjaf = Xjafi + Qjafi−1 , ∀j ∈ {1, ..., naf }
i
Ytdi =
ndf
X
k=1
Ytkdf
i
Ytkdf = Ykdfi + Qkdfi−1 , ∀k ∈ {1, ..., ndf }
i
Qjafi = Xjafi − Xjaf ei
(29)
(30)
(31)
D. Problem Solving
Given values of (Xai , Ydi ) (arrival and departure demands
for a time interval ‘i’) and a number of possible values of (Cai ,
Cdi ), the problem consists in assigning a value to each variable
by respecting the constraints described above. The problem
FCSPAC is a problem of decision. It is also an optimization
problem.
Now our constraint network (CSP) is well identified. In particular the variables, the domains of values of these variables
and the constraints of the problem are explicitly defined. It
remains to apply one of CSP reasoning procedure (Branch and
Bound, Local Search, Backtracking, ...) to solve the capacity
allocation problem. We think in particular to use the Branch
and Bound algorithm and its alternatives, as presented in [19],
[20], [21] [24] and [26].
Let us note that in the case of congestion during
Pnaaftime interval ‘i’ the sum of numbers of effective arrivals j=1
(Xjaf ei )
Pndf
and of effective departures
(Y
)
is
lower
than the
ei
k=1 kdfP
naf
(X
)
and of
sum of numbers of arrival demands
jafi
j=1
Pndf
departure demands k=1
(Ykdfi ). In other words, only a part
of the demands, arrivals or departures, is been served (see
equation (24)).
When solving a CSP, a number of questions should be
asked. For example :
• Is there any solution for this problem ? In other words,
is the problem satisfiable ?
• How can we find a solution ?
• How can we find all solutions ?
• ...
(32)
For our F CSP AC, we are interested in the following
questions:
+ Qjafi−1 , ∀j ∈ {1, ..., naf } (33)
• Is it possible to serve all the requests without any delay:
Qkdfi = Ykdfi − Ykdf ei + Qkdfi−1 , ∀k ∈ {1, ..., ndf } (34)
N
X
Qjafi−1 , for j = 1, ..., naf (equation (33)) is the number
of arrival planes delayed at the fix j for j = 1, ..., naf during the time interval ‘i-1’ and Qkdfi−1 , for k = 1, ..., ndf
(equation (34)) is the number of departure planes delayed
at the fix k for k = 1, ..., ndf during the time interval ‘i-1’.
These two numbers should be added respectively to the
arrival and departure demands during the time interval
‘i’. So the total number of arrival demands during ‘i’
is the sum of demands effectively expressed Xjaf ei , for
j = 1, ..., naf during ‘i’ increased by the number of
planes Qjafi−1 , for j = 1, ..., naf which were waiting
in the queue due to the fact that they were not been
served during ‘i-1’. Similarly, the total departure demand
during ‘i’ is the sum of demands really expressed Ykdf ei
for k = 1, ..., ndf during ‘i’ increased by the number of
planes Qkdfi−1 at the fix k for k = 1, ..., ndf which were
waiting in the queue due to the fact that they were not
been served during the time interval ‘i-1’.
The planes which were waiting in the queue during the
time interval ‘i-1’ are served in priority during the time
interval ‘i’.
i=1
[
naf
X
j=1
Qjafi +
ndf
X
Qkdfi ] = 0 ?
k=1
Where Qjafi and Qkdfi are respectively the number of
arrival planes delayed at the fix j for j = 1, ..., naf and
departure planes delayed at the fix k for k = 1, ..., ndf ,
during the time interval ‘i’ and N is the number of
intervals constituting the given time period.
•
Is it possible to serve all the requests with the given delay
‘1’ :
naf
ndf
N
X
X
X
Qkdfi ] = 1 ?
[
Qjafi +
•
...
•
Is it possible to serve all the requests with the given delay
‘r’:
naf
ndf
N
X
X
X
[
Qjafi +
Qkdfi ] = r ?
i=1
i=1
j=1
j=1
k=1
k=1
•
Is it possible to serve all the requests without exceeding
a certain given delay ‘rmax ’:
N
X
[
i=1
naf
X
Qjafi +
j=1
ndf
X
Ntar
Qkdfi ] ≤ rmax ?
i=1
[
naf
X
j=1
naf
ndf
N X
X
X
Qkdfi ]
=
[
Qjafi +
i=1 j=1
k=1
Our objective is to minimize
the sum of the numbers of
PN
arrival planes delayed ( i=1 [Qjafi ]) ∀j ∈ {1, ..., naf } and to
minimize
the sum of the numbers of departure planes delayed
PN
( i=1 [Qkdfi ]) ∀k ∈ {1, ..., ndf } during a given time period
(15*N minutes). In other words, we seek to minimize:
N
X
is estimated as (Ntar ∗ 15) minutes. Indeed, over one period
constituted by N intervals of time one will have:
Qjafi +
ndf
X
Qkdfi ]
(35)
Qjafi = Xjafi − Xjaf ei + Qjafi−1 , ∀j ∈ {1, ..., nafi }
and
Qkdfi = Ykdfi − Ykdf ei + Qkdfi−1 , ∀k ∈ {1, ..., ndfi }
are respectively the number of planes arriving and leaving
which are delayed during a time interval ‘i=15 min’.
The function of penalty becomes:
k=1
ϕF =
subject to the constraints (6)-(34). We recall that:
N
X
i=1
[
naf
X
Qjafi +
j=1
ndf
X
Qkdfi ]
(38)
k=1
To optimize the function of penalty thus means to minimize
the function ϕF where :
Qjafi = Xjafi − Xjaf ei + Qjafi−1 , ∀j ∈ {1, ..., naf }
naf
ndf
N
X
X
X
ϕF =
[ (Xjafi − Xjaf ei ) +
(Ykdfi − Ykdf ei )] (39)
and
Qkdfi = Ykdfi − Ykdf ei + Qkdfi−1 , ∀k ∈ {1, ..., ndf }
i=1
We propose to place this problem within a general framework of an optimization under time limit constraint. Because
we estimate that in the event of breakdown or other, it is well
necessary to provide a solution even if it is not of very good
quality. This issue will be the subject of our next studies.
For the moment we propose below an optimization method
under constraints without taking in consideration the time limit
constraint.
IV. O PTIMIZATION M ODEL
When one assigns a capacity to a request consisting of
several arrival or departure flight demands, then if this request
is totally satisfied, one passes to the next one. If this request
is not totally satisfied then we consider that the difference
between the request and the available capacity is a penalty p
such as (D − C = p), where D is the number of demands
in the request, C is the capacity allowed to this request. The
problem is to reduce the sum of these penalties of all the fixes.
For that, we propose an optimization model which we called
COF CAP .
A COF CAP is an optimization problem under a set of
constraints. It consists to reduce to the minimum a function of
penalty ϕ under a set of constraints C. This function of penalty
ϕ is the sum of the penalties of the unsatisfied constraints. Let
us note p(c) the penalty of an unsatisfied constraint. Then the
problem COF CAP can be characterized by:
minimize [ϕ] where
X
ϕ=
p(c|c ∈ C and c unsatisf ied)
where
k=1
(36)
(37)
In addition, the penalty can be the total time of delay. It
can be also the total number of planes delayed (Ntar ) over
all the period considered. The total time of delay for this last
j=1
k=1
V. A LGORITHM
The resolution and the optimization of an F CSP AC problem proceed in several stages:
• Initially they consist of the distribution of the arrival
demands of an airport on its arrival fixes and of the
distribution of the departure demands on its departure
fixes.
• In the second time they consist to select the pertinent
couple of arrival and departure capacity (value) (Cai ,
Cdi ) among the various couples of capacities (dictated
by airport managers) to affect at the same time to arrival
and departure demands (variables) of the different fixes.
As mentioned above, this choice is selected according to
the heuristic which allows to respect all the constraints
and to minimize the number of delayed planes.
• In the third time they consist of the distribution of this
pertinent couple of arrival and departure capacity (value)
(Cai , Cdi ) found above respectively on the capacities of
arrival fixes C1afi , ..., Cnafi and on the capacities of
departure fixes C1dfi , ..., Cndfi .
• Finally they consist of the assignment of these capacities
respectively to the variables Xjaei for j=1, ..., nafi and
Ykdei for k=1, ..., ndfi so that all the constraints (6)-(34)
PN Pnaf
Pdaf
are satisfied and that i=1 [ j=1i Qjafi + k=1i Qkdfi ]
is minimized.
• The delays are the sum of differences between the
predicted demands for arrival and the arrival planes
effectively served added to the sum of differences between the predicted demands for departure and the planes
effectively leaved. In other words, the total number of
planes delayed is ϕF where:
ϕF =
naf
N X
X
i=1 j=1
(Xtjaf − Xtjaf e ) +
i
i
ndf
X
(Ytkdf − Ytkdf e )
k=1
i
i
We implemented this approach and compared it to the other
methods suggested in the literature (in particular [10] and
[11]). The results obtained for the various instances of the
problem show that our approach gives solutions at least as
good as those found by other techniques.
We present below an example of F CSP AC problem and
one of its optimal solutions satisfying constraints (6)-(34) and
minimizing
N
X
i=1
[
naf
X
j=1
Qjafi +
ndf
X
Qkdfi ]
k=1
VI. E XAMPLE OF AN F CSP AC PROBLEM FOLLOWED BY
ITS RESOLUTION AND ITS OPTIMIZATION
A. Presentation of an F CSP AC problem
Given a time period composed of a number of time intervals, the predicted arrival and departure demands in each of
these intervals, and a set of compatible repartitions of airport
capacity between arrival and departure flights, our problem
is to determine the real repartition of the demands among the
fixes, and the real repartition of arrival and departure capacities
among fixes, according to the actual demands of each fix.
We illustrate the resolution of this problem by an example
in this section.
We consider that there are 4 arrival fixes noted af1, af2, af3
and af4 and 4 departure fixes noted df1, df2, df3 and df4 at
the airport (figure 1). In this example, the period is 3 hours
(12h00-15h00) and the number of intervals N forming this
period is equal to 12.
The predicted arrival and departure demands are those given
in table I and the available capacities are values (domain)
given as an example in subsection C of section III (D =
{(18, 29), (24, 24), (26, 19), (28, 15), (17, 30), (20, 27)}).
In table II, column 0 shows the time intervals of 15
minutes. For each interval, columns 1 and 2 respectively show
the number of predicted arrival demands Xai and departure
demands Ydi of the airport.
These predicted demands will be distributed among the
different fixes. The arrival predicted demands of the airport
will be distributed among the arrival demands of its fixes
X1afi , X2afi , X3afi and X4afi as indicated respectively in
columns 3, 4, 5 and 6 for arrival fixes af1, af2, af3 and af4
(see table II).
In the same way, the departure predicted demands of the
airport will be distributed among the departure demands of its
fixes Y1dfi , Y2dfi , Y3dfi and Y4dfi as indicated respectively in
columns 7, 8, 9 and 10 for departure fixes df1, df2, df3 and
df4 (table II).
In table III, columns 11, 12, 13 and 14 present respectively
arrival flows served (or effective) X1af ei , X2af ei , X3af ei and
X4af ei at the fixes af1, af2, af3 and af4. Columns 15, 16, 17
and 18 of the same table III present respectively departure
flows Y1df ei , Y2df ei , Y3df ei and Y4df ei at the fixes df1, df2, df3
and df4.
In table IV, columns 21, 22, 23 and 24 present respectively
the arrival demands which were not served (queues or delays)
Q1afi , Q2afi , Q3afi and Q4afi at the arrival fixes af1, af2,
TABLE II
A RRIVAL AND DEPARTURE DEMANDS FORMULATED AT AN AIRPORT
( COLUMNS 1 AND 2). T HESE ARRIVAL AND DEPARTURE DEMANDS
( COLUMNS 1 AND 2) OF THE AIRPORT ARE RESPECTIVELY DISTRIBUTED
ON ITS 4 ARRIVAL FIXES ( COLUMNS 3,...,6) AND ON ITS 4 DEPARTURE
FIXES ( COLUMNS 7,...,10).
af3 and af4. Columns 25, 26, 27 and 28 of the same table IV
present respectively the departure demands which were not
served Q1dfi , Q2dfi , Q3dfi and Q4dfi at the departure fixes
df1, df2, df3 and df4. The arrival/departure demands which
were not served are queues which must be taken into account
during the next time interval.
B. Example of the F CSP AC problem resolution
The problem consists in distributing the values of column
1 on columns 3, 4, 5 and 6 of arrival fixes af1, af2, af3 and
af4 (figure 1) and in distributing the values of column 2 on
columns 7, 8, 9 and 10 of departure fixes df1, df2, df3 and
df4. This distribution is arbitrary and is more or less equitable.
Then, it acts of the distribution of the arrival capacities of
the terminal on the capacities of arrival fixes which are in
columns 11, 12, 13 and 14, as well as the distribution of
the departure capacities of the terminal on the capacities of
departure fixes which are in columns 15, 16, 17 and 18.
But before that, let us notice, for each line, the choice of
the capacity (value) among the various capacities to affect to
TABLE III
TABLE IV
A RRIVAL AND DEPARTURE CAPACITIES DICTATED BY THE MANAGERS OF
AN AIRPORT ( MENTIONED HERE IN COLUMNS 19 AND 20). F OR EACH
LINE , ONE COUPLE OF CAPACITY FORMED RESPECTIVELY BY ARRIVAL
R EAL DELAYS RECORDED ON THE DIFFERENT FIXES OF THE AIRPORT.
AND DEPARTURE CAPACITY IS SELECTED AND DISTRIBUTED
4 ARRIVAL FIXES ( COLUMNS 11,...,14) AND ON ITS
4 DEPARTURE FIXES ( COLUMNS 15,...,18).
RESPECTIVELY ON ITS
arrival and departure demands (variables) of the different fixes.
This choice is selected according to the heuristic which allows
to respect all the constraints and to minimize the number of
delayed planes. The resolution is done in an incremental way
and follows the law of line per line.
In columns 21, 22, 23 and 24 of table IV we recover at each
time interval i the number of the planes which had required
to arrive at the one of the arrival fixes af1, af2, af3 and af4
but which has not been served. These demands which has not
been served will respectively constitute arrival queues for these
same fixes for the next time interval i+1.
In the same way, in columns 25, 26, 27 and 28 of table IV
we recover at each time interval i the number of the planes
which had required to leave at one of the departure fixes
df1, df2, df3 and df4 but which has not been served. These
demands which has not been served will respectively constitute
departure queues for these same fixes for the next time interval
i+1.
The objective of this study is to reduce at the minimum the
sum of the numbers of delayed planes which are in columns
21, 22, 23 and 24 for arrival fixes and in columns 25, 26, 27
and 28 for departure fixes.
In a general way, we seek to reduce the total sum of all
these planes delayed.
C. Example of the F CSP AC problem optimization
For the example of the problem presented above, the function of penalty ϕF which is the sum of the columns 21, 22,
23 and 24 for arrival fixes and of the columns 25, 26, 27 and
28 for the departure fixes (table IV) will be written:
ϕF =
12 X
4
4
X
X
[
Qjafi +
Qkdfi ]
i=1 j=1
(40)
k=1
That is to say :
ϕF =
naf
ndf
12 X
X
X
(Ykdfi − Ykdf ei )] (41)
[ (Xjafi − Xjaf ei ) +
i=1 j=1
k=1
number of the arrival planes delayed is:
PThe
12 Pnaf =4
[
j=1 (Xjafi − Xjaf ei )] = 37 + 35 + 30 + 41 = 143
i=1
(columns 21, 22, 23 and 24), and
The number of the departure planes delayed is:
P12 Pndf =4
i=1 [
k=1 (Ykdfi − Ykdf ei )] = 18 + 21 + 17 + 21 = 77
(columns 25, 26, 27 and 28).
The total number of planes delayed is 143 + 77 = 220.
We find the same results just like if we had considered the
existence of only one arrival fix and only one departure fix
and one terminal (figure 2) as we proposed it in [17].
The following formula allows us to solve partly this problem
of congestion and to ensure a certain balance between arrivals
and departures.
N
X
[αi Poci −γi Ydei +βi Pvi −δi Xaei ] ≤
i=1
VII. C ONTROL AND REGULATION OF CAPACITIES
We saw in inequation (7) that
0 ≤ Pvi + Poci ≤ CTi .
Let us suppose that one wishes to reduce or increase the
capacities of an airport and of its fixes for a some reason. In
otherwise, one wishes to control and regulate these capacities.
Then in this case it is necessary to introduce some control
parameters. The formula above becomes :
αi Poci + βi Pvi = λi CTi
(42)
The parameters αi , βi and λi control the capacities. They
depend on the wishes of regulation expressed by the airport
managers according to certain conditions weather, priorities,
authorities, rush hours, etc. They have values between 0 and
1 such that αi + βi = 1.
For example, if αi = 0.5, βi = 0.5 and λi = 0.25, we
will have 0.5Poci + 0.5Pvi = 0.25CTi . This formula becomes
Poci + Pvi = 0.5CTi . The total capacity is not entirely available: meaning that only half of the airport capacity is available.
This restriction can occur because of certain conditions like
weather, authorities, etc. It is thus necessary to adapt these
new capacities for the arrival and departure demands.
We are thus brought to introduce these parameters into all
the formulas which we put until now. More precisely, we will
have the following formulas:
αi Xaei ≤ λi Cai ≤ βi Pvi
(43)
αi0 Ydei ≤ λ0i Cdi ≤ βi0 Poci
(44)
Such that Cai and Cdi are the total capacities respectively of
arrival and departure fixes (figure 1) during the time interval
‘i’ and αi0 , λ0i and βi0 are parameters of control which can be
respectively equal or different of αi , βi and λi .
Let us note that Cai ≤ CTai and that Cdi ≤ CTdi where
CTai and CTdi are respectively the arrival and departure
capacities of the terminal ‘T’ during ‘i’ such as CTai +CTdi =
CTi . We point out that CTi is the total capacity of the terminal.
Over one period of (15*N) minutes, we will have :
N
X
[αi Xaei + αi0 Ydei ] ≤
i=1
N
X
[λi Cai + λ0i Cdi ]
i=1
≤
N
X
[βi Pvi + βi0 Poci ]
(45)
i=1
Concretely if the capacity is rather large so that no congestion takes place, the problem would not arise. But if there are
more arrival demands than departure demands during one time
period, the problem of the congestion appears. In this case, it
is necessary to minimize the number of delayed planes or to
maximize departure flows as well as possible.
N
X
[λi CTdi +µi CTai ]
i=1
(46)
N represents the number of intervals over one period of
(15*N) minutes;
Poci is the number of occupied places during the time interval
‘i’;
Pvi is the number of empty places during the time interval
‘i’;
Ydei is the number of effective departures during ‘i’;
Xaei is the number of effective arrivals during ‘i’;
CTdi is the departure capacity at the terminal during ‘i’;
CTai is the arrival capacity at the terminal during ‘i’;
αi , βi , γi , δi , λi and µi are parameters of regulation.
The response to this load balancing problem between arrivals and departures consists in minimizing the following
function which we call the new objective function.
N
X
min { [αi Poci − γi Ydei ] + [βi Pvi − δi Xaei ]}
(47)
i=1
We will treat in details these concepts of optimization in a
forthcoming paper.
VIII. C ONCLUSION
In this paper, we presented a modelling of the capacity allocation problem of an airport (CAP) and of its fixes (F CAP )
in form of a constraint satisfaction problem (CSP) called
F CSP AC. The variables represent the number of arrival and
departure demands, the number of planes effectively arrived
and leaved, the number of arrival and departure planes delayed,
the number of demands for arrivals at the different fixes,
the number of effective arrivals at the different fixes, the
number of departure demands at the different fixes, the number
of effective departures at the different fixes, the number of
arrivals delayed at the different fixes, the number of departures
delayed at the different fixes, as well as the number of empty
and occupied places at the terminal. All these variables admit
a common domain which is Di = {0, 1, 2, ..., CT }. The values
which can take the variables Cai and Cdi are dictated by airport managers. The constraints of the problem are the various
equations and inequations governing the available capacities
at the airport. Thereafter, we proposed an optimization model
under constraints for the capacity allocation problem of the
fixes. We showed an example of illustration implementing our
procedures of modelling and optimization. The experimental
results showed that our approach gives results at least as good
as those found in the literature using other techniques. We also
showed how to control and regulate these capacities in order
to assist the managers to deal with unforeseen events.
R EFERENCES
[1] Adams M., S. Kolitz, J. Milner, A.R. Odoni, “Evolutionary concepts for
decentralized air traffic flow management”. Air Traffic Control Quarterly
4, (1996).
[2] Andreatta G., L. Brunetta, “Multiairport ground holding problem: A
computational evaluation of exact algorithms”. In Operations Research
46 (1998).
[3] Andreatta G., Brunetta L. and Guastalla G., “The flow management
problem: Present computational algorithm”. In Proceedings of the Eighth
IFAC Symposium on Transportation System, Chania, (1997).
[4] “Airport Capacity Benchmark”. Report FAA, May (2001).
http://www.faa.gov/events/benchmarks
[5] Dell’Olmo P., G. Lulli. “A Dynamic Programming Approach for the Airport Capacity Allocation Problem”. Technical Paper, Dip. di Statistica,
Probabilità e Statistiche Applicate, Univ. di Roma, (2003).
[6] Dell’Olmo P., G. Lulli, “A new hierarchical architecture for Air Traffic
Management: Optimisation of airway capacity in a Free Flight scenario”.
European Journal of Operational Research 144, (2003).
[7] Duong Vu, “Free route experimental encounter resolution : Initial
results”. Technical Paper, Eurocontrol, Bretigny, (1999).
[8] Fearnsides J. J., “Free flight as a system concept”. In Proceedings of
CAAASD, (1995).
[9] Gilbo E. P., “Airport capacity: Representation, estimation, optimization”.
IEEE Trans. Contr. Syst. Technol., vol. 1, (1993).
[10] Gilbo E. P., “Optimizing Airport Capacity Utilization in Air Traffic Flow
Management Subject to Constraints at Arrival and Departure Fixes”.
IEEE Trans. on Control Systems Technology, Vol. 5, (1997).
[11] Gilbo E.P. and K. Howard, “Collaborative Optimization of Airport
Arrival and Departure Traffic Flow Management Strategies for CDM”.
3rd USA/Europe Air Traffic Management R&D Seminar, Italy, (2000).
[12] Gilbo E. P., “Arrival/Departure Capacity Tradeoff Optimization: a Case
Study at the St. Louis Lambert International Airport (STL)”. 5th
USA/Europe Air Traffic Management R&D Seminar, Budapest, (2003).
[13] Hamzawi S. G., “Lack of airport capacity: Exploration of alternative
solutions”. Transportation Res.-A, vol. 26-A, no. 1, (1992).
[14] Hoffman R. L., M.O. Ball. “A comparison of formulations for the single
airport ground holding problem with banking constraints”. Operations
Research 48, 578-590, (2000).
[15] Idrissi A., “CASC : Aide à la décision au contrôleur aérien par satisfaction de contraintes”. In proceedings of RIVF’05, Can Tho, Viêtnam,
(2005).
[16] Idrissi A. and Chu Min Li, “Modelling Constraint Optimization Problems: Airport Capacity Allocation as a Case Study”. In proceedings of
ICMS’05, Marrakech, (2005).
[17] Idrissi A. and Chu Min Li, “CSPAC : un modèle de résolution du
problème d’allocation de capacités”. In proceedings of MajecStic’05,
Rennes, (2005).
[18] Jenny M., “Foreword”. Air Traffic Control Quarterly 5, (1997).
[19] Korf R., “Linear-space best first search”. Artificial Intelligence, 62:4178, (1993).
[20] Larrosa J. and P. Meseguer, “Exploiting the use of DAC in max-csp”.
In proceedings of CP-96, Cambridge, (1996).
[21] Li Chu Min, F. Manya and J. Planes, “Exploiting Unit Propagation to
Compute Lower Bounds in Branch and Bound Max-SAT Solvers”. In
proceedings of CP-05, (2005).
[22] Montanari U., “Networks of constraints : Fundamental properties and
applications to picture processing”. Information Sciences, 7: 95-132,
(1974).
[23] Odoni A. R., J. Bowman, et al., “Existing and required modeling
capabilities for evaluating ATM system and concepts”. Final Report
International Centre for Air Transportation, MIT, Cambridge, MA,
(1997).
[24] Schiex T., H. Fargier and G. Verfaillie, “Valued constraint satisfaction problems: hard and easy problems”. In proceedings of IJCAI-95.
Montréal, Canada, (1995).
[25] Wambsganss M. C., “Collaborative decision making air traffic management”. In L. Bianco, P. Dell’Olmo, A.R. Odoni (Eds.), New Concepts
and Methods in Air Traffic Management, Springer, Berlin, (2001).
[26] Zhang W., “Modeling and analyzing soft constraint optimization :
Ressource allocation as a case study”. Technical report WU-CIC Working Note 1, Computational Intelligence Center, Washington University,
St Louis, MO 63130, (2001).