Professional Documents
Culture Documents
net/publication/269697792
Genetic Algorithms and Game Theory for Airport Departure Decision Making:
GeDMAN and CoDMAN
CITATIONS READS
5 252
5 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Li Weigang on 23 February 2015.
1 Introduction
The flight delay rate at some of the biggest Brazilian airports is usually around
6% and this value can triplicate on days of bad weather [1]. According to Amer-
ican Federal Aviation Administration (FAA), the delayed flights produce a loss
of almost 22 billions of dollars each year for the airlines all over the world [2].
When delays occur the aircraft spend a significant amount of time in a queue
waiting for the clearance of the runway for departure. Usually, an aircraft per-
forms the waiting process with the engines turned on. These results in a waste
of fuel and emission of pollutants in the atmosphere, in addition to impact to
the airlines that increase operating costs, to the airports that extend the stay
time of aircraft and employees and to the passengers that lose time and money.
The departure management of aircrafts is the pre-departure process, in which
all the flights are distributed to time slots. This process is also a slot allocation
procedure. The aircraft should be allocate within its time slot for departing.
2 Deborah Ferreira et al.
When a flight is delayed, its slot needs to be relocated to another flight, so the
runway can always be in use.
In Brazil, the departure management is done empirically by air traffic con-
trollers, they must negotiate the slots with the airlines with First-Come First-
Served (FCFS) and other unclear criteria. This is a challenge for the flight con-
trollers that already have to deal with some complex situations.
Collaborative Decision Making (CDM) is a paradigm that attempts to im-
prove the exchange of information among the various stakeholders involved in
airport, such as Air Traffic Control (ATC) agency, airlines and airport managers.
Based on this concept, in this paper, two computational departure management
systems are developed. Both of them are created to support air traffic con-
trollers in the creation and modification of the departure sequence. The first
model named as GeDMAN uses Genetic Program to find the best departure se-
quence. And the second one is defined as CoDMAN uses Game Theory to play a
game between the aircraft to decide the order of departure in the slot allocation
process.
Both systems have the goal of reducing the flight delays and benefit the air-
lines that collaborate with the flight controllers, informing problems and delays
in advance, for example. Some researches financed by NASA, about the cause
of flight delays, have shown that automatized tools for departure sequencing
are beneficial for the flight controllers and for the aircraft operators [3]. Using
real flight data from Brasilia International Airport (BSB), the simulation results
show that the developed systems have the capability to manage the departure
sequence automatically and reduce the total flight delay efficiently.
This paper is structured in the following manner: section 2 presents the two
decision methods: 1) HOTRAN, that is used in Brasilia International Airport
and others Brazilian airports and 2) CDM that has been used in many airports
in Europe and USA. Section 3 presents the modeling and tests with GeDMAN.
Section 4 presents how CoDMAN was modeled and tested. Section 5 exhibits a
comparison between GeDMAN and CoDMAN. And, finally, section 6 concludes
this work with the discussions.
Civil) that formalize the concessions for passengers and cargo airlines explo-
ration. It contains the departure and arrival time, number of flights, frequency,
aircraft types and number of seats.
Each flight has its departure and arrival time defined by HOTRAN, however
it is very common that this schedule does not correspond with the real arrival
and departure time. Delays often occur, mainly due to rain and other weather
conditions. When delays occur, if the runway is available, the first aircraft to
make a departure request will be served in First-Come First-Served (FCFS)
manner. Similarly, when a flight is canceled, any other aircraft can make a request
to use the runway in that free slot.
This method has the advantage of being very simple and is easily to imple-
ment, it was one of the first methods used in the aviation history for controlling
the use of the runway. But this simplicity does not always ensure efficiency in
terms of impacting by delays, fuel use, number of passengers and others factors.
This method has many disadvantages, such as it does not allow a planned
optimization of the runway use. The air traffic controller has to keep waiting
for the aircraft request to use the runway, until then, he does not know which
aircraft will be the next to use the runway. In that case, the air traffic controller
has little control over the runway use. Furthermore, long queues can be formed
to use the runway, causing the aircraft to stand still with the engine running,
unnecessarily spending fuel and generating a greater emission of pollutants.
The Compression [7] is another CDM slot allocation algorithm that aims to
improve slot use, filling the empty slots. Compression will adjust all the flights
to try to fill the empty slots left by cancelled and delayed flights.
Based on the algorithms of CDM, some researches have been developed to
improve these basic concepts and to implement the methods and system to
the airports [1,7,11,13,14,15]. This paper intends to compare the developed two
models: GeDMAN and CoDMAN to disseminate the application of CDM in
more airports.
In this solution only the departure has been taken into account, to simplify
the implementation. In Brasilia International Airport (BSB), the slot allocation
process has still not been implemented. In first step, it is needed to create slots
based on the flight’s HOTRAN. The slot size was defined as 1 minute and each
scheduled flight were assigned to the slot closest to its defined departure time.
To use GA, we need to create a genetic representation of the problem, with
the following genetic algorithm structures: gene, chromosome, population, selec-
tion, mutation, crossover and fitness function.
Gene In this solution, the genes are represented by the flights and each slot
will be a locus, i.e., the gene position. The flights departure sequence will be
the population’s individuals. Each flight only can belong to one slot, or else the
flight would departure at two different slots. Each flight is identified by its flight
number and its airline, Figure 1 shows an example with 6 genes (6 flights) and
their 6 positions(6 slots).
GeDMAN and CoDMAN 5
Mutation The mutation will occur with a low probability Pm between 0.01 and
0.005, this probability must be low because a high value of Pm will expand the
search space and the convergence will take more time to happen.
Crossover The crossover has to consider that no flight can be removed from a
chromosome and duplicate flights can not occur, so the type of crossover used
was the order bases crossover, used for permutation problems. The crossover
probability Pc used in this work is between 0.7 and 0.9.
Fitness Function The fitness function should reflect this work objectives, the
used fitness function is presented on Equation 1:
1
Pn × e, if e > 0
( j=1 tdelayj )
f= 1 (1)
Pn
, if e = 0
( j=1 tdelayj )
The previous equation presents the fitness function, where n is the number of
flights, tdelay is the delay time of flight j. The airline is supposed to communicate
early if an delay will occur, if the airline does so, it will be recompensed with
better slots in a new allocation, so the variable e represents the number of airlines
6 Deborah Ferreira et al.
that benefit from this new allocation, considering the airlines that collaborated.
If at least one airline were benefited, this will increase the fitness value(case 1 of
Equation 1). If no airline is benefited by the new slot allocation, then the fitness
value will be computed using case 2 on Equation 1.
(
tcdj − tpdj , if tcdj − tpdj ≥ −5
tdelay = (2)
|tcdj − tpdj |, if tcdj − tpdj < −5
The Equation 2 presentes how tdelay is computed, tcd is the confirmed de-
parture time and tpd is the predicted departure time, if tcdj − tpdj < −5 then
tdelay will be computed using case 1 of Equation 2, that means that the flight
can be advanced for a maximum of 5 minutes, if tcdj − tpdj < 0 that means that
the flight is advanced. If tcdj − tpdj ≥ −5(case 2 of Equation 2)it means that the
flight is very advanced and this solution should not be used, so we will consider
the absolute value and the fitness value will decrease, making this solution less
likely to be selected for reproduction.
3.3 Results
Case 1 - Rush Hour All departed flights at time between 8 a.m. and 12 a.m.
on 20 november of 2013. The total real delay of all flights was 21 hours and 37
minutes.
Case 2 - Not Rush Hour All departed flights at time between 4 p.m. and 8
p.m. The total real delay of all flights was 11 hours and 12 minutes.
The results with the application of GeDMAN are presented on Tables 1 and 2.
The first column presents the total delay of all flights, the second column presents
the average delay by flight, third column shows the mutation probability and
the last column presents crossover probability.
The best result obtained on Case 1 is on the last row of Table 1, with Pm =
0.005 and Pc = 0.7.
The best result obtained on Case 2 is on the last row of Table 2, with Pm =
0.01 and Pc = 0.7.
To verify the efficiency of this model, we compared the real flight data with
the obtained results from GeDMAN and the comparison is presented on Table 3.
First column shows the scenario, second and third column show the mutation
and crossover probability. The fourth column presents the difference between
real delay and the obtained delay and the last column presents the reduction
obtained in percentage.
GeDMAN and CoDMAN 7
Table 3 show that the best result is obtained in Case 2, with Pm = 0.01 and
Pc = 0.7. Case 2 has more empty slots than Case 1, so it has more space for
the GeDMAN to explore a better solution. In all the studied cases the results
obtained with GeDMAN were better than the real data.
CoDMAN proposes the use of Game Theory concepts in the tactical planning
of the flight realization, distributing in an efficient and coordinate manner the
existing slots, according with the demand. It is necessary that the involved agents
participate in a game with perfect information, i.e., where everyone have the
knowledge about themselves and the others.
The contribution of this approach is the fact that the responsibility for airport
incidents, which resides with the air traffic controllers, is transferred directly
to rational agents. They act in a scenario where the negotiation strength will
be a determinant factor in the slot allocation success, with all the operational
requirements imposed by legislation and airport restrictions.
Firstly, the slots for the aircraft are allocate in FIFO (first in, first out) order.
Each aircraft is queued to zero or void, in the ideal case. However, when a flight
wants to occupy a non-empty position, this newly inserted aircraft forces the
queue manipulation, then inflicting unpredicted delay to some (or all) of the
considered flights.
An aircraft can adopt the following actions during a negociation:
1. Try catch the next slot, advancing a position in the line;
2. Spare its slot to another aircraft, losing a position in the line;
3. Remain in its slot.
When a conflict of interest is detected, a situation of negotiation arises and
the agent receives a payoff. This payoff is not directly used as priority attribute,
but it is useful for a measure of negotiation quality.
The importance of the negotiation is that it gives the aircraft an opportunity
to adjust its departure slots. It is important to have its individual interests in
mind, i.e., the aircraft selfishly searches for the best position in the departure
queue. Its interests must be approved by the airport, that will validate or not
the proposition.
Therefore, each aircraft, playing the role of proposer, can propose an accept-
able offer. The consequence is the elaboration of an offer that will bring it a
bigger benefit, while the restrictions of the arbitrator are observed and the op-
ponent aircraft eventually consider the offer as plausible(even if it is not accepted
at first).
If an offer is considered invalid, the proponent aircraft is automatically as-
sumed to be the loser of this negotiation and the last opponents valid offer is
applied. Otherwise, the negotiation will eventually end without an agreement
GeDMAN and CoDMAN 9
and the arbiter will impose the decision among all of those that were offered. It
is important to point out that priority is given to the incoming aircraft instead
of those who wish to take off. This factor has a crucial influence in the arbiter’s
decision.
4.3 Results
To test the applicability and efficiency of CoDMAN, two case studies were cre-
ated with the real data from Brasilia International Airport as scenario:
Case 1: Strategic Allocation In this case, the departure sequence is con-
sidered as a static process. There is no data manipulation in real time. The
departure queue is traversed and the aircraft that are interested in exchanging
slots are identified and an attempt to negotiate with the next aircraft on the
queue is made. The flight plans of 202 aircraft were analyzed in this case.
Case 2: Dynamic Allocation In this case we need to consider two factors:
1. The high demand for arrivals;
2. Conflicting schedules of takeoff between two or more aircrafts, when an ad-
just of the departure time is needed.
The first factor has direct influence in the airport. The limitations referring
to the runway used must be followed. When the airport capacity for departure
reaches zero, a security delay is applied for the next aircraft on the queue. When
the next aircraft is delayed, all other aircraft will also be delayed for departure.
In this case 518 flights were analyzed.
Tables 4 and 5 presents the results for Case 1, the first column shows the
evaluated attributes and second column shows its value. Table 4 presents the
flight information without use of CoDMAN, and Table 5 shows the results with
the negotiations provided by CoDMAN. Comparing these two tables, we can see
that the delay and cost were reduced with use of CoDMAN, the total delay was
reduce by 16.4%.
Attribute Result
Flights out of schedule 35
Average delay by delayed aircraft 2.09 minutes
Average cost by delayed aircraft 17.779
Total delay 73 minutes
Total delay cost in this case 622.267
Tables 6 and 7 present the results for Case 2. Table 6 presents the flight infor-
mation without negotiation, and Table 7 presents the results with negotiations.
The total delay was reduced by 15.1%
10 Deborah Ferreira et al.
Attribute Result
Flights out of schedule 36
Average delay by delayed aircraft 1.69 minutes
Average cost by delayed aircraft 14.3916
Total delay 61 minutes
Total delay cost in this case 518.097
Attribute Result
Flights out of schedule 43
Average delay by delayed aircraft 2 minutes
Average cost by delayed aircraft 17.0643
Total delay 86 minutes
Total delay cost in this case 733.767
Attribute Result
Flights out of schedule 45
Average delay by delayed aircraft 1.62 minutes
Average cost by delayed aircraft 13.1282
Total delay 73 minutes
Total delay cost in this case 590.77
GeDMAN CoDMAN
Reduction of 41% of total delay time Reduction of 16% of total delay time
Aims to reduce total delay Aims to reduce cost of delay
Optimistic model Pessimistic model
6 Conclusion
Collaborative Decision Making is an important mechanism to improve the effi-
ciency of the traffic flow management in airports. To determine the departure
sequence is one basic process in Airport CDM system. In this research, we suc-
cessfully developed two information models: GeDMAN and CoDMAN to manage
the departure sequence to support the decision making by air traffic controller.
The study is important with the motivation to improve the actual situation that
today is empirically done in Brazil. ATC agency, airlines, airports and passen-
gers would be benefited by the new procedure of the decision making. With the
common sharing the information, the delay time would be reduced and the flight
controllers would have less workload on their hands, reducing the risks of human
error.
Both models have been proven, through several test cases with real data from
Brasilia International Airport to have the potential to reduce the total delay time
and help flight controllers with the departure management task.
One possible future work is to implement GeDMAN and CoDMAN models to
airports and Brazilian Air Navigation Management Center (CGNA) and related
ATC units to assist the air traffic controllers and to realize the collaborative
management among the ATC agency, airlines and airport managers in real day
and day operation.
References
1. Rosa, L.P., Ferreira, D.M., Cruciol, Leonardo L.B.V., Weigang L., Deng X.J.: Ge-
netic algorithms for management of taxi scheduling. In: The 2013 International
Conference on Artificial Intelligence, 2013, Las Vegas. CSREA Press (2013)
2. FAA: Federal aviation administration. Available online, http://www.faa.gov/ (2013)
3. Anagnostakis, I., Clarke, J.P.: Runway operations planning: a two-stage solution
methodology. In System Sciences, 2003. Proceedings of 36th Annual Hawaii Inter-
national Conference (2003)
12 Deborah Ferreira et al.