António J.M. Castro and Eugénio Oliveira: LIACC-NIAD&R, FEUP, DEI, University of Porto Portugal

Disruption Management in Airline Operations Control An Intelligent Agent-Based Approach 107
Disruption Management in Airline Operations Control An Intelligent

Agent-Based Approach
Antnio J.M. Castro and Eugnio Oliveira
X

Disruption Management in Airline Operations
Control An Intelligent Agent-Based Approach

Antnio J.M. Castro
1
and Eugnio Oliveira
1

1
LIACC-NIAD&R, FEUP, DEI, University of Porto
Portugal

1. Introduction
Operations control is one of the most important areas for an airline company. Through
operations control mechanisms an airline company monitors all the flights checking if they
follow the schedule that was previously defined by other areas of the company.
Unfortunately, some problems may arise during this stage (Clausen et al., 2005). Those
problems can be related with crewmembers, aircrafts and passengers. The Airline
Operations Control Centre (AOCC) includes teams of experts specialized in solving the
above problems under the supervision of an operation control manager. Each team has a
specific goal contributing to the common and general goal of having the airline operation
running under as few problems as possible. The process of solving these kinds of problems
is known as Disruption Management (Kohl et al., 2004) or Operations Recovery.
To select the best solution to a specific problem, it is necessary to include the actual costs in
the decision process. One can separate the costs in two categories: Direct Operational Costs
(easily quantifiable costs) and Quality Operational Costs (less easily quantifiable costs).
Direct operational costs are, for example, crew related costs (salaries, lodgement, extra-crew
travel, etc.) and aircraft/flights cost (fuel, approach and route taxes, handling services, line
maintenance, etc.). The quality operational costs that AOCC is interested in calculating are,
usually, related with passengers satisfaction. Specifically, we want to include in the decision
process the estimated cost of delaying or cancelling a flight from the passenger point of
view, that is, in terms of the importance that such a delay will have to the passenger.
In this chapter we present our intelligent agent-based approach to help the AOCC solving
the disruption management problem. It is organized as follows: In Section 2 we present
some related regarding operations recovery, a classification of current tools and systems in
use in some airline companies and a brief summary of the current use of software agents
technology in other domains. Section 3 introduces the Airline Operations Control Centre
(AOCC), including typical organizations and problems, the current disruption management
(DM) process and a description of the main costs involved. Section 4 is the main section of
this chapter and presents our agent-based approach to this problem. This section presents:
(i) the reasons that made us adopt the software agents and multi-agent system (MAS)
paradigm; (ii) the MAS architecture including the specific agents, roles and protocols as well
as some relevant agent characteristics like autonomy and social-awareness; (iii) decision
6
Web Intelligence and Intelligent Agents 108

mechanisms, including costs criteria and negotiation protocols and (iv) examples of the
problem solving algorithms used. In Section 5 we present the experimental setup and, in
Section 6, we evaluate our approach, presenting and discussing the results. Finally, in
Section 7, we conclude and give some insights on the future work.

2. Related Work and Current Tools and Systems
The goal of this section is threefold. In Section 2.1 we present the related work regarding
operations recovery. Research in this area has been made, mainly, through Operations
Research (OR) techniques. Barnhart et al., (Barnhart et al., 2003) gives an overview of OR-
based applications in the air transport industry. In Section 2.2 we describe and classify the
current tools and systems in use at some worldwide airlines and in Section 2.3 we present
some interesting examples of how agents are used in other applications domains and
problems.

2.1 Related Work
We divided the bibliography we have analyzed in three main areas: aircraft recovery, crew
recovery and integrated recovery. For a more detailed explanation of those papers as well as
for older papers related with each of these subjects, please consult (Clausen et al., 2005).
Aircraft Recovery: In (Liu et al., 2008) the authors propose a multi-objective genetic
algorithm to generate an efficient time-effective multi-fleet aircraft routing algorithm in
response to disruption of flights. It uses a combination of a traditional genetic algorithm
with a multi-objective optimization method, attempting to optimize objective functions
involving flight connections, flight swaps, total flight delay time and ground turn-around
times. According to the authors () the proposed method has demonstrated the ability to
solve the dynamic and complex problem of airline disruption management. As in other
approaches, the authors do use the delay time in the objective functions but nothing is
included regarding passengers quality of services costs.
Mei Yang Ph.D. thesis (Yang, 2007) investigates the use of advanced tabu search
methodologies to solve the aircraft-grounding problem and the reduced station capacity
problem. The objective is to minimize the schedule recovery costs associated with flight
schedule modifications and deviations from the original route. Mei introduces cancellation
and delay costs in the objective function. For the delay costs, Mei uses a value of $20 if the
delay is less than 15 minutes and $20 each minute if the delay is greater or equal to 15
minutes. For flight cancellations it uses a combination of lost revenue, loss of passenger
goodwill and other negative effects, specific and predefined for each flight. The main
difference regarding our approach is that we allow the definition of profiles for passengers
of each flight (Mei and others, do not consider passengers profiles). Each one with an
associated cost formula, that reflects the delay costs from the passenger point of view.
In (Rosenberger et al., 2001) the authors formulate the problem as a Set Partitioning master
problem and a route generating procedure. The goal is to minimize the cost of cancellation
and retiming, and it is the responsibility of the controllers to define the parameters
accordingly. It is included in the paper a testing process using SimAir (Rosenberger et al.,
2002), simulating 500 days of operations for three fleets ranging in size from 32 to 96 aircraft
servicing 139-407 flights. Although the authors do try to minimize flight delays, nothing is
included regarding the importance of using quality costs.

Crew Recovery: In (Abdelgahny et al., 2004) the flight crew recovery problem for an airline
with a hub-and-spoke network structure is addressed. The paper details and sub-divides the
recovery problem into four categories: misplacement problems, rest problems, duty
problems and unassigned problems. The proposed model is an assignment model with side
constraints. Due to the stepwise approach, the proposed solution is sub-optimal. Results are
presented for a situation involving a US airline taking into account 18 different problems.
This work also omits the use of quality costs for deriving an appropriate solution.
Integrated Recovery: In (Bratu & Barnhart, 2006) the author presents two models that
considers aircraft and crew recovery and through the objective function focuses on
passenger recovery. They include delay costs that capture relevant hotel costs and ticket
costs if passengers are recovered by other airlines. According to the authors, it is possible to
include, although hard to calculate, estimations of delay costs to passengers and potential
costs of loosing future ticket sales. To test those models an AOCC simulator was developed,
simulating domestic operations of a major US airline. It involves 302 aircrafts divided into 4
fleets, 74 airports and 3 hubs. Furthermore, 83869 passengers on 9925 different passengers
itineraries per day are used. For all scenarios solutions are generated with reductions in
passenger delays and disruptions. The difference comparing with our approach is that we
propose a generic model to calculate the delay cost to passengers, based on their specific
profile and opinion (obtained through frequent surveys).
In (Kohl et al., 2004) the author reports on the experiences obtained during the research and
development of project DESCARTES (a large scale project supported by EU) on airline
disruption management. The current (almost manual) mode of dealing with recovery is
presented. They also present the results of the first prototype of a multiple resource decision
support system. Passenger delay costs are calculated regarding the delay at the destination
and not at departure (we include both in our proposal) and takes into consideration the
commercial value of the passenger based on the booked fare class and frequent flyer
information. The main difference regarding our proposal is that we use the opinion of the
passengers when calculating the importance of the delay.
Lettovskys Ph.D. thesis (Lettovsky, 1997) is the first presentation of a truly integrated
approach in the literature, although only parts of it are implemented. The thesis presents a
linear mixed-integer mathematical problem that maximizes total profit to the airline while
capturing availability of the three most important resources: aircraft, crew and passengers.
The formulation has three parts corresponding to each of the resources, that is, crew
assignment, aircraft routing and passenger flow. In a decomposition scheme these are three
parts of a master problem known as the Schedule Recovery Model. Although the author
takes into consideration the passenger, it does so concerning finding the best solution for the
disrupted passengers. The difference of our approach is that we use the opinion of the
passengers regarding the delay (expressed through a mathematical formula) to reach the
best possible solution concerning delaying the flight. We still do not approach (at least at
present time) the, also important, issue of finding the best itinerary for disrupted
passengers.

2.2 Current Tools and Systems
In previous work (Castro, 2008) we have classified the current tools (or systems that provide
those tools) in use at AOCCs in one of these three categories:


mechanisms, including costs criteria and negotiation protocols and (iv) examples of the
problem solving algorithms used. In Section 5 we present the experimental setup and, in
Section 6, we evaluate our approach, presenting and discussing the results. Finally, in
Section 7, we conclude and give some insights on the future work.

2. Related Work and Current Tools and Systems
The goal of this section is threefold. In Section 2.1 we present the related work regarding
operations recovery. Research in this area has been made, mainly, through Operations
Research (OR) techniques. Barnhart et al., (Barnhart et al., 2003) gives an overview of OR-
based applications in the air transport industry. In Section 2.2 we describe and classify the
current tools and systems in use at some worldwide airlines and in Section 2.3 we present
some interesting examples of how agents are used in other applications domains and
problems.

2.1 Related Work
We divided the bibliography we have analyzed in three main areas: aircraft recovery, crew
recovery and integrated recovery. For a more detailed explanation of those papers as well as
for older papers related with each of these subjects, please consult (Clausen et al., 2005).
Aircraft Recovery: In (Liu et al., 2008) the authors propose a multi-objective genetic
algorithm to generate an efficient time-effective multi-fleet aircraft routing algorithm in
response to disruption of flights. It uses a combination of a traditional genetic algorithm
with a multi-objective optimization method, attempting to optimize objective functions
involving flight connections, flight swaps, total flight delay time and ground turn-around
times. According to the authors () the proposed method has demonstrated the ability to
solve the dynamic and complex problem of airline disruption management. As in other
approaches, the authors do use the delay time in the objective functions but nothing is
included regarding passengers quality of services costs.
Mei Yang Ph.D. thesis (Yang, 2007) investigates the use of advanced tabu search
methodologies to solve the aircraft-grounding problem and the reduced station capacity
problem. The objective is to minimize the schedule recovery costs associated with flight
schedule modifications and deviations from the original route. Mei introduces cancellation
and delay costs in the objective function. For the delay costs, Mei uses a value of $20 if the
delay is less than 15 minutes and $20 each minute if the delay is greater or equal to 15
minutes. For flight cancellations it uses a combination of lost revenue, loss of passenger
goodwill and other negative effects, specific and predefined for each flight. The main
difference regarding our approach is that we allow the definition of profiles for passengers
of each flight (Mei and others, do not consider passengers profiles). Each one with an
associated cost formula, that reflects the delay costs from the passenger point of view.
In (Rosenberger et al., 2001) the authors formulate the problem as a Set Partitioning master
problem and a route generating procedure. The goal is to minimize the cost of cancellation
and retiming, and it is the responsibility of the controllers to define the parameters
accordingly. It is included in the paper a testing process using SimAir (Rosenberger et al.,
2002), simulating 500 days of operations for three fleets ranging in size from 32 to 96 aircraft
servicing 139-407 flights. Although the authors do try to minimize flight delays, nothing is
included regarding the importance of using quality costs.

Crew Recovery: In (Abdelgahny et al., 2004) the flight crew recovery problem for an airline
with a hub-and-spoke network structure is addressed. The paper details and sub-divides the
recovery problem into four categories: misplacement problems, rest problems, duty
problems and unassigned problems. The proposed model is an assignment model with side
constraints. Due to the stepwise approach, the proposed solution is sub-optimal. Results are
presented for a situation involving a US airline taking into account 18 different problems.
This work also omits the use of quality costs for deriving an appropriate solution.
Integrated Recovery: In (Bratu & Barnhart, 2006) the author presents two models that
considers aircraft and crew recovery and through the objective function focuses on
passenger recovery. They include delay costs that capture relevant hotel costs and ticket
costs if passengers are recovered by other airlines. According to the authors, it is possible to
include, although hard to calculate, estimations of delay costs to passengers and potential
costs of loosing future ticket sales. To test those models an AOCC simulator was developed,
simulating domestic operations of a major US airline. It involves 302 aircrafts divided into 4
fleets, 74 airports and 3 hubs. Furthermore, 83869 passengers on 9925 different passengers
itineraries per day are used. For all scenarios solutions are generated with reductions in
passenger delays and disruptions. The difference comparing with our approach is that we
propose a generic model to calculate the delay cost to passengers, based on their specific
profile and opinion (obtained through frequent surveys).
In (Kohl et al., 2004) the author reports on the experiences obtained during the research and
development of project DESCARTES (a large scale project supported by EU) on airline
disruption management. The current (almost manual) mode of dealing with recovery is
presented. They also present the results of the first prototype of a multiple resource decision
support system. Passenger delay costs are calculated regarding the delay at the destination
and not at departure (we include both in our proposal) and takes into consideration the
commercial value of the passenger based on the booked fare class and frequent flyer
information. The main difference regarding our proposal is that we use the opinion of the
passengers when calculating the importance of the delay.
Lettovskys Ph.D. thesis (Lettovsky, 1997) is the first presentation of a truly integrated
approach in the literature, although only parts of it are implemented. The thesis presents a
linear mixed-integer mathematical problem that maximizes total profit to the airline while
capturing availability of the three most important resources: aircraft, crew and passengers.
The formulation has three parts corresponding to each of the resources, that is, crew
assignment, aircraft routing and passenger flow. In a decomposition scheme these are three
parts of a master problem known as the Schedule Recovery Model. Although the author
takes into consideration the passenger, it does so concerning finding the best solution for the
disrupted passengers. The difference of our approach is that we use the opinion of the
passengers regarding the delay (expressed through a mathematical formula) to reach the
best possible solution concerning delaying the flight. We still do not approach (at least at
present time) the, also important, issue of finding the best itinerary for disrupted
passengers.

2.2 Current Tools and Systems
In previous work (Castro, 2008) we have classified the current tools (or systems that provide
those tools) in use at AOCCs in one of these three categories:


1. Database Query Systems (DBQS)
2. Decision Support Systems (DSS)
3. Automatic or Semi-Automatic Systems (ASAS)
The DBQS Database Query Systems (the most common situation at airlines) allows the
AOCC human operators to perform queries on the existing databases to monitor the airline
operation and to obtain other data essential for decision-making. For example, the aircraft
and/or crew roster, aircraft maintenance schedule, passenger reservations, and so on. These
systems are useful and relatively easy to implement and/or acquire but they have some
important disadvantages, for example, to find the best solution and to take the best decision
is completely dependent on the human operator. As we have explained in (Castro, 2008)
there are two problems when airline companies use only this type of systems: (1) the
solution quality is dependent on knowledge and experience of the human operator and, (2)
due to the usual difficulty of the human being in leading with large volumes of data
simultaneously, they do not use all the necessary information (variables) to take the best
decision.
The DSS - Decision Support Systems, besides having the same characteristics of the DBQS,
also include additional functionalities to support the human operators on the decision-
making. For example, after a request made by a human operator, these systems are able to
recommend the best solution to solve a problem related with a delayed aircraft. Some of
them may just recommend a flight re-scheduling but others are able to justify the candidate
solution as well as to present the solution cost. DSS systems eliminate some of the
disadvantages of the DBQS systems. Namely, they are able to analyze large volumes of data
and, because of that, propose solutions that take into consideration more information
(variables). The decision-making still is on the human operator side but, now, he is able to
take better decisions. Unfortunately, one of the big problems with airline companies is the
absence and/or complexity of the computerized information system keeping all the
operational information. These are of paramount importance for the success of the decision
support tools. This problem, referred in (Kohl et al., 2004) as the Data Quality and System
Accessibility Problem, gains more importance when we start to implement decision support
tools and/or automatic or semi-automatic systems.
The goal of the third type of systems, ASAS Automatic or Semi-Automatic Systems, is to
automate as much as possible the AOCC, replacing the functional part by computerized
programs. Specifically, these systems try to automate the repetitive tasks and also the tasks
related with searching for the best solution (problem solving). In a totally automatic system,
decision-making is also taken by the system. In a semi-automatic system, the final decision
is taken by the human operator. In ASAS type of systems, the AOCC does not need as much
human operators as in the previous ones, to operate correctly. Usually, roles or functions
related with operation monitoring, searching for solutions related with aircraft, crew or
passenger problems and re-allocation of resources, are performed by specialists agents
(Castro & Oliveira, 2007) replacing the human specialists. The final decision regarding the
application of the solution found by these systems on the environment (for example, making
the necessary changes on the airline operational plan database) depends on the human
supervisor. According to (Wooldridge, 2009) and (Castro, 2007) the agent and multi-agent
systems paradigm is more appropriate to be used in this domain than any other paradigm.

2.3
To
rep
com
(Ca
bee
do
de
Wo
ma
pro
age
by

3.
In
air
the
exp
AO
we

3.1
Ac
by
ma
on
rep

Fig

3 Other Applicat
o the best of our
present all roles
mmon overall go
astro, 2007), (Ca
en applied both
omains. A brief a
veloped a multi-
olfe et al., use ag
anagement (Wol
oposed the use o
ent system for th
y Ouelhadj (Ouelh
Airline Operat
this section we i
rline disruption m
e AOCP precedin
plain what an air
OCC organization
ell as the main co
1 Airline Schedu
ccording to (Kohl
y the long and sh
ain dimensions o
ne represents the
presents resource
g. 1. The airline sc
tion Domains
knowledge, we
of an AOCC, in
oal of solving the
stro & Oliveira,
to other problem
and incomplete
-agent algorithm
gents to compare
fe et al., 2007).
of multi-agent sy
he integrated dyn
hadj, 2003), (Cow
tions Control
introduce the airl
management pro
ng problem know
rline operational
ns. The typical p
sts involved are a
uling Problem
l et al., 2004) the
hort-term phases p
r views: (1) passe
seats available to
es that will be allo
cheduling proces
were the first to
ncluding specialis
e unexpected pro
2007). However,
ms in air transpor
list of such app
for traffic flow m
e routing selection
For ATC Towe
ystems (Jonker et
namic scheduling
ling et al., 2003).
line operations co
blem). To contex
wn as the Airlin
l control centre (A
roblems, the curr
also introduced.
scheduling proc
presented in Figu
enger view; (2) ai
o be sold to the a
ocated.
s
o propose an org
st agents that co
oblems arising du
, agents and mu
rtation domain an
lications follows
management (Tu
n strategies in co
er operations, Jo
t al., 2005). As a
g of steel product
ontrol problem
xtualize, we start
ne Scheduling Pro
AOCC) is and w
rent disruption m
cess of an airline
ure 1. The schedu
ircraft view and
airline customers
ganization of age
ooperate to achie
uring airline oper
ulti-agent systems
nd in other appli
s. Tumer and Ag
mer & Agogino,
ollaborative traffi
onker et al., hav
last example, a
tion has been pro
AOCP (also kno
by briefly introd
oblem (ASP). Th
we present some t
management proc
company is com
uling process has
(3) crew view. Th
s. The other two

ents to
eve the
rations
s have
ication
gogino
2007).
ic flow
ve also
multi-
oposed
own as
ducing
hen we
typical
cess as
mposed
s three
he first
views,

1. Database Query Systems (DBQS)
2. Decision Support Systems (DSS)
3. Automatic or Semi-Automatic Systems (ASAS)
The DBQS Database Query Systems (the most common situation at airlines) allows the
AOCC human operators to perform queries on the existing databases to monitor the airline
operation and to obtain other data essential for decision-making. For example, the aircraft
and/or crew roster, aircraft maintenance schedule, passenger reservations, and so on. These
systems are useful and relatively easy to implement and/or acquire but they have some
important disadvantages, for example, to find the best solution and to take the best decision
is completely dependent on the human operator. As we have explained in (Castro, 2008)
there are two problems when airline companies use only this type of systems: (1) the
solution quality is dependent on knowledge and experience of the human operator and, (2)
due to the usual difficulty of the human being in leading with large volumes of data
simultaneously, they do not use all the necessary information (variables) to take the best
decision.
The DSS - Decision Support Systems, besides having the same characteristics of the DBQS,
also include additional functionalities to support the human operators on the decision-
making. For example, after a request made by a human operator, these systems are able to
recommend the best solution to solve a problem related with a delayed aircraft. Some of
them may just recommend a flight re-scheduling but others are able to justify the candidate
solution as well as to present the solution cost. DSS systems eliminate some of the
disadvantages of the DBQS systems. Namely, they are able to analyze large volumes of data
and, because of that, propose solutions that take into consideration more information
(variables). The decision-making still is on the human operator side but, now, he is able to
take better decisions. Unfortunately, one of the big problems with airline companies is the
absence and/or complexity of the computerized information system keeping all the
operational information. These are of paramount importance for the success of the decision
support tools. This problem, referred in (Kohl et al., 2004) as the Data Quality and System
Accessibility Problem, gains more importance when we start to implement decision support
tools and/or automatic or semi-automatic systems.
The goal of the third type of systems, ASAS Automatic or Semi-Automatic Systems, is to
automate as much as possible the AOCC, replacing the functional part by computerized
programs. Specifically, these systems try to automate the repetitive tasks and also the tasks
related with searching for the best solution (problem solving). In a totally automatic system,
decision-making is also taken by the system. In a semi-automatic system, the final decision
is taken by the human operator. In ASAS type of systems, the AOCC does not need as much
human operators as in the previous ones, to operate correctly. Usually, roles or functions
related with operation monitoring, searching for solutions related with aircraft, crew or
passenger problems and re-allocation of resources, are performed by specialists agents
(Castro & Oliveira, 2007) replacing the human specialists. The final decision regarding the
application of the solution found by these systems on the environment (for example, making
the necessary changes on the airline operational plan database) depends on the human
supervisor. According to (Wooldridge, 2009) and (Castro, 2007) the agent and multi-agent
systems paradigm is more appropriate to be used in this domain than any other paradigm.

2.3
To
rep
com
(Ca
bee
do
de
Wo
ma
pro
age
by

3.
In
air
the
exp
AO
we

3.1
Ac
by
ma
on
rep

Fig

3 Other Applicat
o the best of our
present all roles
mmon overall go
astro, 2007), (Ca
en applied both
omains. A brief a
veloped a multi-
olfe et al., use ag
anagement (Wol
oposed the use o
ent system for th
y Ouelhadj (Ouelh
Airline Operat
this section we i
rline disruption m
e AOCP precedin
plain what an air
OCC organization
ell as the main co
1 Airline Schedu
ccording to (Kohl
y the long and sh
ain dimensions o
ne represents the
presents resource
g. 1. The airline sc
tion Domains
knowledge, we
of an AOCC, in
oal of solving the
stro & Oliveira,
to other problem
and incomplete
-agent algorithm
gents to compare
fe et al., 2007).
of multi-agent sy
he integrated dyn
hadj, 2003), (Cow
tions Control
introduce the airl
management pro
ng problem know
rline operational
ns. The typical p
sts involved are a
uling Problem
l et al., 2004) the
hort-term phases p
r views: (1) passe
seats available to
es that will be allo
cheduling proces
were the first to
ncluding specialis
e unexpected pro
2007). However,
ms in air transpor
list of such app
for traffic flow m
e routing selection
For ATC Towe
ystems (Jonker et
namic scheduling
ling et al., 2003).
line operations co
blem). To contex
wn as the Airlin
l control centre (A
roblems, the curr
also introduced.
scheduling proc
presented in Figu
enger view; (2) ai
o be sold to the a
ocated.
s
o propose an org
st agents that co
oblems arising du
, agents and mu
rtation domain an
lications follows
management (Tu
n strategies in co
er operations, Jo
t al., 2005). As a
g of steel product
ontrol problem
xtualize, we start
ne Scheduling Pro
AOCC) is and w
rent disruption m
cess of an airline
ure 1. The schedu
ircraft view and
airline customers
ganization of age
ooperate to achie
uring airline oper
ulti-agent systems
nd in other appli
s. Tumer and Ag
mer & Agogino,
ollaborative traffi
onker et al., hav
last example, a
tion has been pro
AOCP (also kno
by briefly introd
oblem (ASP). Th
we present some t
management proc
company is com
uling process has
(3) crew view. Th
s. The other two

ents to
eve the
rations
s have
ication
gogino
2007).
ic flow
ve also
multi-
oposed
own as
ducing
hen we
typical
cess as
mposed
s three
he first
views,

Everything starts with publishing the flights timetable for a specific period of time (usually six
months). After publishing the timetable, the revenue management phase starts. Here the goal
is to maximize the revenue obtained selling tickets. At the same time, the scheduling of the
two most important resources starts: aircrafts and crew. Regarding the aircraft, the first step
is the fleet assignment. Here, the goal is to assign the aircraft type or aircraft fleet that will
perform the flights. It is an important step because the aircraft type/fleet will define the
number of available seats in each flight. Near to the day of operations, the assignment of the
specific aircraft to each flight is performed. This step is known as tail assignment. After the
fleet assignment step, it is possible to start to schedule the crew. The first step is the crew
pairing. The goal is to define the crew duty periods (pairings) that will be necessary to cover
all the flights of the airline for a specific period of time (typical one month). Having the
pairings, it is possible to start the crew rostering step that is, assign crewmembers to the
pairings. The output of this step is an individual crew roster that is distributed or published
in the crew web portal. Finally and until the day of operations, it is necessary to
change/updated the crew roster (roster maintenance), to include any changes that might
appear after publishing the roster. The airline scheduling problem (ASP) is composed of all
the previous phases and steps and ends some hours or days (depends on the airline policy)
before the day of operation. The global objective of the ASP is to maximize the airline
operating profit. For more detailed information please consult (Grosche, 2009) specially
Section 2.1 to Section 2.4.

3.2 AOCC Organization
The airline operations control problem (AOCP) starts where the airline scheduling problem
stops. In Figure 1 the AOCP is represented by the disruption management square. If
everything goes as planned the airline just needs to monitor the execution of the plan.
Unfortunately, several unexpected events appear during this phase that can disrupt the
plan. To monitor those events and solve the problems that arise from these disruptions and
return to the previous plan as soon as possible, it is necessary to define and follow a
disruption management process. Airline companies have an entity called Airline Operations
Control Centre (AOCC) that is responsible for the disruption management process. There
are three main AOCC organizations (Castro, 2008):
Decision Centre: The aircraft controllers share the same physical space. The other
roles or support functions (crew control, maintenance service, etc.) are in a
different physical space. In this type of Collective Organization all roles need to
cooperate to achieve the common goal.
Integrated Centre: All roles share the same physical space and are hierarchically
dependent of a supervisor. For small companies we have a Simple Hierarchy
Organization. For bigger companies we have a Multidimensional Hierarchy
Organization. Figure 2 shows an example of this kind of AOCC organization.
Hub Control Centre (HCC): Most of the roles are physically separated at the airports
where the airline companies operate a hub. In this case, if the aircraft controller role
stays physically outside the hub we have an organization called Decision Centre
with a hub. If the both the aircraft controller and crew controller roles are physically
outside the hub we have an organization called Integrated Centre with a hub. The
main advantage of this kind of organization is to have the roles that are related

Th
(fo
pre
po
an
AO
fro

Fig

Th
an

with airpo
etc.) physi
he organization a
or example, hub-a
esent the organiz
oint out the role o
d, also, the ope
OCC. This operat
om 72 to 24 hours
g. 2. Integrated ai
he roles or suppo
d (Castro, 2008),
Flight Disp
Traffic Co
Europe, fo
Aircraft Co
the operat
Crew Cont
out, updat
appear du
Maintenan
short-term
short-term
Passenger
passengers
the decisio
bigger com
ort operations (cu
ically closer to the
dopted depends
and-spoke) and g
zation of a typica
of the supervisor,
eration time-win
tion time-window
s before to 12 to 24
irline operational
ort functions mor
are the following
patch: Prepares th
ontrol (ATC) enti
or example).
ontrol: Manages t
ional control.
trol: Manages the
tes and changes
uring the operatio
ce Services: Respo
m maintenance sc
m maintenance (m
Services: Decisio
s. The responsibi
ons on passenger
mpanies is part of
ustomer service,
e operation.
on several factor
geographic distri
al Integrated Oper
a characteristic th
dow that marks
w is different fro
4 hours after the
control centre
e common in an
g:
he flight plans a
ities (FAA in No
he resource aircr
e resource crew.
the crew roster a
on.
onsible for the un
heduling. Chang
maintenance canno
ons taken on the
lity of this role is
rs. Typical this ro
f the HCC organi
catering, cleanin
rs like airline siz
ibution of the ope
rational Control Ce
hat makes this or
s the responsibil
m airline to airli
day of operation
AOCC, accordin
nd requests new
orth America an
raft. It is the centr
Monitors the cre
according to the
nplanned mainte
ges on aircraft rot
ot be done at all s
e AOCC will ha
s to consider and
ole is performed
zation.
ng, passengers tra
ze, airline networ
eration. In Figure
entre. It is import
rganization hierar
lity boundaries
ne but, usually, r
.

ng to (Kohl et al.,
w flight slots to th
nd EUROCONTR
ral coordination
ew check-in and
disruptions that
enance services a
tations may impa
stations).
ave an impact o
minimize the imp
on the airports a
ansfer,
rk type
e 2 we
tant to
rchical
of the
ranges
, 2004)
he Air
ROL in
role in
check-
might
and for
act the
on the
pact of
and for

Everything starts with publishing the flights timetable for a specific period of time (usually six
months). After publishing the timetable, the revenue management phase starts. Here the goal
is to maximize the revenue obtained selling tickets. At the same time, the scheduling of the
two most important resources starts: aircrafts and crew. Regarding the aircraft, the first step
is the fleet assignment. Here, the goal is to assign the aircraft type or aircraft fleet that will
perform the flights. It is an important step because the aircraft type/fleet will define the
number of available seats in each flight. Near to the day of operations, the assignment of the
specific aircraft to each flight is performed. This step is known as tail assignment. After the
fleet assignment step, it is possible to start to schedule the crew. The first step is the crew
pairing. The goal is to define the crew duty periods (pairings) that will be necessary to cover
all the flights of the airline for a specific period of time (typical one month). Having the
pairings, it is possible to start the crew rostering step that is, assign crewmembers to the
pairings. The output of this step is an individual crew roster that is distributed or published
in the crew web portal. Finally and until the day of operations, it is necessary to
change/updated the crew roster (roster maintenance), to include any changes that might
appear after publishing the roster. The airline scheduling problem (ASP) is composed of all
the previous phases and steps and ends some hours or days (depends on the airline policy)
before the day of operation. The global objective of the ASP is to maximize the airline
operating profit. For more detailed information please consult (Grosche, 2009) specially
Section 2.1 to Section 2.4.

3.2 AOCC Organization
The airline operations control problem (AOCP) starts where the airline scheduling problem
stops. In Figure 1 the AOCP is represented by the disruption management square. If
everything goes as planned the airline just needs to monitor the execution of the plan.
Unfortunately, several unexpected events appear during this phase that can disrupt the
plan. To monitor those events and solve the problems that arise from these disruptions and
return to the previous plan as soon as possible, it is necessary to define and follow a
disruption management process. Airline companies have an entity called Airline Operations
Control Centre (AOCC) that is responsible for the disruption management process. There
are three main AOCC organizations (Castro, 2008):
Decision Centre: The aircraft controllers share the same physical space. The other
roles or support functions (crew control, maintenance service, etc.) are in a
different physical space. In this type of Collective Organization all roles need to
cooperate to achieve the common goal.
Integrated Centre: All roles share the same physical space and are hierarchically
dependent of a supervisor. For small companies we have a Simple Hierarchy
Organization. For bigger companies we have a Multidimensional Hierarchy
Organization. Figure 2 shows an example of this kind of AOCC organization.
Hub Control Centre (HCC): Most of the roles are physically separated at the airports
where the airline companies operate a hub. In this case, if the aircraft controller role
stays physically outside the hub we have an organization called Decision Centre
with a hub. If the both the aircraft controller and crew controller roles are physically
outside the hub we have an organization called Integrated Centre with a hub. The
main advantage of this kind of organization is to have the roles that are related

Th
(fo
pre
po
an
AO
fro

Fig

Th
an

with airpo
etc.) physi
he organization a
or example, hub-a
esent the organiz
oint out the role o
d, also, the ope
OCC. This operat
om 72 to 24 hours
g. 2. Integrated ai
he roles or suppo
d (Castro, 2008),
Flight Disp
Traffic Co
Europe, fo
Aircraft Co
the operat
Crew Cont
out, updat
appear du
Maintenan
short-term
short-term
Passenger
passengers
the decisio
bigger com
ort operations (cu
ically closer to the
dopted depends
and-spoke) and g
zation of a typica
of the supervisor,
eration time-win
tion time-window
s before to 12 to 24
irline operational
ort functions mor
are the following
patch: Prepares th
ontrol (ATC) enti
or example).
ontrol: Manages t
ional control.
trol: Manages the
tes and changes
uring the operatio
ce Services: Respo
m maintenance sc
m maintenance (m
Services: Decisio
s. The responsibi
ons on passenger
mpanies is part of
ustomer service,
e operation.
on several factor
geographic distri
al Integrated Oper
a characteristic th
dow that marks
w is different fro
4 hours after the
control centre
e common in an
g:
he flight plans a
ities (FAA in No
he resource aircr
e resource crew.
the crew roster a
on.
onsible for the un
heduling. Chang
maintenance canno
ons taken on the
lity of this role is
rs. Typical this ro
f the HCC organi
catering, cleanin
rs like airline siz
ibution of the ope
rational Control Ce
hat makes this or
s the responsibil
m airline to airli
day of operation
AOCC, accordin
nd requests new
orth America an
raft. It is the centr
Monitors the cre
according to the
nplanned mainte
ges on aircraft rot
ot be done at all s
e AOCC will ha
s to consider and
ole is performed
zation.
ng, passengers tra
ze, airline networ
eration. In Figure
entre. It is import
rganization hierar
lity boundaries
ne but, usually, r
.

ng to (Kohl et al.,
w flight slots to th
nd EUROCONTR
ral coordination
ew check-in and
disruptions that
enance services a
tations may impa
stations).
ave an impact o
minimize the imp
on the airports a
ansfer,
rk type
e 2 we
tant to
rchical
of the
ranges
, 2004)
he Air
ROL in
role in
check-
might
and for
act the
on the
pact of
and for

3.3
In
tho
du
fro
dia
de
tha
pro
dif
ha

Fig

As
Fli
pre
is n
de
de
air
de
de
pa
du
or

3.4
As
de
de
mi
pro

3 Typical Proble
the previous sect
ose organizations
uring the executio
om (Kohl & Karis
agram we have a
parture delays a
at the problems
oblems on differ
fficult to be solve
ve on the AOCC.
g. 3. Typical AOC
s we can see in F
ight Departure De
evious flights. If
not enough, then
parture delay. Fr
lay (besides the
rcraft malfunction
lay it is necessa
parture delays, t
ssenger delays as
uty, air traffic con
at arrival) are the
4 Current Disrup
s we can see from
lays. AOCCs hav
lays are minimiz
inimum operatio
ocess in use at mo
ems
tion we presented
s. Now, it is imp
on of the airline
sch, 2004), we fo
lso included the i
s well as the rela
might propagate
rent flights. This
ed optimally in
.
CC problems and
Figure 3 there is
elays. Most of th
we have an arriv
n, if the AOCC d
rom the diagram
delay on depar
and flight divers
ary a cooperation
the main reason
s a consequence o
ntrol reasons, airc
e other main reas
ption Manageme
m the previous se
ve a process to m
zed with the min
onal cost. In Figu
ost of the airlines
d typical AOCC o
portant to unders
operation. From
ound the typical p
impact that each
ation that exist b
e due to the rela
s propagation ch
a real time and
relations
s an obvious rela
he flights are pe
val delay and the
oes not find an a
m we can also see
rture) are: En-rou
sion. In the previ
n between the p
s are: crew dela
of an arrival dela
craft malfunction
sons for departure
ent Process
ection, there are s
monitor the event
nimum impact o
ure 4 we presen
s. This process ha
organizations and
stand the typical
our observation
problems presen
problem might h
between them. Th
ation between th
haracteristic mak
dynamic environ
ation between Fli
erformed by airc
aircraft turn-aro
alternative solutio
that the main re
ute air traffic, en
ious cases and to
pilot, the AOCC
ys, cargo/bagga
ay. Crewmember
ns and weather co
e delays.
several problems
ts and solve the p
on passenger and
nt the current di
s five steps:
d the roles that ex
l problems that a
ns in a real AOCC
nted in Figure 3.
have on flight arr
he diagram also
hem, and generat
kes the problem
nment, like the o

ight Arrival Delay
crafts that are u
ound time at the a
on, we will also h
easons for flight a
n-route weather, en
o minimize the a
C and ATC. Rega
ge loading delay
s that do not rep
onditions (at dep
that might cause
problems, so that
d, preferably, wi
isruption manag
xist on
appear
C, and
In this
rival or
shows
te new
m more
one we
ys and
sed in
airport
have a
arrival
n-route
arrival
arding
ys and
port for
parture
e flight
t flight
ith the
gement

1. Operation Monitoring: In this step the flights are monitored to see if anything is not
going according the plan. The same happens in relation with crewmembers,
passenger check-in and boarding, cargo and baggage loading, etc.
2. Take Action: If an event happens, like for example, a crewmember is delayed or an
aircraft malfunction, a quick assessment is performed to see if an action is required.
If not, the monitoring continues. If an action is necessary than we have a problem
that needs to be solved.
3. Generate and Evaluate Solutions: Having all the information regarding the problem
the AOCC needs to find and evaluate the candidate solutions. Although there are
several costs involved in this process, we found that the AOCC relies heavily on
the experience of their controllers and in some rules-of-thumb (a kind of hidden
knowledge) that exist on the AOCC.
4. Take Decision: Having the candidate solutions a decision needs to be taken.
5. Apply Decision: After the decision the final solution needs to be applied in the
environment, that is, the operational plan needs to be updated accordingly.

Fig. 4. AOCC disruption management process

In our opinion, this process can greatly benefit from an intelligent agent based approach to
the problem, as we will explain in Section 4.

3.5 Main Costs Involved
In the step Generate and Evaluate Solutions of the disruption management process on the
previous section, we should consider the main costs involved in generating and choosing
from candidate solutions. According to our observations these are the main costs involved
when generating and evaluating a solution for a specific disruption:

1. Crew Costs: the average or real salary costs of the crewmembers, additional work
hours and perdiem days to be paid, hotel costs and extra-crew travel costs.
2. Flight Costs: airport costs (approach and taxing taxes, for example), service costs
(cleaning services, handling services, line maintenance, etc.), and average
maintenance costs for the type of aircraft, ATC en-route charges and fuel
consumption.

3.3
In
tho
du
fro
dia
de
tha
pro
dif
ha

Fig

As
Fli
pre
is n
de
de
air
de
de
pa
du
or

3.4
As
de
de
mi
pro

3 Typical Proble
the previous sect
ose organizations
uring the executio
om (Kohl & Karis
agram we have a
parture delays a
at the problems
oblems on differ
fficult to be solve
ve on the AOCC.
g. 3. Typical AOC
s we can see in F
ight Departure De
evious flights. If
not enough, then
parture delay. Fr
lay (besides the
rcraft malfunction
lay it is necessa
parture delays, t
ssenger delays as
uty, air traffic con
at arrival) are the
4 Current Disrup
s we can see from
lays. AOCCs hav
lays are minimiz
inimum operatio
ocess in use at mo
ems
tion we presented
s. Now, it is imp
on of the airline
sch, 2004), we fo
lso included the i
s well as the rela
might propagate
rent flights. This
ed optimally in
.
CC problems and
Figure 3 there is
elays. Most of th
we have an arriv
n, if the AOCC d
rom the diagram
delay on depar
and flight divers
ary a cooperation
the main reason
s a consequence o
ntrol reasons, airc
e other main reas
ption Manageme
m the previous se
ve a process to m
zed with the min
onal cost. In Figu
ost of the airlines
d typical AOCC o
portant to unders
operation. From
ound the typical p
impact that each
ation that exist b
e due to the rela
s propagation ch
a real time and
relations
s an obvious rela
he flights are pe
val delay and the
oes not find an a
m we can also see
rture) are: En-rou
sion. In the previ
n between the p
s are: crew dela
of an arrival dela
craft malfunction
sons for departure
ent Process
ection, there are s
monitor the event
nimum impact o
ure 4 we presen
s. This process ha
organizations and
stand the typical
our observation
problems presen
problem might h
between them. Th
ation between th
haracteristic mak
dynamic environ
ation between Fli
erformed by airc
aircraft turn-aro
alternative solutio
that the main re
ute air traffic, en
ious cases and to
pilot, the AOCC
ys, cargo/bagga
ay. Crewmember
ns and weather co
e delays.
several problems
ts and solve the p
on passenger and
nt the current di
s five steps:
d the roles that ex
l problems that a
ns in a real AOCC
nted in Figure 3.
have on flight arr
he diagram also
hem, and generat
kes the problem
nment, like the o

ight Arrival Delay
crafts that are u
ound time at the a
on, we will also h
easons for flight a
n-route weather, en
o minimize the a
C and ATC. Rega
ge loading delay
s that do not rep
onditions (at dep
that might cause
problems, so that
d, preferably, wi
isruption manag
xist on
appear
C, and
In this
rival or
shows
te new
m more
one we
ys and
sed in
airport
have a
arrival
n-route
arrival
arding
ys and
port for
parture
e flight
t flight
ith the
gement

1. Operation Monitoring: In this step the flights are monitored to see if anything is not
going according the plan. The same happens in relation with crewmembers,
passenger check-in and boarding, cargo and baggage loading, etc.
2. Take Action: If an event happens, like for example, a crewmember is delayed or an
aircraft malfunction, a quick assessment is performed to see if an action is required.
If not, the monitoring continues. If an action is necessary than we have a problem
that needs to be solved.
3. Generate and Evaluate Solutions: Having all the information regarding the problem
the AOCC needs to find and evaluate the candidate solutions. Although there are
several costs involved in this process, we found that the AOCC relies heavily on
the experience of their controllers and in some rules-of-thumb (a kind of hidden
knowledge) that exist on the AOCC.
4. Take Decision: Having the candidate solutions a decision needs to be taken.
5. Apply Decision: After the decision the final solution needs to be applied in the
environment, that is, the operational plan needs to be updated accordingly.

Fig. 4. AOCC disruption management process

In our opinion, this process can greatly benefit from an intelligent agent based approach to
the problem, as we will explain in Section 4.

3.5 Main Costs Involved
In the step Generate and Evaluate Solutions of the disruption management process on the
previous section, we should consider the main costs involved in generating and choosing
from candidate solutions. According to our observations these are the main costs involved
when generating and evaluating a solution for a specific disruption:

1. Crew Costs: the average or real salary costs of the crewmembers, additional work
hours and perdiem days to be paid, hotel costs and extra-crew travel costs.
2. Flight Costs: airport costs (approach and taxing taxes, for example), service costs
(cleaning services, handling services, line maintenance, etc.), and average
maintenance costs for the type of aircraft, ATC en-route charges and fuel
consumption.

3. Passenger Costs: passenger airport meals, passenger hotel costs and passenger
compensations.

Finally, there is a less easily quantifiable cost that is also included: the cost of delaying or
cancelling a flight from the passenger point of view. Most airlines use some kind of rule-of-
thumb when they are evaluating the impact of the decisions on passengers. Others just
assign a monetary cost to each minute of delay and evaluate the solutions taking into
consideration this value. We propose a different way of calculating this cost component.

4. A MAS for Disruption Management in Airline Operations Control
In Section 3 we introduced the Airline Scheduling Problem and the Airline Operations
Control Problem (or Disruption Management Problem). We have described the AOCC
organization and roles as well as the typical problems that appear during the execution of
the operational plan. The disruption management process used by airlines was presented as
well as the main costs involved in generating and evaluating the solutions. In this section we
present our intelligent agent based approach to solve the Disruption Management Problem
in the airline domain. The MAS was developed using Java
1
and JADE (Bellifemine et al.,
2004) as the development platform and as the run-time environment that provides the basic
services for agents to execute.

4.1 Why an Agent and Multi-Agent System Paradigm?
Considering the agent and multi-agent system characteristics as specified in (Wooldridge,
2009) and (Elamy, 2005), the following ones make us adopt this paradigm to the Airline
Operations Control Problem:
Autonomy: MAS models problems in terms of autonomous interacting component-
agents, which are a more natural way of representing task allocation, team
planning, and user preferences, among others. In Figure 5 the PaxManager,
AircraftManager and CrewManager agents (among others) are agents that can choose
to respond or not to the requests according to their own objectives.
Agents are a Natural Metaphor: The AOCC is naturally modelled as a society of
agents cooperating with each other to solve such a complex problem.
Reactivity: Agents are able to perceive and react to the changes in their
environment. The Monitor agent in Figure 5 is an example of such an agent.
Distribution of resources: With a MAS we can distribute the computational resources
and capabilities across a network of interconnected agents avoiding problems
associated with centralized systems. Airline companies of some dimension have
different operational bases. We use a MAS for each operational base, taking
advantage of this important characteristic. Due to the social awareness characteristics
of some of our agents (for example, Monitoring agent in Figure 5) they are able to
distribute their tasks among other agents with similar behaviour.
Modularity and Scalability: A MAS is extensible, scalable, robust, maintainable,
flexible and promotes reuse. These characteristics are very important in systems of

1

http://www.java.com

4.2
It i
aft
(Ca
pro

Fig

Fig
age
gre
pro
age
im

this dimen
more oper
specific pr
Concurrenc
parallel. T
CrewCBR a
Additiona
responsibi
failures by
different c
Legacy Sys
systems. W
interact wi
2 MAS Architect
is important to p
ter performing a
astro & Oliveira
ocess and the bas
g. 5. MAS architec
gure 5 shows the
ents and the na
een lines represe
oblem-solving pr
ents and the nor
mportant to clarify
nsion and compl
rational bases a
roblems.
cy/Parallelism: Ag
This provides flex
and CrewHillClim
ally and accord
ilities are suffici
y one or more ag
computers. If one
stems: The AOCC
We can wrap the
ith other software
ure
point out that we
an analysis and
a, 2008). The age
se for this architec
cture
e architecture of
arrow black dash
ent the interaction
rocess. The narro
rmal black lines r
y that Figure 5 rep
lexity. Our MAS
s well as in sup
gents are capable
xibility and speed
mb agents in Figur
ding to (Stone
iently shared am
gents. Our MAS
or more agents f
C needs informati
legacy compone
e components.
e arrived to the a
design using an
ent model and se
cture.
our multi-agent
h lines represent
n between agents
ow gray lines rep
represent the inte
presents only one
is able to scale i
pporting differen
e of reasoning an
ds up computatio
re 5, are example
& Veloso, 20
mong agents, th
can be totally or
ail, the global obj
on that exists in
ents in an agent l
architecture of o
agent-oriented
ervice model we
system approach
t requests/propo
s regarding nego
present interaction
eractions after a
e instance of the
in terms of supp
nt algorithms to
nd performing ta
on. The CrewSimA
es of concurrent a
000) if contro
he system can to
r partially replica
jective is not affec
obsolete but func
layer, enabling th
ur multi-agent sy
software method
ere the outputs o

h. The boxes rep
osals made. The
otiation and distr
n within a hierar
solution is found
MAS. We can rep
porting
o solve
asks in
Anneal,
agents.
l and
olerate
ated in
cted.
ctional
hem to
ystem,
dology
of this
present
larger
ributed
rchy of
d. It is
plicate

3. Passenger Costs: passenger airport meals, passenger hotel costs and passenger
compensations.

Finally, there is a less easily quantifiable cost that is also included: the cost of delaying or
cancelling a flight from the passenger point of view. Most airlines use some kind of rule-of-
thumb when they are evaluating the impact of the decisions on passengers. Others just
assign a monetary cost to each minute of delay and evaluate the solutions taking into
consideration this value. We propose a different way of calculating this cost component.

4. A MAS for Disruption Management in Airline Operations Control
In Section 3 we introduced the Airline Scheduling Problem and the Airline Operations
Control Problem (or Disruption Management Problem). We have described the AOCC
organization and roles as well as the typical problems that appear during the execution of
the operational plan. The disruption management process used by airlines was presented as
well as the main costs involved in generating and evaluating the solutions. In this section we
present our intelligent agent based approach to solve the Disruption Management Problem
in the airline domain. The MAS was developed using Java
1
and JADE (Bellifemine et al.,
2004) as the development platform and as the run-time environment that provides the basic
services for agents to execute.

4.1 Why an Agent and Multi-Agent System Paradigm?
Considering the agent and multi-agent system characteristics as specified in (Wooldridge,
2009) and (Elamy, 2005), the following ones make us adopt this paradigm to the Airline
Operations Control Problem:
Autonomy: MAS models problems in terms of autonomous interacting component-
agents, which are a more natural way of representing task allocation, team
planning, and user preferences, among others. In Figure 5 the PaxManager,
AircraftManager and CrewManager agents (among others) are agents that can choose
to respond or not to the requests according to their own objectives.
Agents are a Natural Metaphor: The AOCC is naturally modelled as a society of
agents cooperating with each other to solve such a complex problem.
Reactivity: Agents are able to perceive and react to the changes in their
environment. The Monitor agent in Figure 5 is an example of such an agent.
Distribution of resources: With a MAS we can distribute the computational resources
and capabilities across a network of interconnected agents avoiding problems
associated with centralized systems. Airline companies of some dimension have
different operational bases. We use a MAS for each operational base, taking
advantage of this important characteristic. Due to the social awareness characteristics
of some of our agents (for example, Monitoring agent in Figure 5) they are able to
distribute their tasks among other agents with similar behaviour.
Modularity and Scalability: A MAS is extensible, scalable, robust, maintainable,
flexible and promotes reuse. These characteristics are very important in systems of

1

http://www.java.com

4.2
It i
aft
(Ca
pro

Fig

Fig
age
gre
pro
age
im

this dimen
more oper
specific pr
Concurrenc
parallel. T
CrewCBR a
Additiona
responsibi
failures by
different c
Legacy Sys
systems. W
interact wi
2 MAS Architect
is important to p
ter performing a
astro & Oliveira
ocess and the bas
g. 5. MAS architec
gure 5 shows the
ents and the na
een lines represe
oblem-solving pr
ents and the nor
mportant to clarify
nsion and compl
rational bases a
roblems.
cy/Parallelism: Ag
This provides flex
and CrewHillClim
ally and accord
ilities are suffici
y one or more ag
computers. If one
stems: The AOCC
We can wrap the
ith other software
ure
point out that we
an analysis and
a, 2008). The age
se for this architec
cture
e architecture of
arrow black dash
ent the interaction
rocess. The narro
rmal black lines r
y that Figure 5 rep
lexity. Our MAS
s well as in sup
gents are capable
xibility and speed
mb agents in Figur
ding to (Stone
iently shared am
gents. Our MAS
or more agents f
C needs informati
legacy compone
e components.
e arrived to the a
design using an
ent model and se
cture.
our multi-agent
h lines represent
n between agents
ow gray lines rep
represent the inte
presents only one
is able to scale i
pporting differen
e of reasoning an
ds up computatio
re 5, are example
& Veloso, 20
mong agents, th
can be totally or
ail, the global obj
on that exists in
ents in an agent l
architecture of o
agent-oriented
ervice model we
system approach
t requests/propo
s regarding nego
present interaction
eractions after a
e instance of the
in terms of supp
nt algorithms to
nd performing ta
on. The CrewSimA
es of concurrent a
000) if contro
he system can to
r partially replica
jective is not affec
obsolete but func
layer, enabling th
ur multi-agent sy
software method
ere the outputs o

h. The boxes rep
osals made. The
otiation and distr
n within a hierar
solution is found
MAS. We can rep
porting
o solve
asks in
Anneal,
agents.
l and
olerate
ated in
cted.
ctional
hem to
ystem,
dology
of this
present
larger
ributed
rchy of
d. It is
plicate

almost all agents with the exception of the Supervisor agent because it is the one that
interacts with the human supervisor (an application domain restriction). Each one of the
agents Monitoring, PaxManager, AircraftManager, CrewManager and Supervisor has a specific
role in the AOCC. The Monitoring agent monitors the operational plan looking for events
that may represent any of the usual three problem dimensions, that is, aircraft, crew and/or
passenger problems. In case there are other instances of this agent, they recognize and
interact with each other, splitting the monitoring task. For example, if each instance
corresponds to an operational base, each one will monitor the corresponding operation plan.
This is one example of the social-awareness characteristic of our agents. The agent is
autonomous in the sense that it will consider an event as a problem only if the event has
certain characteristics.
The PaxManager agent has the responsibility to find solutions for passenger problems. The
AircraftManager and CrewManager agents have the responsibility for finding solutions for
aircraft and crew problems, respectively. These agents are autonomous in the sense that
they can choose not to respond to the information received from the Monitor agent, i.e., if the
problem is not related with their field of expertise or if they do not have local resources to
solve that problem. These agents have similar social-awareness characteristics of the Monitor
agent. Although not yet implemented, these agents may decide to participate with their
expertise in the integrated and distributed problem solving approach of the system.
The AircraftManager and CrewManager agents manage a team of specialized agents (Castro &
Oliveira, 2007). Each team should have several specialist agents, each one implementing a
different problem solving algorithm, making them heterogeneous regarding this
characteristic. The ACTabuSearch agent, ACCBR agent and ACHillClimb agent implements
algorithms dedicated to solve aircraft problems and present the candidate solutions they
find to the AircraftManager agent. The CrewSimAnneal agent, CrewHillClimb agent and
CrewCBR agent implements algorithms dedicated to solve crew problems and present the
candidate solutions to the CrewManager.
The agent Supervisor and agent EventType are the only ones that interact with a human user
of the AOCC. The Supervisor agent presents the solutions to the human supervisor, ranked
according to the criteria in use by the airline (more information on the next section),
including details about the solution to help the human to decide. After getting approval
from the human supervisor, the Supervisor agent requests ApplySolution agent to apply it on
the environment.
All agents are able to act and observe the environment that is represented by the Operational
and MAS database, in our diagram. The operational database includes information
regarding the flight, aircraft and crew schedule as well as airport and company specific
information. The other database is related with the learning characteristics of our system
and is used, mainly, by the Learning agent. The learning characteristics of our system are not
yet implemented. In Section 7, the interested reader can find more information about the
way we expect to apply learning in our MAS. Finally, the protocols we use are the following
FIPA compliant ones:
Fipa-Request: This protocol allows one agent to request another to perform some
action and the receiving agent to perform the action or reply, in some way, that it
cannot perform it. Fipa-request is used in interactions between the Monitor,
PaxManager, AircraftManager and CrewManager agents.

4.3
Ou
bet
coo
on
an
com
sec
spe
200
Fig

Fig
sim
sho

Fipa-Query
action on
AircraftMa
EventType
Fipa-Contr
role of ma
agents (th
characteriz
the intera
ACTabuSea
agents, i.e
how we us
3 Decision Mech
ur system uses n
tween PaxManag
operate so that a
ne that considers t
d passengers. As
mpletely implem
cond level is the
ecialist agents of
02) (Smith, 1980).
g. 6. Simplified co
gure 6 shows the
mplicity only the
own). After receiv
y: This protocol
another agent. I
anager, CrewMana
agent and, finally
act.net: In the c
anager which wis
he Participants)
zes the task(Fip
ctions that entai
arch, ACCBR an
., CrewSimAnneal
se this protocol is
hanisms
negotiation at two
ger, CrewManager
an integrated solu
the impact on the
s of the writing o
mented. Section 7
Specialist Agents
the team. At this
.
ontract net protoc
e simplified cont
e interaction betw
ving a request fro
allows one agen
It is used in the
ager and Supervis
y, EventType and
ontract net proto
shes to have som
and further w
a, 2002). We use
il the AircraftMa
nd ACHillClimb;
l, CrewHillClimb
s presented in the
o levels. The firs
r and AircraftMa
ution can be foun
e three dimensio
of this paper, we
7 gives a glimpse
or Team level, i.e
s level we have u
col
tract.net protocol
ween CrewManag
om the Monitoring
nt to request to
e interactions tha
sor agent; Superv
Monitoring agent
ocol, one agent (
me task performed
wishes to optim
a simplified ver
anager and its s
and CrewManag
and CrewCBR. M
e next section.
st level is the Ma
anager agents. At
nd. We define an
ns of the problem
e do not have th
e of how we are
., between each m
used a simplified

l applied to the C
ger and one of t
g agent and case
perform some k
at involve PaxMa
visor, ApplySolutio
t.
(the Initiator) tak
d by one or more
mize a function
sion of this proto
pecialized agent
ger and its speci
More information
anager Agents lev
t this level the
n integrated solut
m, that is, aircraft
his negotiation pr
e implementing i
manager agent an
fipa-contract.net
CrewManager tea
the specialist age
the CrewManager
kind of
anager,
on and
kes the
e other
n that
ocol in
ts, i.e.,
ialized
n about
el, i.e.,
agents
tion as
t, crew
rotocol
it. The
nd the
t (Fipa,
m (for
ents is
r agent

almost all agents with the exception of the Supervisor agent because it is the one that
interacts with the human supervisor (an application domain restriction). Each one of the
agents Monitoring, PaxManager, AircraftManager, CrewManager and Supervisor has a specific
role in the AOCC. The Monitoring agent monitors the operational plan looking for events
that may represent any of the usual three problem dimensions, that is, aircraft, crew and/or
passenger problems. In case there are other instances of this agent, they recognize and
interact with each other, splitting the monitoring task. For example, if each instance
corresponds to an operational base, each one will monitor the corresponding operation plan.
This is one example of the social-awareness characteristic of our agents. The agent is
autonomous in the sense that it will consider an event as a problem only if the event has
certain characteristics.
The PaxManager agent has the responsibility to find solutions for passenger problems. The
AircraftManager and CrewManager agents have the responsibility for finding solutions for
aircraft and crew problems, respectively. These agents are autonomous in the sense that
they can choose not to respond to the information received from the Monitor agent, i.e., if the
problem is not related with their field of expertise or if they do not have local resources to
solve that problem. These agents have similar social-awareness characteristics of the Monitor
agent. Although not yet implemented, these agents may decide to participate with their
expertise in the integrated and distributed problem solving approach of the system.
The AircraftManager and CrewManager agents manage a team of specialized agents (Castro &
Oliveira, 2007). Each team should have several specialist agents, each one implementing a
different problem solving algorithm, making them heterogeneous regarding this
characteristic. The ACTabuSearch agent, ACCBR agent and ACHillClimb agent implements
algorithms dedicated to solve aircraft problems and present the candidate solutions they
find to the AircraftManager agent. The CrewSimAnneal agent, CrewHillClimb agent and
CrewCBR agent implements algorithms dedicated to solve crew problems and present the
candidate solutions to the CrewManager.
The agent Supervisor and agent EventType are the only ones that interact with a human user
of the AOCC. The Supervisor agent presents the solutions to the human supervisor, ranked
according to the criteria in use by the airline (more information on the next section),
including details about the solution to help the human to decide. After getting approval
from the human supervisor, the Supervisor agent requests ApplySolution agent to apply it on
the environment.
All agents are able to act and observe the environment that is represented by the Operational
and MAS database, in our diagram. The operational database includes information
regarding the flight, aircraft and crew schedule as well as airport and company specific
information. The other database is related with the learning characteristics of our system
and is used, mainly, by the Learning agent. The learning characteristics of our system are not
yet implemented. In Section 7, the interested reader can find more information about the
way we expect to apply learning in our MAS. Finally, the protocols we use are the following
FIPA compliant ones:
Fipa-Request: This protocol allows one agent to request another to perform some
action and the receiving agent to perform the action or reply, in some way, that it
cannot perform it. Fipa-request is used in interactions between the Monitor,
PaxManager, AircraftManager and CrewManager agents.

4.3
Ou
bet
coo
on
an
com
sec
spe
200
Fig

Fig
sim
sho

Fipa-Query
action on
AircraftMa
EventType
Fipa-Contr
role of ma
agents (th
characteriz
the intera
ACTabuSea
agents, i.e
how we us
3 Decision Mech
ur system uses n
tween PaxManag
operate so that a
ne that considers t
d passengers. As
mpletely implem
cond level is the
ecialist agents of
02) (Smith, 1980).
g. 6. Simplified co
gure 6 shows the
mplicity only the
own). After receiv
y: This protocol
another agent. I
anager, CrewMana
agent and, finally
act.net: In the c
anager which wis
he Participants)
zes the task(Fip
ctions that entai
arch, ACCBR an
., CrewSimAnneal
se this protocol is
hanisms
negotiation at two
ger, CrewManager
an integrated solu
the impact on the
s of the writing o
mented. Section 7
Specialist Agents
the team. At this
.
ontract net protoc
e simplified cont
e interaction betw
ving a request fro
allows one agen
It is used in the
ager and Supervis
y, EventType and
ontract net proto
shes to have som
and further w
a, 2002). We use
il the AircraftMa
nd ACHillClimb;
l, CrewHillClimb
s presented in the
o levels. The firs
r and AircraftMa
ution can be foun
e three dimensio
of this paper, we
7 gives a glimpse
or Team level, i.e
s level we have u
col
tract.net protocol
ween CrewManag
om the Monitoring
nt to request to
e interactions tha
sor agent; Superv
Monitoring agent
ocol, one agent (
me task performed
wishes to optim
a simplified ver
anager and its s
and CrewManag
and CrewCBR. M
e next section.
st level is the Ma
anager agents. At
nd. We define an
ns of the problem
e do not have th
e of how we are
., between each m
used a simplified

l applied to the C
ger and one of t
g agent and case
perform some k
at involve PaxMa
visor, ApplySolutio
t.
(the Initiator) tak
d by one or more
mize a function
sion of this proto
pecialized agent
ger and its speci
More information
anager Agents lev
t this level the
n integrated solut
m, that is, aircraft
his negotiation pr
e implementing i
manager agent an
fipa-contract.net
CrewManager tea
the specialist age
the CrewManager
kind of
anager,
on and
kes the
e other
n that
ocol in
ts, i.e.,
ialized
n about
el, i.e.,
agents
tion as
t, crew
rotocol
it. The
nd the
t (Fipa,
m (for
ents is
r agent

decides to reply, a Call for Proposal (cfp) is issued to initiate the negotiation process. Table 1
shows an example of a message sent in this step.

(cfp
:sender (agent-identifier :name CrewManager@masdima:1099/JADE
:addresses (sequence http://masdima:7778/acc))
:receiver (set (agent-identifier :name CrewSimAnneal@masdima:1099/JADE
:addresses (sequence http://masdima:7778/acc)))
:X-JADE-Encoding Base64
:content "ABXN rO0AAAA eHB4h3CAAAAAsAAAAAeHB4"
:language fipa-sl
:conversation-id cfp_crew_solution)
Table 1. CFP message sent by CrewManager agent

Please note that the content of the FIPA-ACL message is a serialized Java object (see Table
2), that contains the event description, as well as the deadline for receiving an answer
(propose or refuse) and the deadline for receiving the candidate solution (i.e., the
CrewSimAnneal agent needs to send a candidate solution before a specific period of time).

import java.io.Serializable;
import java.util.ArrayList;
import java.util.GregorianCalendar;

public class CrewProblem implements Serializable {
private ArrayList<Event> events;
private int numSeconds;
private int maxCost;
private int numMinutesTimeWindow;
private GregorianCalendar bidDeadline;
private GregorianCalendar candSolutionDeadline;
public ArrayList<Event> getEvents() {return events;}
()
public int getMaxCost() {return maxCost;}
public void setMaxCost(int maxCost) {this.maxCost = maxCost;}
public GregorianCalendar getBidDeadline() {return bidDeadline;}
public void setBidDeadline(GregorianCalendar bidDeadline) {this.bidDeadline = bidDeadline;}
public GregorianCalendar getCandSolutionDeadline() {return candSolutionDeadline;}
public void setCandSolutionDeadline(GregorianCalendar candSolutionDeadline) {
this.candSolutionDeadline = candSolutionDeadline;}
public CrewProblem(ArrayList<Event> events, int numSeconds, int maxCost, int numMinutesTimeWindow,
GregorianCalendar bidDeadline, GregorianCalendar candSolutionDeadline)
{
this.maxCost = maxCost;
this.events = events;
this.numSeconds = numSeconds;
this.numMinutesTimeWindow = numMinutesTimeWindow;
this.bidDeadline = bidDeadline;
this.candSolutionDeadline = candSolutionDeadline;
}
}
Table 2. Partial example of a Serialized Java object included in the CFP message

The CrewSimAnneal agent may choose to answer refuse or propose. In our approach the
CrewSimAnneal propose performative only means that it will look for a candidate solution
according to the conditions of the cfp. The CrewManager agent will automatically answer
back with an accept-proposal. Here we simplified the contract.net protocol to speed-up the
communication between our agents. In our case, the answer we get from specialist agents is
a simple yes or no, because we want all available agents (i.e., that are not busy looking for
candidate-solutions for other requests) to work in parallel to find candidate solutions.
Be
pro
pe
me
res
sol

Ta

Ta

At
sol
To
Per
Th
Qu

Co

cause of that we
oblem during th
rformative statin
essage content. If
sult performative
lution.
(in
:

:

:
:
:
:
able 3. Example of
im
im

pu
p
p
p
p

p
p
p
p
p
p
p
p
p
ini
{

}
}
able 4. Serialized J
t the team level,
lutions that were
tal Operational Co
rformance, are bei
he Total Operation
uality Operational
oefficient is used
e do not need to
he execution of
ng the reasons for
the agent is able
e (Table 3) that in
nform
:sender (agent-identifier :n
:addresses (sequence ht
:receiver (set (agent-ident
:content "eHB4h3CAAAA
:language fipa-sl
:conversation-id cfp_crew
f a Failure and In
mport java.io.Serializable;
mport java.util.ArrayList;
ublic class CrewSolution im
private int cost;
private int initialCost;
private String description
private ArrayList<Flight>
public int getCost() {retur
public void setCost(int co
public int getInitialCost()
public void setInitialCost(
public String getDescripti
public void setDescription
public ArrayList<Flight>
public void setSolution(A
public CrewSolution(Arr
tialCost)
{
setCost(cost);
setInitialCost(initialCos
setDescription(descript
setSolution(solution);
}
Java object includ
the manager age
found by the spe
ost as the only cr
ing tested but we
al Cost (tc) of a sp
Costs (qc) and is g
d to define the we
o choose between
the task, the
r the failure, in th
to perform the ta
ncludes the serial
name CrewSimAnneal @m
ttp://masdima:7778/acc)
ifier :name CrewManager
4
AAsAAAAAeHABXNrO0
w_solution)
nform message
mplements Serializable {
n;
> solution;
rn cost;}
ost) {this.cost = cost;}
{return initialCost;}
(int initialCost) {this.initia
ion() {return description;}
n(String description) {this
getSolution() {return solu
ArrayList<Flight> solution
rayList<Flight> solution,
st);
tion);
ded in the Inform
ent needs to sele
ecialist agents. As
riteria for the sele
do not have any
pecific solution in
given by Equation
eight of quality c
n all the answers
CrewSimAnneal a
he serialized Java
ask with success,
ized object (Tabl
masdima:1099/JADE
))
r@masdima:1099/JADE
)))
0AAAAB4"
alCost = initialCost;}

s.description = description
ution;}
n) {this.solution = solution
String description, int
m Message
ct the best soluti
s of the writing o
ection. Other crit
results at this mo
ncludes Direct Op
n 1.

osts.
s received. If the
agent issues a
a object included
it will issue an in
e 4) with the can
n;}
n;}
cost, int
ion from the can
f this paper, we u
teria, like AOCC
oment.
perational Costs (d
re is a
failure
in the
nform-
ndidate
ndidate
use the
Global
dc) and
(1)

decides to reply, a Call for Proposal (cfp) is issued to initiate the negotiation process. Table 1
shows an example of a message sent in this step.

(cfp
:sender (agent-identifier :name CrewManager@masdima:1099/JADE
:addresses (sequence http://masdima:7778/acc))
:receiver (set (agent-identifier :name CrewSimAnneal@masdima:1099/JADE
:addresses (sequence http://masdima:7778/acc)))
:content "ABXN rO0AAAA eHB4h3CAAAAAsAAAAAeHB4"
:language fipa-sl
:conversation-id cfp_crew_solution)
Table 1. CFP message sent by CrewManager agent

Please note that the content of the FIPA-ACL message is a serialized Java object (see Table
2), that contains the event description, as well as the deadline for receiving an answer
(propose or refuse) and the deadline for receiving the candidate solution (i.e., the
CrewSimAnneal agent needs to send a candidate solution before a specific period of time).

import java.io.Serializable;
import java.util.ArrayList;
import java.util.GregorianCalendar;

public class CrewProblem implements Serializable {
private ArrayList<Event> events;
private int numSeconds;
private int maxCost;
private int numMinutesTimeWindow;
private GregorianCalendar bidDeadline;
private GregorianCalendar candSolutionDeadline;
public ArrayList<Event> getEvents() {return events;}
()
public int getMaxCost() {return maxCost;}
public void setMaxCost(int maxCost) {this.maxCost = maxCost;}
public GregorianCalendar getBidDeadline() {return bidDeadline;}
public void setBidDeadline(GregorianCalendar bidDeadline) {this.bidDeadline = bidDeadline;}
public GregorianCalendar getCandSolutionDeadline() {return candSolutionDeadline;}
public void setCandSolutionDeadline(GregorianCalendar candSolutionDeadline) {
this.candSolutionDeadline = candSolutionDeadline;}
public CrewProblem(ArrayList<Event> events, int numSeconds, int maxCost, int numMinutesTimeWindow,
GregorianCalendar bidDeadline, GregorianCalendar candSolutionDeadline)
{
this.maxCost = maxCost;
this.events = events;
this.numSeconds = numSeconds;
this.numMinutesTimeWindow = numMinutesTimeWindow;
this.bidDeadline = bidDeadline;
this.candSolutionDeadline = candSolutionDeadline;
}
}
Table 2. Partial example of a Serialized Java object included in the CFP message

The CrewSimAnneal agent may choose to answer refuse or propose. In our approach the
CrewSimAnneal propose performative only means that it will look for a candidate solution
according to the conditions of the cfp. The CrewManager agent will automatically answer
back with an accept-proposal. Here we simplified the contract.net protocol to speed-up the
communication between our agents. In our case, the answer we get from specialist agents is
a simple yes or no, because we want all available agents (i.e., that are not busy looking for
candidate-solutions for other requests) to work in parallel to find candidate solutions.
Be
pro
pe
me
res
sol

Ta

Ta

At
sol
To
Per
Th
Qu

Co

cause of that we
oblem during th
rformative statin
essage content. If
sult performative
lution.
(in
:

:

:
:
:
:
able 3. Example of
im
im

pu
p
p
p
p

p
p
p
p
p
p
p
p
p
ini
{

}
}
able 4. Serialized J
t the team level,
lutions that were
tal Operational Co
rformance, are bei
he Total Operation
uality Operational
oefficient is used
e do not need to
he execution of
ng the reasons for
the agent is able
e (Table 3) that in
nform
:sender (agent-identifier :n
:receiver (set (agent-ident
:content "eHB4h3CAAAA
:language fipa-sl
:conversation-id cfp_crew
f a Failure and In
mport java.io.Serializable;
mport java.util.ArrayList;
ublic class CrewSolution im
private int cost;
private int initialCost;
private String description
private ArrayList<Flight>
public int getCost() {retur
public void setCost(int co
public int getInitialCost()
public void setInitialCost(
public String getDescripti
public void setDescription
public ArrayList<Flight>
public void setSolution(A
public CrewSolution(Arr
tialCost)
{
setCost(cost);
setInitialCost(initialCos
setDescription(descript
setSolution(solution);
}
Java object includ
the manager age
found by the spe
ost as the only cr
ing tested but we
al Cost (tc) of a sp
Costs (qc) and is g
d to define the we
o choose between
the task, the
r the failure, in th
to perform the ta
ncludes the serial
name CrewSimAnneal @m
ifier :name CrewManager
4
AAsAAAAAeHABXNrO0
w_solution)
nform message
mplements Serializable {
n;
> solution;
rn cost;}
ost) {this.cost = cost;}
{return initialCost;}
(int initialCost) {this.initia
ion() {return description;}
n(String description) {this
getSolution() {return solu
ArrayList<Flight> solution
rayList<Flight> solution,
st);
tion);
ded in the Inform
ent needs to sele
ecialist agents. As
riteria for the sele
do not have any
pecific solution in
given by Equation
eight of quality c
n all the answers
CrewSimAnneal a
he serialized Java
ask with success,
ized object (Tabl
masdima:1099/JADE
))
r@masdima:1099/JADE
)))
0AAAAB4"
alCost = initialCost;}

s.description = description
ution;}
n) {this.solution = solution
String description, int
m Message
ct the best soluti
s of the writing o
ection. Other crit
results at this mo
ncludes Direct Op
n 1.

osts.
s received. If the
agent issues a
a object included
it will issue an in
e 4) with the can
n;}
n;}
cost, int
ion from the can
f this paper, we u
teria, like AOCC
oment.
perational Costs (d
re is a
failure
in the
nform-
ndidate
ndidate
use the
Global
dc) and
(1)

Di
rel
Pas

Th
ad
cos
by

Th
by
dis
the
con
ad

Th
the
an
of

Qu
an
im
Eq
Ol

rect Operational C
lated with the o
ssenger Costs (pc).
he Crew Cost (cc) f
ditional work ho
sts (Hotel) and ex
y Equation 3.
he Flight Cost (fc)
y the airport ope
spatch, line main
e type of aircraft
nsumption (Fuel)
ditional extra fue
he Passenger Cost
e airline has to su
d any compensat
the delayed pass
uality Operational
d are related w
mportant to increa
quation 6 present
iveira, 2009).
Costs (dc) of a spec
operation of the
. It is given by Eq
for a specific fligh
ours to be paid (H
xtra-crew travel c
for a specific flig
erator like appr
ntenance, cleaning
t that performs t
), i.e., fuel to go
el required. The F
(pc) of the delaye
upport when a fli
tion to the passen
engers for a spec
Costs (qc) of a sp
with passenger s
ase the quality l
ts a generic expre
cific solution are c
flights, namely,
quation 2.

ht includes the sa
Hour), additional p
costs (Dhc). The C
ght includes the a
oaching and tax
g services and oth
the flight (Maint)
from the origin
Flight Cost for a sp
ed passengers for
ight is delayed o
ngers according to
ific solution is giv
pecific solution a
satisfaction. The
level of an airlin
ession that calcu
costs that are eas
, Crew Costs (cc)
alary costs of all c
perdiem days to b
Crew Cost for a sp
airport costs (Airp
xing; service cos
her costs; average
; ATC en-route c
to the destinatio
pecific solution is

r a specific flight
r cancelled (Meal
o regulations (Co
ven by Equation

are costs that are
e quantification
ne company whe
ulates this value a
ily quantifiable a
), Flight Costs (fc
crew members (S
be paid (Perdiem)
pecific solution is

p), i.e., charges a
sts (Service), i.e.,
e maintenance co
charges (Atc); an
on (trip fuel) plu
given by Equatio
includes airport
ls), hotels costs (P
omp). The Passeng
5.
not easily quant
of this value is
en facing a disru
according to (Ca
and are
fc) and
(2)
Salary),
), hotel
s given
(3)
applied
flight
osts for
nd fuel
us any
on 4.
(4)
t meals
PHotel)
ger Cost
(5)
tifiable
s very
uption.
astro &

(6)

Now that we know the main costs involved, it is time to understand how each manager
agent selects the best (or the best x candidate solutions). Once the participant agent has
completed the task (for example, agent CrewHillClimb in Figure 5), it sends a completion
message to the initiator (agent CrewManager in Figure 5) in the form of an inform-result
performative (Table 3), with the details of the candidate solution (Table 4) including the Total
Operational Cost. The manager agent sorts, in ascending order, all candidate solutions
received by total operational cost. The top three solutions are selected (Castro & Oliveira,
2007).

4.4 Problem Solving Algorithms
As it is possible to see in Figure 5 (Section 4.2), the aircraft and crew dimension have, each
one, a team of specialist agents. Each agent should implement a heterogeneous problem
solving algorithm on the team they belong to. Preliminary results show that a single
problem solving algorithm is not able to solve, dynamically and within the required time
restriction, all types of problems that we have identified during our observations (see
Section 3.3). Taking advantage of the modularity, scalability and distributed characteristics
of the MAS paradigm, we are able to add as many specialist agents as required, so that all
types of problems are covered. As we have seen in Section 4.2 and 4.3, the idea is to have all
specialist agents of a team looking for solutions concurrently.
In this section we are going to show how we have implemented one of the specialist agents
of the crew team, namely, CrewHillClimb. This agent implements a hill climb algorithm. For
more details regarding how we have implemented this and other specialist agents, please
read (Mota, 2007).

The hill climbing agent solves the problem iteratively by following the steps:

1. Obtains the flights that are in the time window of the problem. This time window
starts at the flight date, and ends at a customizable period in the future. This will be
the initial solution of the problem. The crew members exchanges are made
between flights that are inside the time window of the problem.
2. While some specific and customizable time has not yet passed, or a solution below
a specific and customizable cost has not been found, repeats steps 3 and 4.
3. Generates the successor of the initial solution (the way a successor is generated is
described below).
4. Evaluates the cost of the solution. If it is smaller than the cost of the current
solution, accepts the generated solution as the new current solution. Otherwise,
discards the generated solution. The way a solution is evaluated is described
below.

Di
rel
Pas

Th
ad
cos
by

Th
by
dis
the
con
ad

Th
the
an
of

Qu
an
im
Eq
Ol

rect Operational C
lated with the o
ssenger Costs (pc).
he Crew Cost (cc) f
ditional work ho
sts (Hotel) and ex
y Equation 3.
he Flight Cost (fc)
y the airport ope
spatch, line main
e type of aircraft
nsumption (Fuel)
ditional extra fue
he Passenger Cost
e airline has to su
d any compensat
the delayed pass
uality Operational
d are related w
mportant to increa
quation 6 present
iveira, 2009).
Costs (dc) of a spec
operation of the
. It is given by Eq
for a specific fligh
ours to be paid (H
xtra-crew travel c
for a specific flig
erator like appr
ntenance, cleaning
t that performs t
), i.e., fuel to go
el required. The F
(pc) of the delaye
upport when a fli
tion to the passen
engers for a spec
Costs (qc) of a sp
with passenger s
ase the quality l
ts a generic expre
cific solution are c
flights, namely,
quation 2.

ht includes the sa
Hour), additional p
costs (Dhc). The C
ght includes the a
oaching and tax
g services and oth
the flight (Maint)
from the origin
Flight Cost for a sp
ed passengers for
ight is delayed o
ngers according to
ific solution is giv
pecific solution a
satisfaction. The
level of an airlin
ession that calcu
costs that are eas
, Crew Costs (cc)
alary costs of all c
perdiem days to b
Crew Cost for a sp
airport costs (Airp
xing; service cos
her costs; average
; ATC en-route c
to the destinatio
pecific solution is

r a specific flight
r cancelled (Meal
o regulations (Co
ven by Equation

are costs that are
e quantification
ne company whe
ulates this value a
ily quantifiable a
), Flight Costs (fc
crew members (S
be paid (Perdiem)
pecific solution is

p), i.e., charges a
sts (Service), i.e.,
e maintenance co
charges (Atc); an
on (trip fuel) plu
given by Equatio
includes airport
ls), hotels costs (P
omp). The Passeng
5.
not easily quant
of this value is
en facing a disru
according to (Ca
and are
fc) and
(2)
Salary),
), hotel
s given
(3)
applied
flight
osts for
nd fuel
us any
on 4.
(4)
t meals
PHotel)
ger Cost
(5)
tifiable
s very
uption.
astro &

(6)

Now that we know the main costs involved, it is time to understand how each manager
agent selects the best (or the best x candidate solutions). Once the participant agent has
completed the task (for example, agent CrewHillClimb in Figure 5), it sends a completion
message to the initiator (agent CrewManager in Figure 5) in the form of an inform-result
performative (Table 3), with the details of the candidate solution (Table 4) including the Total
Operational Cost. The manager agent sorts, in ascending order, all candidate solutions
received by total operational cost. The top three solutions are selected (Castro & Oliveira,
2007).

4.4 Problem Solving Algorithms
As it is possible to see in Figure 5 (Section 4.2), the aircraft and crew dimension have, each
one, a team of specialist agents. Each agent should implement a heterogeneous problem
solving algorithm on the team they belong to. Preliminary results show that a single
problem solving algorithm is not able to solve, dynamically and within the required time
restriction, all types of problems that we have identified during our observations (see
Section 3.3). Taking advantage of the modularity, scalability and distributed characteristics
of the MAS paradigm, we are able to add as many specialist agents as required, so that all
types of problems are covered. As we have seen in Section 4.2 and 4.3, the idea is to have all
specialist agents of a team looking for solutions concurrently.
In this section we are going to show how we have implemented one of the specialist agents
of the crew team, namely, CrewHillClimb. This agent implements a hill climb algorithm. For
more details regarding how we have implemented this and other specialist agents, please
read (Mota, 2007).

The hill climbing agent solves the problem iteratively by following the steps:

1. Obtains the flights that are in the time window of the problem. This time window
starts at the flight date, and ends at a customizable period in the future. This will be
the initial solution of the problem. The crew members exchanges are made
between flights that are inside the time window of the problem.
2. While some specific and customizable time has not yet passed, or a solution below
a specific and customizable cost has not been found, repeats steps 3 and 4.
3. Generates the successor of the initial solution (the way a successor is generated is
described below).
4. Evaluates the cost of the solution. If it is smaller than the cost of the current
solution, accepts the generated solution as the new current solution. Otherwise,
discards the generated solution. The way a solution is evaluated is described
below.

Th
cur
fol

Wh
(2)
cho
pro
tha
ele
If t
an
flig
cho
ele
ele
du
flig
Th
typ

Th

Ta
5. Send the c
have seen
he generation of a
rrent solution by
llowing operation
Swap two
time wind
Swap a cr
window of
hen choosing the
) choose an elem
oosing an elemen
oceed faster to g
at is not late prob
ement probably w
the decision is to
d the first delaye
ght is picked, and
oosing a crewm
ement that is goi
ements of flights
uty. The choice is
ghts, which was g
he evaluation of t
pes of costs:
The crew c
The penali
The penali
highest, be
he Hill Climbing O
able 5 shows the im
GregorianCalen
int secondsExec
while(!Shared.t
{
// get succes
current solution
in Section 4.3.
a new solution is
y one unit, that is
ns:
crewmembers b
dow of the problem
rewmember of a
f the problem wit
e first element to s
ment that is dela
nt that is late, wh
good solutions (e
bably wont reduc
wont compensate
o exchange an ele
ed element is chos
d a crewmember
member, which w
ng to swap with
or (2) swap betw
made based on
given a value of 0
the solution is do
cost according to
ization for exchan
ization for delay
ecause the goal is
Objective Function
mplementation o
ndar currentDate = new G
cution = (int) ((currentDa
to(problem.getNumSecon
ssor
to the CrewMana
s made by finding
s, the successor is
between flights th
m.
a flight that belo
th a crewmember
swap, there are tw
ayed. The choice
hich was given a
xchanges are hig
ce the cost, as a p
e the penalization
ement that is dela
sen. If the decisio
or the aircraft is c
was given a valu
h the first, there a
ween an element o
the probability o
0.5.
one by an objectiv
Equation 3;
nging elements;
yed elements. The
s to have no delay
(hc) is given by E
of the hill climbin
GregorianCalendar();
ate.getTimeInMillis() - star
nds(), secondsExecution, p
ager agent follow
g a successor tha
s obtained by on
hat belong to the
ongs to the flight
r that isnt on dut
wo possibilities:
e is made based
value of 0.9, so t
ghly penalized, s
possible saving by
n associated with
ayed, the list of fl
on is to choose ran
chosen, dependin
ue of 0.85. When
are two possibili
of a flight and an
of choosing a swa
ve function that m
e cost associated
yed elements.
Equation 7.

g algorithm in Jav
rtDateResolution.getTime
problem.getMaxCost(), cu
wing the protocol
at distances itself
ne, and only one,
e flights that are
ts that are in th
ty, but is on stand
(1) choose random
on the probabi
that the algorithm
so choosing an el
y choosing a less
the exchange).
lights will be exa
ndomly, then a ra
ng on the probab
n choosing the s
ities: (1) swap be
n element that is
ap between eleme
measures the foll
with this aspect
va.
eInMillis()) / 1000);
urrentSolutionCost))
as we
to the
of the
in the
e time
dby.
mly or
ility of
ms can
lement
costly
amined
andom
ility of
second
etween
not on
ents of
lowing
t is the
(7)

successor = Shared.generateSuccessor(Shared.copyArrayList(currentSolution));
// checks if successor has an inferior solution cost
successorCost = Shared.calculateCost(successor, initialPlainSolution);
System.out.println("Successor Cost: " + successorCost + "\n");
if(sucessorCost < currentSolutionCost)
{
currentSolution = successor;
currentSolutionCost = successorCost;
}
currentDate = new GregorianCalendar();
secondsExecution = (int) ((currentDate.getTimeInMillis() - startDateResolution.getTimeInMillis()) / 1000);
}
Table 5. Implementation of the Hill-Climbing algorithm in Java

5. Experimental Setup

To evaluate our approach we have setup a scenario that includes 3 operational bases (A, B
and C). Each base includes their crewmembers each one with a specific roster. The data used
corresponds to a real airline operation of June 2006 of base A. We have simulated a situation
where 15 crewmembers, with different ranks, did not report for duty in base A. A
description of the information collected for each event is presented in Table 6.

Attribute Description
Event ID A number that represents the ID of the event. For tracking purposes only
Duty Date Time The start date and time of the duty in UTC for which the crew did not report.
Duty ID A string that represents the ID of the duty for which the crew did not report.
Flt Dly Flight delay in minutes
C Pax Number of passengers in business class
Y Pax Number of passengers in economy class
End Date Time The end date and time of the duty in UTC for which the crew did not report.
Ready Date Time The date and time at which the crew member is ready for another duty after this one.
Delay The delay of the crewmember. We have considered 10 minutes in our scenario.
Credit Minutes The minutes of this duty that will count for payroll.
Crew Group The crew group (Technical = 1; Cabin = 2) that the crewmember belongs to.
Crew Rank CPT = Captain; OPT = First Officer; CCB = Chief Purser; CAB = Purser.
Crew Number The employee number.
Crew Name The employee name.
Base ID The base where the event happened. We considered all events in base A.
Open Positions The number of missing crews for this duty and rank. We used a fixed number of 1.
Table 6. Description of the information collected for each event

The events did not happen at the same day and each one corresponds to a crewmember that
did not report for a specific duty in a specific day. Table 7 shows the data for each of the
events created. As you can see we have omitted the information regarding Delay, Base ID
and Open Positions because we have used fixed values as indicated in Table 6. For example,
the event 10 corresponds to the following situation: Crew Peter B, with number 32 and rank
CPT (captain) belonging to the crew group 1 (technical crew), did not report for the duty
with ID 1ZRH12X with briefing time (duty date time) at 15:25 in 15-06-2006. This flight did
not delay on departure and has 5 passengers in business class and 115 in economy class. The
event was created after a 10 minutes delay of the crewmember in reporting for duty and
happened at base A. It is necessary to find another crewmember to be assigned to this duty.
The duty ends at 09:30 on 17-06-2006 and the crewmember assigned to this duty will be
ready for another one at 21:30 in 17-06-2006. The duty will contribute with 1318 minutes
(21h58) for the payroll. The new crewmember must belong to the same rank and group.

Th
cur
fol

Wh
(2)
cho
pro
tha
ele
If t
an
flig
cho
ele
ele
du
flig
Th
typ

Th

Ta
5. Send the c
have seen
he generation of a
rrent solution by
llowing operation
Swap two
time wind
Swap a cr
window of
hen choosing the
) choose an elem
oosing an elemen
oceed faster to g
at is not late prob
ement probably w
the decision is to
d the first delaye
ght is picked, and
oosing a crewm
ement that is goi
ements of flights
uty. The choice is
ghts, which was g
he evaluation of t
pes of costs:
The crew c
The penali
The penali
highest, be
he Hill Climbing O
able 5 shows the im
GregorianCalen
int secondsExec
while(!Shared.t
{
// get succes
current solution
in Section 4.3.
a new solution is
y one unit, that is
ns:
crewmembers b
dow of the problem
rewmember of a
f the problem wit
e first element to s
ment that is dela
nt that is late, wh
good solutions (e
bably wont reduc
wont compensate
o exchange an ele
ed element is chos
d a crewmember
member, which w
ng to swap with
or (2) swap betw
made based on
given a value of 0
the solution is do
cost according to
ization for exchan
ization for delay
ecause the goal is
Objective Function
mplementation o
ndar currentDate = new G
cution = (int) ((currentDa
to(problem.getNumSecon
ssor
to the CrewMana
s made by finding
s, the successor is
between flights th
m.
a flight that belo
th a crewmember
swap, there are tw
ayed. The choice
hich was given a
xchanges are hig
ce the cost, as a p
e the penalization
ement that is dela
sen. If the decisio
or the aircraft is c
was given a valu
h the first, there a
ween an element o
the probability o
0.5.
one by an objectiv
Equation 3;
nging elements;
yed elements. The
s to have no delay
(hc) is given by E
of the hill climbin
GregorianCalendar();
ate.getTimeInMillis() - star
nds(), secondsExecution, p
ager agent follow
g a successor tha
s obtained by on
hat belong to the
ongs to the flight
r that isnt on dut
wo possibilities:
e is made based
value of 0.9, so t
ghly penalized, s
possible saving by
n associated with
ayed, the list of fl
on is to choose ran
chosen, dependin
ue of 0.85. When
are two possibili
of a flight and an
of choosing a swa
ve function that m
e cost associated
yed elements.
Equation 7.

g algorithm in Jav
rtDateResolution.getTime
problem.getMaxCost(), cu
wing the protocol
at distances itself
ne, and only one,
e flights that are
ts that are in th
ty, but is on stand
(1) choose random
on the probabi
that the algorithm
so choosing an el
y choosing a less
the exchange).
lights will be exa
ndomly, then a ra
ng on the probab
n choosing the s
ities: (1) swap be
n element that is
ap between eleme
measures the foll
with this aspect
va.
eInMillis()) / 1000);
urrentSolutionCost))
as we
to the
of the
in the
e time
dby.
mly or
ility of
ms can
lement
costly
amined
andom
ility of
second
etween
not on
ents of
lowing
t is the
(7)

successor = Shared.generateSuccessor(Shared.copyArrayList(currentSolution));
// checks if successor has an inferior solution cost
successorCost = Shared.calculateCost(successor, initialPlainSolution);
System.out.println("Successor Cost: " + successorCost + "\n");
if(sucessorCost < currentSolutionCost)
{
currentSolution = successor;
currentSolutionCost = successorCost;
}
currentDate = new GregorianCalendar();
secondsExecution = (int) ((currentDate.getTimeInMillis() - startDateResolution.getTimeInMillis()) / 1000);
}
Table 5. Implementation of the Hill-Climbing algorithm in Java

5. Experimental Setup

To evaluate our approach we have setup a scenario that includes 3 operational bases (A, B
and C). Each base includes their crewmembers each one with a specific roster. The data used
corresponds to a real airline operation of June 2006 of base A. We have simulated a situation
where 15 crewmembers, with different ranks, did not report for duty in base A. A
description of the information collected for each event is presented in Table 6.

Attribute Description
Event ID A number that represents the ID of the event. For tracking purposes only
Duty Date Time The start date and time of the duty in UTC for which the crew did not report.
Duty ID A string that represents the ID of the duty for which the crew did not report.
Flt Dly Flight delay in minutes
C Pax Number of passengers in business class
Y Pax Number of passengers in economy class
End Date Time The end date and time of the duty in UTC for which the crew did not report.
Ready Date Time The date and time at which the crew member is ready for another duty after this one.
Delay The delay of the crewmember. We have considered 10 minutes in our scenario.
Credit Minutes The minutes of this duty that will count for payroll.
Crew Group The crew group (Technical = 1; Cabin = 2) that the crewmember belongs to.
Crew Rank CPT = Captain; OPT = First Officer; CCB = Chief Purser; CAB = Purser.
Crew Number The employee number.
Crew Name The employee name.
Base ID The base where the event happened. We considered all events in base A.
Open Positions The number of missing crews for this duty and rank. We used a fixed number of 1.
Table 6. Description of the information collected for each event

The events did not happen at the same day and each one corresponds to a crewmember that
did not report for a specific duty in a specific day. Table 7 shows the data for each of the
events created. As you can see we have omitted the information regarding Delay, Base ID
and Open Positions because we have used fixed values as indicated in Table 6. For example,
the event 10 corresponds to the following situation: Crew Peter B, with number 32 and rank
CPT (captain) belonging to the crew group 1 (technical crew), did not report for the duty
with ID 1ZRH12X with briefing time (duty date time) at 15:25 in 15-06-2006. This flight did
not delay on departure and has 5 passengers in business class and 115 in economy class. The
event was created after a 10 minutes delay of the crewmember in reporting for duty and
happened at base A. It is necessary to find another crewmember to be assigned to this duty.
The duty ends at 09:30 on 17-06-2006 and the crewmember assigned to this duty will be
ready for another one at 21:30 in 17-06-2006. The duty will contribute with 1318 minutes
(21h58) for the payroll. The new crewmember must belong to the same rank and group.

After setting-up the scenario we found the solutions for each crew event using three
methods.

Duty
DateTime
DutyID Flt
Dly
C
Pax
Y
Pax
End
DateTime
Ready
DateTime
Cred
Min
Crew
Grp
Rnk Crw
Nr
Crew
Name
1 05-06 07:25 1ORY149S 0 7 123 05-06 13:35 06-06 01:35 370 2 CAB 80 John A
2 05-06 07:25 1ORY149S 10 11 114 05-06 13:35 06-06 01:35 370 2 CAB 45 Mary A
3 05-06 07:25 1ORY85P 0 10 112 05-06 13:35 06-06 01:35 370 1 CPT 35 Anthony
4 15-06 04:10 2LIS24X 30 0 90 16-06 16:15 17-06 04:15 1757 2 CAB 99 Paul M
5 15-06 04:10 3LIS25X 25 3 77 15-06 09:20 15-06 21:20 632 2 CAB 56 John B
6 15-06 12:50 2LHR63P 5 25 85 16-06 20:45 17-06 08:45 1549 1 CPT 57 Paul S
7 15-06 12:50 2LHR63P 0 20 95 16-06 20:45 17-06 08:45 1549 1 OPT 53 Mary S
8 15-06 14:15 1LHR31P 0 23 52 15-06 20:55 16-06 08:55 843 2 CCB 23 Sophie
9 15-06 15:25 2LHR19P 10 27 105 16-06 20:45 17-06 08:45 1341 2 CCB 34 Angel
10 15-06 15:25 1ZRH12X 0 5 115 17-06 09:30 17-06 21:30 1318 1 CPT 32 Peter B
11 25-06 05:20 1LIS16S 20 3 97 25-06 15:05 26-06 03:05 585 2 CAB 20 Paul G
12 25-06 05:20 1LIS16S 5 2 108 25-06 15:05 26-06 03:05 585 2 CAB 10 Alice
13 25-06 05:20 1LIS158T 0 4 92 25-06 15:05 26-06 03:05 585 2 CAB 15 Daniel
14 25-06 06:15 3LIS174S 0 1 129 27-06 16:15 28-06 04:15 1258 2 CAB 71 George
15 25-06 14:20 4LIS50A 0 2 83 28-06 19:40 29-06 07:40 219 1 OPT 65 Allan
Table 7. Events data used for testing

In the first method we used one of the best users from the AOCC, with current tools
available, to find the solutions. The user uses software that shows the roster of each
crewmember in a Gantt chart for a specific period. The user can scroll down the information,
filter according to the crew rank and base, and sort the information by name, month duty,
etc. Each user has a specific way of trying to find the solutions. However, we have observed
that, in general, they follow these steps:
1. Open the roster for a one month period, starting two days before the current day.
For example, lets suppose that the current day is 7th of June of 2006, they open the
roster from the 5th of June until the 4th of July.
2. Filter the roster by crew rank and base, where the base is equal to the base where
the crew event happened and crew rank is equal to the crewmember rank that did
not report for duty.
3. Order the information by month duty, in an ascendant order and by seniority in a
descendent order.
4. Visually, they scroll down the information until they found a crewmember with an
open space for the period of time that corresponds to the duty to be assigned. This
period of time takes into consideration the start and end time of the duty and also
the time required for resting (ready date time).
5. If they do not found a crewmember in the base specified, they try to find it in
another base, filtering the information accordingly.
6. They assign the duty to the crewmember with less credit hours.
The data collected using this method is presented in Table 8. We point out that the data in
columns marked with an asterisk where calculated manually, according to the equations
presented in chapter 4.3. The reason for this is that the information system that is available
for the users does not include information related with any kind of costs.

Duty ID Base
ID
Crew
Grp
Rank Hour Pay (*) Perdiem
Pay (*)
Quality
Op. Cost
Op. Cost
(*)
1 1ORY149S A 2 CAB 0,00 72,00 0 72,00
2 1ORY149S B 2 CAB 0,00 72,00 0 86,40

3 1ORY85P A 1 CPT 942,90 106,00 0 1048,90
4 2LIS24X A 2 CAB 939,00 144,00 0 1083,00
5 3LIS25X B 2 CAB 0,00 72,00 0 86,40
6 2LHR63P B 1 CPT 777,00 212,00 0 1186,80
7 2LHR63P B 1 OPT 0,00 148,00 0 177,60
8 1LHR31P A 2 CCB 687,65 72,00 0 759,65
9 2LHR19P B 2 CCB 0,00 144,00 0 172,80
10 1ZRH12X C 1 CPT 0,00 212,00 0 296,80
11 1LIS16S A 2 CAB 0,00 72,00 0 72,00
12 1LIS16S C 2 CAB 0,00 72,00 0 100,80
13 1LIS158T B 2 CAB 0,00 72,00 0 86,40
14 3LIS174S A 2 CAB 1051,60 216,00 0 1267,60
15 4LIS50A A 1 OPT 246,40 296,00 0 542,40

Totals 4644,55 1982,00 0 7039,55
Table 8. Data collected (partial) after using method 1 (human user)

In the second method we have used our approach as indicated in Section 4 but with =0 in
Equation 1 (Total Operational Cost), i.e., although we calculate the Quality Operational Cost as
indicated in Equation 6 we did not considered this value in resolution as well as in the
decision process. The data collected is presented in Table 9.

Duty ID Base
ID
Crew
Grp
Rank Hour Pay Perdiem
Pay
Quality
Op. Cost
Direct Op.
Cost
1 1ORY149S A 2 CAB 0,00 72,00 0 72,00
2 1ORY149S B 2 CAB 0,00 72,00 501,31 86,40
3 1ORY85P B 1 CPT 0,00 106,00 0 127,20
4 2LIS24X C 2 CAB 563,40 62,00 1561,76 875,56
5 3LIS25X B 2 CAB 0,00 72,00 1877,73 86,40
6 2LHR63P C 1 CPT 0,00 212,00 658 296,80
7 2LHR63P A 1 OPT 0,00 144,00 687,62 144,00
8 1LHR31P B 2 CCB 229,17 72,00 0 361,40
9 2LHR19P B 2 CCB 0,00 144,00 788,78 172,80
10 1ZRH12X C 1 CPT 0,00 212,00 0 296,80
11 1LIS16S A 2 CAB 0,00 72,00 961,95 72,00
12 1LIS16S C 2 CAB 0,00 72,00 301,48 100,80
13 1LIS158T B 2 CAB 0,00 72,00 0 86,40
14 3LIS174S C 2 CAB 411,00 93,00 0 705,60
15 4LIS50A B 1 OPT 0,00 296,00 449,84 355,20

Totals 1203,57 1773,00 7788,47 3839,36
Table 9. Data collected (partial) after using method 2 (No Quality Costs)

In the third method we have used our approach as indicated in Section 4 but with =1 in
Equation 1, i.e., considering the Quality Operational Cost in the resolution as well as in the
decision process. The Quality Operational Cost was calculated using two passenger profiles
(business and economy classes) and with =0,1. Equation 9 and Equation 10 are the
formulas used to calculate the delay cost of each passenger in business and economy profile,
respectively. For more information about how we reached these equations, please read
(Castro & Oliveira, 2009).

C
business
= 0.16 * x
2
+ 1.38 * x x = minutes of flight delay, x >= 0 (9)

C
economy
= 1.20 * x x = minutes of flight delay, x >= 0 (10)


After setting-up the scenario we found the solutions for each crew event using three
methods.

Duty
DateTime
DutyID Flt
Dly
C
Pax
Y
Pax
End
DateTime
Ready
DateTime
Cred
Min
Crew
Grp
Rnk Crw
Nr
Crew
Name
1 05-06 07:25 1ORY149S 0 7 123 05-06 13:35 06-06 01:35 370 2 CAB 80 John A
2 05-06 07:25 1ORY149S 10 11 114 05-06 13:35 06-06 01:35 370 2 CAB 45 Mary A
3 05-06 07:25 1ORY85P 0 10 112 05-06 13:35 06-06 01:35 370 1 CPT 35 Anthony
4 15-06 04:10 2LIS24X 30 0 90 16-06 16:15 17-06 04:15 1757 2 CAB 99 Paul M
5 15-06 04:10 3LIS25X 25 3 77 15-06 09:20 15-06 21:20 632 2 CAB 56 John B
6 15-06 12:50 2LHR63P 5 25 85 16-06 20:45 17-06 08:45 1549 1 CPT 57 Paul S
7 15-06 12:50 2LHR63P 0 20 95 16-06 20:45 17-06 08:45 1549 1 OPT 53 Mary S
8 15-06 14:15 1LHR31P 0 23 52 15-06 20:55 16-06 08:55 843 2 CCB 23 Sophie
9 15-06 15:25 2LHR19P 10 27 105 16-06 20:45 17-06 08:45 1341 2 CCB 34 Angel
10 15-06 15:25 1ZRH12X 0 5 115 17-06 09:30 17-06 21:30 1318 1 CPT 32 Peter B
11 25-06 05:20 1LIS16S 20 3 97 25-06 15:05 26-06 03:05 585 2 CAB 20 Paul G
12 25-06 05:20 1LIS16S 5 2 108 25-06 15:05 26-06 03:05 585 2 CAB 10 Alice
13 25-06 05:20 1LIS158T 0 4 92 25-06 15:05 26-06 03:05 585 2 CAB 15 Daniel
14 25-06 06:15 3LIS174S 0 1 129 27-06 16:15 28-06 04:15 1258 2 CAB 71 George
15 25-06 14:20 4LIS50A 0 2 83 28-06 19:40 29-06 07:40 219 1 OPT 65 Allan
Table 7. Events data used for testing

In the first method we used one of the best users from the AOCC, with current tools
available, to find the solutions. The user uses software that shows the roster of each
crewmember in a Gantt chart for a specific period. The user can scroll down the information,
filter according to the crew rank and base, and sort the information by name, month duty,
etc. Each user has a specific way of trying to find the solutions. However, we have observed
that, in general, they follow these steps:
1. Open the roster for a one month period, starting two days before the current day.
For example, lets suppose that the current day is 7th of June of 2006, they open the
roster from the 5th of June until the 4th of July.
2. Filter the roster by crew rank and base, where the base is equal to the base where
the crew event happened and crew rank is equal to the crewmember rank that did
not report for duty.
3. Order the information by month duty, in an ascendant order and by seniority in a
descendent order.
4. Visually, they scroll down the information until they found a crewmember with an
open space for the period of time that corresponds to the duty to be assigned. This
period of time takes into consideration the start and end time of the duty and also
the time required for resting (ready date time).
5. If they do not found a crewmember in the base specified, they try to find it in
another base, filtering the information accordingly.
6. They assign the duty to the crewmember with less credit hours.
The data collected using this method is presented in Table 8. We point out that the data in
columns marked with an asterisk where calculated manually, according to the equations
presented in chapter 4.3. The reason for this is that the information system that is available
for the users does not include information related with any kind of costs.

Duty ID Base
ID
Crew
Grp
Rank Hour Pay (*) Perdiem
Pay (*)
Quality
Op. Cost
Op. Cost
(*)
1 1ORY149S A 2 CAB 0,00 72,00 0 72,00
2 1ORY149S B 2 CAB 0,00 72,00 0 86,40

3 1ORY85P A 1 CPT 942,90 106,00 0 1048,90
4 2LIS24X A 2 CAB 939,00 144,00 0 1083,00
5 3LIS25X B 2 CAB 0,00 72,00 0 86,40
6 2LHR63P B 1 CPT 777,00 212,00 0 1186,80
7 2LHR63P B 1 OPT 0,00 148,00 0 177,60
8 1LHR31P A 2 CCB 687,65 72,00 0 759,65
9 2LHR19P B 2 CCB 0,00 144,00 0 172,80
10 1ZRH12X C 1 CPT 0,00 212,00 0 296,80
11 1LIS16S A 2 CAB 0,00 72,00 0 72,00
12 1LIS16S C 2 CAB 0,00 72,00 0 100,80
13 1LIS158T B 2 CAB 0,00 72,00 0 86,40
14 3LIS174S A 2 CAB 1051,60 216,00 0 1267,60
15 4LIS50A A 1 OPT 246,40 296,00 0 542,40

Totals 4644,55 1982,00 0 7039,55
Table 8. Data collected (partial) after using method 1 (human user)

In the second method we have used our approach as indicated in Section 4 but with =0 in
Equation 1 (Total Operational Cost), i.e., although we calculate the Quality Operational Cost as
indicated in Equation 6 we did not considered this value in resolution as well as in the
decision process. The data collected is presented in Table 9.

Duty ID Base
ID
Crew
Grp
Pay
Quality
Op. Cost
Direct Op.
Cost
1 1ORY149S A 2 CAB 0,00 72,00 0 72,00
2 1ORY149S B 2 CAB 0,00 72,00 501,31 86,40
3 1ORY85P B 1 CPT 0,00 106,00 0 127,20
4 2LIS24X C 2 CAB 563,40 62,00 1561,76 875,56
5 3LIS25X B 2 CAB 0,00 72,00 1877,73 86,40
6 2LHR63P C 1 CPT 0,00 212,00 658 296,80
7 2LHR63P A 1 OPT 0,00 144,00 687,62 144,00
8 1LHR31P B 2 CCB 229,17 72,00 0 361,40
9 2LHR19P B 2 CCB 0,00 144,00 788,78 172,80
10 1ZRH12X C 1 CPT 0,00 212,00 0 296,80
11 1LIS16S A 2 CAB 0,00 72,00 961,95 72,00
12 1LIS16S C 2 CAB 0,00 72,00 301,48 100,80
13 1LIS158T B 2 CAB 0,00 72,00 0 86,40
14 3LIS174S C 2 CAB 411,00 93,00 0 705,60
15 4LIS50A B 1 OPT 0,00 296,00 449,84 355,20

Totals 1203,57 1773,00 7788,47 3839,36
Table 9. Data collected (partial) after using method 2 (No Quality Costs)

In the third method we have used our approach as indicated in Section 4 but with =1 in
Equation 1, i.e., considering the Quality Operational Cost in the resolution as well as in the
decision process. The Quality Operational Cost was calculated using two passenger profiles
(business and economy classes) and with =0,1. Equation 9 and Equation 10 are the
formulas used to calculate the delay cost of each passenger in business and economy profile,
respectively. For more information about how we reached these equations, please read
(Castro & Oliveira, 2009).

C
business
= 0.16 * x
2
+ 1.38 * x x = minutes of flight delay, x >= 0 (9)

C
economy
= 1.20 * x x = minutes of flight delay, x >= 0 (10)


The data collected is presented in Table 10.

Duty ID Base
ID
Crew
Grp
Pay
Quality
Op. Cost
Direct Op.
Cost
1 1ORY149S A 2 CAB 0,00 72,00 0 72,00
2 1ORY149S B 2 CAB 0,00 72,00 501,31 86,40
3 1ORY85P B 1 CPT 0,00 106,00 0 127,20
4 2LIS24X C 2 CAB 503,50 144,00 1060,92 906,50
5 3LIS25X C 2 CAB 0,00 72,00 1420,78 100,80
6 2LHR63P B 1 CPT 102,90 212,00 272,10 377,88
7 2LHR63P B 1 OPT 37,22 144,00 0 217,46
8 1LHR31P B 2 CCB 229,17 72,00 0 361,40
9 2LHR19P B 2 CCB 0,00 144,00 788,78 172,80
10 1ZRH12X C 1 CPT 0,00 212,00 0 296,80
11 1LIS16S A 2 CAB 0,00 80,00 593,30 80,00
12 1LIS16S C 2 CAB 0,00 80,00 144,34 112,00
13 1LIS158T B 2 CAB 0,00 72,00 0 86,40
14 3LIS174S C 2 CAB 411,00 93,00 0 705,60
15 4LIS50A A 1 OPT 138,83 288,00 0 426,83

Totals 1422,62 1863,00 4781,53 4130,07
Table 10. Data collected (partial) after using method 3 (Quality Costs)

6. Results and Discussion
Table 11 shows a comparison of the results obtained through the above methods. We point
out that in method 1 (human) we did not calculate the quality costs, and in method 2 (agent-
no-quality) we did calculate the quality costs but they were not used to find the best solution,
although we use that value for comparison purposes. From the results obtained we can see
that on average, method 3 (agent-quality) produced solutions that decreased flight delays in
aprox. 36%. Agent-quality is, on average 3% slower than agent-no-quality in finding a solution
and produces solutions that represent a decrease of 23% on the total operational costs, when
compared with agent-no-quality.
From the results (Table 11) we can see that our approach obtains valid solutions faster and
with fewer direct operational costs when compared with the current method used in a real
airline company (human). Agent-no-quality represents a decrease of aprox. 45,5% and agent-
quality a decrease of aprox. 41%. Agent-quality has a higher direct operational cost than agent-
no-quality because it uses the quality operational cost in the decision process. If we read this
number without any other consideration, we have to say that the goal of having less direct
operational costs was not achieved. An 8% increased on direct operational costs can
represent a lot of money. However, we should read this number together with the flight
delay figure. As we can see, although agent-quality has increased the direct operational costs
(when compared with agent-no-quality) in 8% it was able to choose solutions that decrease, in
average, 36% of the flight delays. This means that, when there are multiple solutions to the
same problem, agent-quality is able to choose the one with less operational cost, less quality
costs (hence, better passenger satisfaction) and, because of the relation between quality costs
and flight delays, the solution that produces shorter flight delays.

Human (M1) Agent-no-Quality (M2) Agent-Quality (M3)
Total % Total % Total %
Base of the solution:
- From the crew event base (A) 7 47% 3 20% 3 20%
- From base B 6 40% 7 47% 7 47%

- From base C 2 13% 5 33% 5 33%

Time to Find Solution (avr sec) 101 100,00% 25 24,75% 26 25,74%

Flight Delays (avr min): 11 100,00% 7 63,64%
- Base A (avr) 14 40% 7 30%
- Base B (avr) 9 26% 4 17%
- Base C (avr) 12 34% 12 52%

Total Direct Operational Costs: 7039,60 100,00% 3839,36 54,54% 4130,07 58,67%
Total by Base:
- Base A 4845,55 92,42% 288,00 11,23% 578,83 14,02%
- Base B 1796,40 34,26% 1275,80 49,77% 1429,54 34,61%
- Base C 397,60 7,58% 2275,56 88,77% 2121,70 51,37%

Total Quality Operational Cost: 7788,47 100% 4781,53 61,39%
Total by Base:
- Base A 1649,57 21,18% 593,30 12,41%
- Base B 3617,66 46,45% 1562,19 32,67%
- Base C 2521,24 32,37% 2626,04 54,92%

Total Operational Costs: 11628,01 165% 8911,60 126,6%
Total by Base:
- Base A 1937,57 16,66% 1172,13 13,15%
- Base B 4088,42 35,16% 2991,73 33,57%
- Base C 4796,80 41,25% 4747,74 53,28%
Table 11. Summary of the results obtained by each method

From this conclusion, one can argue that if we just include the direct operational costs and
the expected flight delay, minimizing both values, the same results could be achieved
having all passengers happy. In general, this assumption might be true. However, when we
have to choose between two solutions with the same direct operational cost and delay time,
which one should we choose? In our opinion, the answer depends on the profile of the
passengers of each flight and on the importance they give to the delays (quality operational
cost), and not only in minimizing the flight delays and direct operational cost. Agent-quality
takes into consideration this important information when making decisions. This is the
reason why we think that one of the main contributions of our work is the generic approach
to quantify the passenger satisfaction regarding delaying a flight, from the passenger point
of view. It is fair to say that we cannot conclude that our MAS will always have this
behaviour. For that we need to evaluate a higher number of scenarios, at different times of
the year (we might have seasonal behaviours) and, then, find an average value.
Additionally, we found that the cooperation between different operational bases has
increased with our approach, because we evaluate all the solutions found (including the
ones from different operational bases where the event happened) and we select the one with
less cost. In human, they choose the first one they find with less credit hours, usually from
the same base where the event was triggered. This cooperation is also possible to be inferred
from the costs by base. In Table 11 is possible to see that the direct operational costs of base
C using human represents only 7,58% of the costs of all bases, whilst in agent-no-quality and
agent-quality it represents 88,77% and 51,73%, respectively. The same is possible to be
inferred from the other bases (although with different figures). This means that our MAS
uses more resources from other bases than the base where the problem happened (base A).


The data collected is presented in Table 10.

Duty ID Base
ID
Crew
Grp
Pay
Quality
Op. Cost
Direct Op.
Cost
1 1ORY149S A 2 CAB 0,00 72,00 0 72,00
2 1ORY149S B 2 CAB 0,00 72,00 501,31 86,40
3 1ORY85P B 1 CPT 0,00 106,00 0 127,20
4 2LIS24X C 2 CAB 503,50 144,00 1060,92 906,50
5 3LIS25X C 2 CAB 0,00 72,00 1420,78 100,80
6 2LHR63P B 1 CPT 102,90 212,00 272,10 377,88
7 2LHR63P B 1 OPT 37,22 144,00 0 217,46
8 1LHR31P B 2 CCB 229,17 72,00 0 361,40
9 2LHR19P B 2 CCB 0,00 144,00 788,78 172,80
10 1ZRH12X C 1 CPT 0,00 212,00 0 296,80
11 1LIS16S A 2 CAB 0,00 80,00 593,30 80,00
12 1LIS16S C 2 CAB 0,00 80,00 144,34 112,00
13 1LIS158T B 2 CAB 0,00 72,00 0 86,40
14 3LIS174S C 2 CAB 411,00 93,00 0 705,60
15 4LIS50A A 1 OPT 138,83 288,00 0 426,83

Totals 1422,62 1863,00 4781,53 4130,07
Table 10. Data collected (partial) after using method 3 (Quality Costs)

6. Results and Discussion
Table 11 shows a comparison of the results obtained through the above methods. We point
out that in method 1 (human) we did not calculate the quality costs, and in method 2 (agent-
no-quality) we did calculate the quality costs but they were not used to find the best solution,
although we use that value for comparison purposes. From the results obtained we can see
that on average, method 3 (agent-quality) produced solutions that decreased flight delays in
aprox. 36%. Agent-quality is, on average 3% slower than agent-no-quality in finding a solution
and produces solutions that represent a decrease of 23% on the total operational costs, when
compared with agent-no-quality.
From the results (Table 11) we can see that our approach obtains valid solutions faster and
with fewer direct operational costs when compared with the current method used in a real
airline company (human). Agent-no-quality represents a decrease of aprox. 45,5% and agent-
quality a decrease of aprox. 41%. Agent-quality has a higher direct operational cost than agent-
no-quality because it uses the quality operational cost in the decision process. If we read this
number without any other consideration, we have to say that the goal of having less direct
operational costs was not achieved. An 8% increased on direct operational costs can
represent a lot of money. However, we should read this number together with the flight
delay figure. As we can see, although agent-quality has increased the direct operational costs
(when compared with agent-no-quality) in 8% it was able to choose solutions that decrease, in
average, 36% of the flight delays. This means that, when there are multiple solutions to the
same problem, agent-quality is able to choose the one with less operational cost, less quality
costs (hence, better passenger satisfaction) and, because of the relation between quality costs
and flight delays, the solution that produces shorter flight delays.

Human (M1) Agent-no-Quality (M2) Agent-Quality (M3)
Total % Total % Total %
Base of the solution:
- From the crew event base (A) 7 47% 3 20% 3 20%
- From base B 6 40% 7 47% 7 47%

- From base C 2 13% 5 33% 5 33%

Time to Find Solution (avr sec) 101 100,00% 25 24,75% 26 25,74%

Flight Delays (avr min): 11 100,00% 7 63,64%
- Base A (avr) 14 40% 7 30%
- Base B (avr) 9 26% 4 17%
- Base C (avr) 12 34% 12 52%

Total Direct Operational Costs: 7039,60 100,00% 3839,36 54,54% 4130,07 58,67%
Total by Base:
- Base A 4845,55 92,42% 288,00 11,23% 578,83 14,02%
- Base B 1796,40 34,26% 1275,80 49,77% 1429,54 34,61%
- Base C 397,60 7,58% 2275,56 88,77% 2121,70 51,37%

Total Quality Operational Cost: 7788,47 100% 4781,53 61,39%
Total by Base:
- Base A 1649,57 21,18% 593,30 12,41%
- Base B 3617,66 46,45% 1562,19 32,67%
- Base C 2521,24 32,37% 2626,04 54,92%

Total Operational Costs: 11628,01 165% 8911,60 126,6%
Total by Base:
- Base A 1937,57 16,66% 1172,13 13,15%
- Base B 4088,42 35,16% 2991,73 33,57%
- Base C 4796,80 41,25% 4747,74 53,28%
Table 11. Summary of the results obtained by each method

From this conclusion, one can argue that if we just include the direct operational costs and
the expected flight delay, minimizing both values, the same results could be achieved
having all passengers happy. In general, this assumption might be true. However, when we
have to choose between two solutions with the same direct operational cost and delay time,
which one should we choose? In our opinion, the answer depends on the profile of the
passengers of each flight and on the importance they give to the delays (quality operational
cost), and not only in minimizing the flight delays and direct operational cost. Agent-quality
takes into consideration this important information when making decisions. This is the
reason why we think that one of the main contributions of our work is the generic approach
to quantify the passenger satisfaction regarding delaying a flight, from the passenger point
of view. It is fair to say that we cannot conclude that our MAS will always have this
behaviour. For that we need to evaluate a higher number of scenarios, at different times of
the year (we might have seasonal behaviours) and, then, find an average value.
Additionally, we found that the cooperation between different operational bases has
increased with our approach, because we evaluate all the solutions found (including the
ones from different operational bases where the event happened) and we select the one with
less cost. In human, they choose the first one they find with less credit hours, usually from
the same base where the event was triggered. This cooperation is also possible to be inferred
from the costs by base. In Table 11 is possible to see that the direct operational costs of base
C using human represents only 7,58% of the costs of all bases, whilst in agent-no-quality and
agent-quality it represents 88,77% and 51,73%, respectively. The same is possible to be
inferred from the other bases (although with different figures). This means that our MAS
uses more resources from other bases than the base where the problem happened (base A).


7. Conclusion and Future Works
In this chapter we have introduced the Airline Operations Control Problem as well as the
Airline Operations Control Centre (AOCC), including typical organizations and problems,
the current disruption management (DM) process and a description of the main costs
involved. We described our agent-based approach to this problem, including the reasons
that make us adopt an agent and multi-agent system (MAS) paradigm; the MAS architecture
with agents, roles and protocols as well as some agent characteristics like autonomy and
social-awareness; the decision mechanisms, including the costs criteria and negotiation
protocols used and examples of the problem solving algorithms. Using data from a real
airline company, we tested our approach and discussed the results obtained by three
different methods. We have shown that our approach is able to select solutions that
contribute to a better passenger satisfaction and that produce shorter flight delays when
compared with methods that only minimize direct operational costs.
We are working on several improvements. Some of them are already implemented.
However, we did not perform, yet, enough tests to have meaningful results. These are our
goals:
- Improve autonomy and learning characteristics of the Monitor agent, so that he is
able to consider new events (or change existing ones) according to the experience
he gets from monitoring the operation, without relying exclusively on the
definition of events created by the human operator.
- Working on a protocol at the Manager Agent team level that allows a better
coordination and improves the distributed problem solving characteristics of our
approach. For example, including in each team, knowledge provided by other
teams to improve the objective function of each specialist agent, with parameters of
the other dimensions (aircraft, crew and passenger).
- Solving problems learning by example, applying Case-Based Reasoning (CBR).
- Increase robustness of future schedules by applying the knowledge gathered from
learning by example.
- Study the behaviour and compare the results, of several problem solving
algorithms, including the ones that implement heuristics to specific problems. The
idea is to classify the algorithms according to their success rate in solving specific
types of problems in this domain.

8. Acknowledgements
The first author is supported by FCT (Fundao para a Cincia e Tecnologia) under research
grant SFRH/BD/44109/2008. The authors are grateful to TAP Portugal for allowing the use
of real data from the airline company.

9. References
Abdelgahny, A., Ekollu, G., Narisimhan, R. & Abdelgahny, K. 2004. A Proactive Crew
Recovery Decision Support Tool for Commercial Airlines during Irregular
Operations, Annals of Operations Research, Vol.127, pp. 309-331.

Barnhart, C., Belobaba, P. & Odoni, A. (2003). Applications of Operations Research in the
Air Transport Industry, Transportation Science, Vol. 37, pp. 368-391.
Bellifemine, F., Caire, G., Trucco, T. & Rimassa, G. (2004). JADE Programmers Guide, JADE
3.3, TILab S.p.A.
Bratu, S. & Barnhart, C. (2006). Flight Operations Recovery: New Approaches Considering
Passenger Recovery, Journal of Scheduling, Vol. 9, No. 3, pp.279-298.
Castro, A.J.M. (2007). Designing a Multi-Agent System for Monitoring and Operations
Recovery for an Airline Operations Control Centre, MsC Thesis, pp.1-133, Porto,
Portugal: University of Porto, Faculty of Engineering.
Castro, A.J.M. & Oliveira, E. (2007). Using Specialized Agents in a Distributed MAS to Solve
Airline Operations Problems: a Case Study, Proceedings of IAT 2007 (Intelligent Agent
Technology Conference), pp. 473-476, Silicon Valley, California, USA 2-5 November
2007, IEEE Computer Society, ISBN: 0-7695-3027-3.
Castro, A. (2008). Centros de Controlo Operacional: Organizao e Ferramentas, Monograph
for Post-graduation in Air Transport Operations, ISEC - Instituto Superior de Educao
e Cincias, Outubro 2008 (In Portuguese).
Castro, A. & Oliveira, E. (2008). The rationale behind the development of an airline
operations control centre using Gaia-based methodology, International Journal of
Agent-Oriented Software Engineering, Vol. 2, No. 3, pp. 350377.
Castro, A.J.M & Oliveira, E. (2009). Using Quality Costs in a Multi-Agent System for Airline
Operations Control, 11th International Conference on Enterprise Information Systems,
Artificial Intelligence and Decision Support Systems, pp. 19-24, Milan, Italy, May 6-
10, 2009. ISBN: 978-989-8111-85-2
Clausen, J., Larsen, A. & Larsen, J. (2005). Disruption Management in the Airline Industry
Concepts, Models and Methods, Technical Report, 2005-01, Informatics and
Mathematical Modeling, Technical University of Denmark, DTU.
Cowling, P.I., Ouelhadj, D., & Petrovic, S. (2003). A Multi-Agent Architecture for Dynamic
Scheduling of Steel Hot Rolling, Journal of Intelligent Manufacturing, Special Issue on
Agent-Based Manufacturing Process Planning and Scheduling, Vol. 15, No. 5, pp. 457-
470.
Elamy, A. (2005). Perspectives in Agents-Based Technology, AgentLink News 18, August
2005, ISSN: 1465-3842.
Fipa. (2002). FIPA Contract Net Interaction Protocol Specification, http://www.fipa.org/specs/
fipa00029/SC00029H.html, Accessed in 2009/06/21.
Grosche, T. (2009). Computational Intelligence in Integrated Airline scheduling, Springer-
Verlag Berlin Heidelberg, 978-3-540-89886-3, Germany.
Jonker, G., Meyer, J.-J., & Dignum, F. (2005). Towards a Market Mechanism for Airport
Traffic Control, 12th Portuguese Conference on Artificial Intelligence (EPIA 2005),
Covilha, Portugal.
Kohl, N. & Karisch, S. (2004). Airline crew rostering. problem types, modeling, and
optimization, Annals of Operations Research, Vol. 127, pp. 223-257.
Kohl, N., Larsen, A., Larsen, J., Ross, A. & Tiourline, S. (2004). Airline Disruption
Management Perspectives, Experiences and Outlook, Technical Report, CRTR-0407,
Carmen Research.
Lettovsky, L. (1997). Airline Operations Recovery: An Optimization Approach, Ph.D.
dissertation, Georgia Institute of Technology, Atlanta, USA.

7. Conclusion and Future Works
In this chapter we have introduced the Airline Operations Control Problem as well as the
Airline Operations Control Centre (AOCC), including typical organizations and problems,
the current disruption management (DM) process and a description of the main costs
involved. We described our agent-based approach to this problem, including the reasons
that make us adopt an agent and multi-agent system (MAS) paradigm; the MAS architecture
with agents, roles and protocols as well as some agent characteristics like autonomy and
social-awareness; the decision mechanisms, including the costs criteria and negotiation
protocols used and examples of the problem solving algorithms. Using data from a real
airline company, we tested our approach and discussed the results obtained by three
different methods. We have shown that our approach is able to select solutions that
contribute to a better passenger satisfaction and that produce shorter flight delays when
compared with methods that only minimize direct operational costs.
We are working on several improvements. Some of them are already implemented.
However, we did not perform, yet, enough tests to have meaningful results. These are our
goals:
- Improve autonomy and learning characteristics of the Monitor agent, so that he is
able to consider new events (or change existing ones) according to the experience
he gets from monitoring the operation, without relying exclusively on the
definition of events created by the human operator.
- Working on a protocol at the Manager Agent team level that allows a better
coordination and improves the distributed problem solving characteristics of our
approach. For example, including in each team, knowledge provided by other
teams to improve the objective function of each specialist agent, with parameters of
the other dimensions (aircraft, crew and passenger).
- Solving problems learning by example, applying Case-Based Reasoning (CBR).
- Increase robustness of future schedules by applying the knowledge gathered from
learning by example.
- Study the behaviour and compare the results, of several problem solving
algorithms, including the ones that implement heuristics to specific problems. The
idea is to classify the algorithms according to their success rate in solving specific
types of problems in this domain.

8. Acknowledgements
The first author is supported by FCT (Fundao para a Cincia e Tecnologia) under research
grant SFRH/BD/44109/2008. The authors are grateful to TAP Portugal for allowing the use
of real data from the airline company.

9. References
Abdelgahny, A., Ekollu, G., Narisimhan, R. & Abdelgahny, K. 2004. A Proactive Crew
Recovery Decision Support Tool for Commercial Airlines during Irregular
Operations, Annals of Operations Research, Vol.127, pp. 309-331.

Barnhart, C., Belobaba, P. & Odoni, A. (2003). Applications of Operations Research in the
Air Transport Industry, Transportation Science, Vol. 37, pp. 368-391.
Bellifemine, F., Caire, G., Trucco, T. & Rimassa, G. (2004). JADE Programmers Guide, JADE
3.3, TILab S.p.A.
Bratu, S. & Barnhart, C. (2006). Flight Operations Recovery: New Approaches Considering
Passenger Recovery, Journal of Scheduling, Vol. 9, No. 3, pp.279-298.
Castro, A.J.M. (2007). Designing a Multi-Agent System for Monitoring and Operations
Recovery for an Airline Operations Control Centre, MsC Thesis, pp.1-133, Porto,
Portugal: University of Porto, Faculty of Engineering.
Castro, A.J.M. & Oliveira, E. (2007). Using Specialized Agents in a Distributed MAS to Solve
Airline Operations Problems: a Case Study, Proceedings of IAT 2007 (Intelligent Agent
Technology Conference), pp. 473-476, Silicon Valley, California, USA 2-5 November
2007, IEEE Computer Society, ISBN: 0-7695-3027-3.
Castro, A. (2008). Centros de Controlo Operacional: Organizao e Ferramentas, Monograph
for Post-graduation in Air Transport Operations, ISEC - Instituto Superior de Educao
e Cincias, Outubro 2008 (In Portuguese).
Castro, A. & Oliveira, E. (2008). The rationale behind the development of an airline
operations control centre using Gaia-based methodology, International Journal of
Agent-Oriented Software Engineering, Vol. 2, No. 3, pp. 350377.
Castro, A.J.M & Oliveira, E. (2009). Using Quality Costs in a Multi-Agent System for Airline
Operations Control, 11th International Conference on Enterprise Information Systems,
Artificial Intelligence and Decision Support Systems, pp. 19-24, Milan, Italy, May 6-
10, 2009. ISBN: 978-989-8111-85-2
Clausen, J., Larsen, A. & Larsen, J. (2005). Disruption Management in the Airline Industry
Concepts, Models and Methods, Technical Report, 2005-01, Informatics and
Mathematical Modeling, Technical University of Denmark, DTU.
Cowling, P.I., Ouelhadj, D., & Petrovic, S. (2003). A Multi-Agent Architecture for Dynamic
Scheduling of Steel Hot Rolling, Journal of Intelligent Manufacturing, Special Issue on
Agent-Based Manufacturing Process Planning and Scheduling, Vol. 15, No. 5, pp. 457-
470.
Elamy, A. (2005). Perspectives in Agents-Based Technology, AgentLink News 18, August
2005, ISSN: 1465-3842.
Fipa. (2002). FIPA Contract Net Interaction Protocol Specification, http://www.fipa.org/specs/
fipa00029/SC00029H.html, Accessed in 2009/06/21.
Grosche, T. (2009). Computational Intelligence in Integrated Airline scheduling, Springer-
Verlag Berlin Heidelberg, 978-3-540-89886-3, Germany.
Jonker, G., Meyer, J.-J., & Dignum, F. (2005). Towards a Market Mechanism for Airport
Traffic Control, 12th Portuguese Conference on Artificial Intelligence (EPIA 2005),
Covilha, Portugal.
Kohl, N. & Karisch, S. (2004). Airline crew rostering. problem types, modeling, and
optimization, Annals of Operations Research, Vol. 127, pp. 223-257.
Kohl, N., Larsen, A., Larsen, J., Ross, A. & Tiourline, S. (2004). Airline Disruption
Management Perspectives, Experiences and Outlook, Technical Report, CRTR-0407,
Carmen Research.
Lettovsky, L. (1997). Airline Operations Recovery: An Optimization Approach, Ph.D.
dissertation, Georgia Institute of Technology, Atlanta, USA.

Liu, T., Jeng, C. & Chang, Y. (2008). Disruption Management of an Inequality-Based Multi-
Fleet Airline Schedule by a Multi-Objective Genetic Algorithm, Transportation
Planning and Technology, Vol. 31, No. 6, pp. 613-639.
Mota, A. (2007). Multi-Agent System for an Airline Operations Control Centre, MsC Thesis,
Porto, Portugal: University of Porto, Faculty of Engineering.
Ouelhadj, D. (2003). A Multi-Agent System for the Integrated Dynamic Scheduling of Steel
Production, Ph.D. dissertation, The University of Nottingham, School of Computer
Science and Information Technology, England, August 2003.
Rosenberger, J., Johnson, E. & Nemhauser, G. (2001). Rerouting aircraft for airline recovery,
Technical Report, TLI-LEC 01-04, Georgia Institute of Technology.
Rosenberger, J., Schaefer, A., Goldsmans, D., Johnson, E., Kleywegt, A. & Nemhauser, G.
(2002). A Stochastic Model of Airline Operations, Transportation Science, Vol. 36, No.
4, pp. 357-377.
Smith, R. (1980). The Contract Net Protocol: High-Level Communication and Control in a
Distributed Problem Solver, IEEE Transactions on Computers, Vol. C, No. 29.
Stone, P. & Veloso, M. (2000). Multi-Agent Systems: a Survey from a Machine Learning
Perspective, Autonomous Robotics, Vol. 8. No. 3.
Tumer, K., & Agogino, A. (2007). Distributed Agent-Based Air Traffic Flow Management,
6th International Joint Conference on Autonomous Agents and Multiagent Systems
(AAMAS 2007), Honolulu, Hawaii.
Wolfe, S. R., Jarvis, P. A., Enomoto, F. Y., & Sierhuis, M. (2007). Comparing Route Selection
Strategies in Collaborative Traffic Flow Management., Proceedings of IAT 2007
(Intelligent Agent Technology Conference), Silicon Valley, California, USA 2-5
November 2007, IEEE Computer Society, ISBN: 0-7695-3027-3.
Wooldridge, M., (2009). When is an Agent-Based Solution Appropriate? An Introduction to
Multiagent Systems, West Sussex, England: John Wiley & Sons, Ltd., 2nd Edition (15
May 2009), pp. 183, ISBN: 978-0470519462.
Yang, M. (2007). Using Advanced Tabu Search Techniques to Solve Airline Disruption
Management Problems, Ph.D. dissertation, The University of Texas at Austin,
December 2007, USA.

António J.M. Castro and Eugénio Oliveira: LIACC-NIAD&R, FEUP, DEI, University of Porto Portugal

Uploaded by

Copyright:

Available Formats

António J.M. Castro and Eugénio Oliveira: LIACC-NIAD&R, FEUP, DEI, University of Porto Portugal

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

António J.M. Castro and Eugénio Oliveira: LIACC-NIAD&R, FEUP, DEI, University of Porto Portugal

Uploaded by

Copyright:

Available Formats

Disruption Management in Airline Operations Control An Intelligent Agent-Based Approach 107

Disruption Management in Airline Operations Control An Intelligent

You might also like