Modelling Passengers Flow at Airport Terminals
Individual Agent Decision Model for Stochastic Passenger Behaviour
Wenbo Ma, Clinton Fookes, Tristan Kleinschmidt and Prasad Yarlagadda
Queensland University of Technology, George St, Brisbane, Australia
Keywords: Agent-based Models, Bayesian Networks.
Abstract: Airport system is complex. Passenger dynamics within it appear to be complicate as well. Passenger
behaviours outside standard processes are regarded more significant in terms of public hazard and service
rate issues. In this paper, we devised an individual agent decision model to simulate stochastic passenger
behaviour in airport departure terminal. Bayesian networks are implemented into the decision making model
to infer the probabilities that passengers choose to use any in-airport facilities. We aim to understand
dynamics of the discretionary activities of passengers.
1 INTRODUCTION
Airport terminal is a particular built environment
where there are large numbers of passengers travel
through daily. It not only handles standard processes
for departure and arrival but also provide in-airport
discretionary services for passengers to use (Ma,
Kleinschmidt et al. 2011). Airports have to satisfy a
myriad of different tasks. Continual legal changes,
security constraints, safety in public facilities and
technological innovations always have a significant
effect on handling passengers. Current models which
mostly use aggregated approaches are hard to adapt
to continuous changes (Gatersleben and Van der
Weij 1999; Takakuwa and Oyama 2003; Andreatta,
Brunetta et al. 2007). They seem only focus on
standard processes and ignore discretionary
components, i.e. duty-free shops, in-airport
restaurants and telephones. Moreover, airports have
been under growing pressure to be financially more
self-sufficient since the early 1990s and bound to be
less reliant on government support (IATA 1997).
Airports rely increasingly on concession services to
bring in more revenues (Fu and Zhang 2010).
Concession services refer to the non-aircraft-related
operations in terminals and on airport land,
including activities such as running or leasing out
shopping concessions of various kinds, car parking
and rental, banking and catering, and so on. ATRS
(2006) reports that most of the world major airports
acquire anywhere between 45 and 80 percent of their
total revenues from non-aviation sectors, a major
part of which is revenue from retail and parking.
Since these operations depend greatly on passenger
throughput of an airport, there is a complementarily
between the demand for aviation services and the
demand for concession services. In the passenger
perspective, the escalating expectations of
passengers make airport system become
complicating. Passengers nowadays are accustomed
to sophisticated, fast-changing technology
environments at home and at work. They have
grown to expect painless self-service and instant,
unfettered access to resources and information. Like
customers in other industries, passengers expect
better, cheaper, and faster services from airlines and
airports. They want real-time information about
flight delays and demand streamlined processes for
check-in, transit, and boarding, and want
increasingly higher levels of personalized services.
All in all, it is difficult to handle efficiency and
security of airport processes and to balance all the
stakeholders’ interests.
Optimizing airport processes and infrastructure
therefore becomes very important. Desired models
are expected to be able to analyse the performance
of an airport system, plan resources for a given
future flight schedules, assist in planning changes
and determine effects on the overall level of
services. Individual-based models allow for a
scientifically reliable and detailed evaluation of the
behavioural processes, considering agent demands,
environmental perception and individual
109
Ma W., Fookes C., Kleinschmidt T. and Yarlagadda P..
Modelling Passengers Flow at Airport Terminals - Individual Agent Decision Model for Stochastic Passenger Behaviour.
DOI: 10.5220/0004055701090113
In Proceedings of the 2nd International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH-2012),
pages 109-113
ISBN: 978-989-8565-20-4
Copyright
c
2012 SCITEPRESS (Science and Technology Publications, Lda.)
interactions. Agent-based models of human
movement address that individual agent is
autonomous. That is, surpassing conventional
mathematical analysis, one simply instantiates a
population having some distribution of initial states.
Individual agents representing walking human with
initial states are situated in a representation of an
environment and interact with the environment and
themselves, acting out possible macroscopic
emergent behaviours. Agent-based models is
“solved” merely by executing it, where results are
dynamic at each simulation runs.
Modeling pedestrian dynamics gathers more and
more attention because of safety issues in public
facilities. However, current models hardly try to
hypothesize complex passenger decision-making
which would dictate the likelihood of individual
agents entering discretionary areas (Kleinschmidt et
al. 2011). In order to enable more intuitive stochastic
passenger movements, Ma et al. (2011) tried to
incorporate more detailed passenger information into
the simulation model, and therefore provide airport
operators with more realistic passenger flow models.
In this paper, we envisaged an individual agent
decision model aiming to represent intuitive
passenger behaviours. In section 2 we addressed the
significance of studying stochastic passenger
behaviours in air terminals. In Section 3 we propose
Bayesian artificial intelligent for agent decision
model. In section 4 we demonstrate the approach by
have a Bays net simulation case study. In section 5,
we summarize our conclusions and propose some
future areas of research.
2 STOCHASTIC PASSENGER
DYNAMICS
Airport is a complex system. It consists of many
standard sub-systems such as Check-in, Security,
Custom and Boarding. However among the intervals
of them, passenger dynamics are regarded as
stochastic and complex. Passengers might have
difference preferences to use any (if all) in-airport
facilities outside standard processes, for example
duty-free shops, cafe, telephone and bank. We
phrased such events as discretionary activities of
passengers. Since passenger activities outside
standard processing areas account for large
significance regarding safety and airport revenue
(Takakuwa and Oyama 2003, Ma et al. 2011), we
found it is important to study the discretionary
behaviours of passengers.
In order to understand the activities that passengers
use discretionary facilities, we first investigate what
sorts of discretionary facilities an airport terminal
might have and then investigate corresponding impact
factors which have affection on passengers’ choice to
use those facilities (Table 1).
Table 1: Discretionary facilities and corresponding impact
factors.
Discretionary facilities Impact factors
Relaxation facilities Physical tired
Technological self-
service kiosks
Technology preference
Information kiosks New users
Currency service Cash in need
Communication service
Business/entertaining
purposes
Dietary places Hunger level
Shopping places Desire to shopping
Discretionary facilities can be categorised into
the following parts according to utilities. They are
relaxation facilities, technological self-service kiosks,
information kiosks, currency service,
communication service, dietary places and shopping
places. It is true that passenger must use standard
processing facilities to get access through airport.
However, when utilising discretionary facilities,
passengers would usually spend plenty more time.
Ma et al. (2011) use a revised social force model
to simulate the basic motion of passengers in airport
terminals. It enables passenger agents in the
simulation can avoid collision by the repulsive
forces.
Advance path choosing of each passenger agent
are governed by one of the artificial intelligence
decision theories. We choose Bayesian belief
network as the tool for the study in this paper (Kevin
B. K. and Ann E. N., 2011). Bayesian belief
networks is used to generate the possibility that
passengers choose to go to a certain service facility,
which is an innovation comparing to conventional
passenger flow models who pre-assign the
probabilities. We aim to find the relationship
between passenger traits and the possibility of using
certain service facilities. Besides basic traits of
passengers, such as age, gender, we also devise the
advanced traits of passengers in Section 3. They can
be inferred from basic traits within the graph model
of Bayesian networks, and will be used for
individual decision making in our simulation
environment.
Bayesian inference computes the posterior
probability by conditioning, according to the rule of
Bayes. Advanced traits stand for mental preferences
of passengers, which mean that passengers could
SIMULTECH2012-2ndInternationalConferenceonSimulationandModelingMethodologies,Technologiesand
Applications
110
have sorts of probabilities to use different facilities
when they need to make decision. To compute the
posterior probabilities that a passenger would prefer
to use a certain in-airport facility, we suppose a
series of advanced trait which can be used to
indicate preferences of passengers. The Bayesian
network model is illustrated in detail at the next
section.
3 AGENT DECISION MODEL
The supposed agent decision model was devised
aiming to explain the complex stochastic behaviour
of passenger’s motion. Fig 1 shows the model
framework. In order to tackle the probabilities of
passengers choosing to use specific sorts of in-
airport facilities, we use Bayesian brief networks to
infer certain types of passengers. For example, if a
passenger who are a visitor and travel through the
airport firstly, he/she would be regarded as a “desire
shopping” passenger. Passengers who belong to this
type have a great possibility to use duty-free shops
as long as simulation-based components permit.
Simulation-based components are currently defined
as two parts: planned time and endurable walking
distance. Planned time refer to the time left till
boarding for departure process. Whether a passenger
will go to duty-free shops depends on if there is
enough time left to get on board. For inbound
however passengers seem have no restriction on
time. They may stay at airport any longer as they
wish. Endurable walking distance is parameter
which defines the normal longest distance a
pedestrian can walk along. Currently very few study
reveals walking distance issue at airport. We took a
reference of a survey of walking distance guidelines
used by North American companies, which
addressed that the value is between 400m and 800m
(Walking Distance Research – TOD Committee
http://www.fairfaxcounty.gov/planning/tod_docs/wa
lking_distance_abstracts.pdf accessed 2 March,
2012).
Figure 1: Agent decision model framework.
Passenger categories can be specified in terms of
the six advanced traits (Fig 2). Different passenger
categories are inferred through the devised Bayesian
brief networks. Parent nodes are the basic traits of
passengers which are not difficult to be found from
the information of passengers’ air tickets. We
investigated the ticket information and have the
major seven so as to represent them as the seven
basic trait nodes.
Figure 2: Inferring advanced traits of a passenger.
Table 1 explains the data type of the seven traits
in our simulation. For example, the trait “Age” is
calculated based on an equation that if the
registration information on the air ticket is over than
60 we deem the passenger as old, otherwise we
simply regard the passenger is young. We also can
acquire information about whether a passenger is a
frequent flyer who must use the airport more than
three times. We give frequency of travel a Boolean
data type. Other passengers who haven’t use the
airport before or use the airport only a couple of
times get the trait as non-frequent flyer. At this stage,
generating the trait value for passenger agent in our
simulation are due to the help of expert extrapolation
and historical data.
Table 2: The basic traits of passengers.
Traits category Data type Example
Age Boolean Old/Young
Gender Boolean Male/Female
Baggage Integer 0,1,2
Nationality Boolean Local/Foreigner
Travel Class Boolean Business/Economy
Frequency of
Travel
Boolean Frequent flyer/ non
frequent flyer
Travel group
size
Integer 0,1, ... , n (n >0, n is
an integer)
Bayesian networks are used here to infer the six
advanced traits for each passenger agent. Basically,
it calculates the conditional probabilities of
advanced traits. For example, in Fig 3, we select
four major nodes of basic traits to infer the
ModellingPassengersFlowatAirportTerminals-IndividualAgentDecisionModelforStochasticPassengerBehaviour
111
conditional probabilities that a passenger would
have this kind of proportion preference to use in-
airport shop facilities. In the same theory, we also
can acquire the conditional probabilities of the other
five advanced traits of an individual passenger. All
passenger agents can possess any or at least one type
of the advanced traits. The whole value representing
advanced traits of a passenger agent are calculated
and stored at the first beginning when a passenger
agent is generated in the simulation environment.
Figure 3: Bayesian network to infer desire shopping trait.
We also need to consider the utilities that a
passenger agent makes a decision to use a series of
service facilities. In order to make an intuitive
simulation of passenger motion in an airport
terminal, we add the components, planned time and
endurable walking distance, into the agent decision
model. Fig 4 illustrates the framework of the
decision-making model. Bayesian network is used to
infer the passenger preferences, which represent that
a passenger is inclined to use certain service
facilities. Simulation components part limits
unreasonable behaviours of passenger in case
passengers miss their flight. The decision graph
calculates the utility that a passenger chooses to use
a specific service facility. It provides the highest
utility results for output to guild a passenger agent to
execute the most feasible action.
Figure 4: Decision making with utility estimation.
In the simulation, basically there are four major
decision points where passenger would behave
autonomously based on the results of their Bayesian
network inferences. They are decision points before
check-in process, after check-in and before security
process, after security and before custom, and after
custom and before gates. The higher results of
utilities in the decision graph denote a passenger will
choose to use a certain facility first. For example, at
a decision point, as long as duty-free shops are
contained in the following possible interval
destinations, passengers who can fulfil their desire to
shopping and also are able to board on time would
choose to use duty-free shops first. They might walk
as many shops as possible as long as the endurable
walking distance is satisfied. Otherwise the
passengers choose to rest on the lounge area.
4 CASE STUDY
Our simulation model takes the Brisbane
international departure terminal as a case study. We
aim to validate the devised agent decision model
through the case study. Fig 5 shows parts of the
simulation within the check-in hall. The blue areas
stand for cafe and restaurants. The red areas
represent shops. Passengers can behave
discretionary activities before and after check-in
process. The probabilities of using any discretionary
are inferred through the devised agent-decision
model.
Figure 5: The simulation environment.
The simulation is able to simulate the whole
departure process. Passengers’ discretionary
activities happen at both check-in hall and gate
lounge. The dwell time that passenger stay at various
discretionary facilities is calculated and put in a
statistics graph (Fig 6). The longest dwell time in
SIMULTECH2012-2ndInternationalConferenceonSimulationandModelingMethodologies,Technologiesand
Applications
112
discretionary facilities during the whole departure
process is about one hour. The average value is a bit
above 10 minutes.
Figure 6: Average utilisation of discretionary facilities.
5 CONCLUSIONS
Passenger dynamics is becoming an important issue
in the study of service rate within built environment
such as transportation hub which includes airport
terminals. The paper demonstrates a devised agent-
decision model which can acquire the results of
utilisation of discretionary facilities. For the future
work, since individual passenger is programmed as
single agent, it is also able to record other possible
behaviours, such as how many shops passengers
walked through and recording the walking routes so
as to facilitate space design and estimation.
ACKNOWLEDGEMENTS
This research forms part of the work undertaken by
the project “Airports of the Future” (LP0990135)
which is funded by the Australian Research Council
Linkage Project scheme. The authors also
acknowledge the contributions made by the many
aviation industry stakeholders also involved in this
project, particularly for Brisbane Airport
Corporation who have permitted the use of their
terminal for this case study. More details on
“Airports of the Future” and its participants can be
found at www.airportsofthefuture.qut.edu.au.
REFERENCES
Andreatta, G., L. Brunetta, et al. (2007). "Evaluating
terminal management performances using SLAM: The
case of Athens International Airport." Computers &
Operations Research 34(6): 1532-1550.
Fu, X. and A. Zhang (2010). "Effects of Airport
Concession Revenue Sharing on Airline Competition
and Social Welfare." Journal of Transport Economics
and Policy (JTEP) 44(2): 119-138.
Gatersleben, M. R. and S. W. Van der Weij (1999).
Analysis and simulation of passenger flows in an
airport terminal. Simulation Conference Proceedings,
1999 Winter.
International Air Transport Association (IATA) (1997):
‘IATA Policy on User Charges Aspects of the
Commercialization of Airports and Air Trac Service
Entities’, International Air Transport Association,
Montreal.
Kevin B. Korb, Ann E. Nicholson, 2011. Bayesian
Aritificial Intelligence, CRC Press Taylor & Francis
Group. NW, 2
nd
edition.
Kleinschmidt, T., X. Guo, et al. (2011). Including airport
duty-free shopping in arriving passenger simulation
and the opportunities this presents. Proceedings of the
2011 Winter Simulation Conference, Phoenix, AZ,
USA.
Ma, W., T. Kleinschmidt, et al. (2011). Check-in
processing : simulation of passengers with advanced
traits. Proceedings of the 2011 Winter Simulation
Conference, Phoenix, AZ, USA.
Takakuwa, S. and T. Oyama (2003). Modeling people
flow: simulation analysis of international-departure
passenger flows in an airport terminal. Proceedings of
the 35th conference on Winter simulation: driving
innovation. New Orleans, Louisiana, Winter
Simulation Conference: 1627-1634.
ModellingPassengersFlowatAirportTerminals-IndividualAgentDecisionModelforStochasticPassengerBehaviour
113
