This is the authors version of a work that was submitted/accepted for pub-

lication in the following source:

Cheng, Lin, Reddy, Vikas, Fookes, Clinton, & Yarlagadda, Prasad K.D.V.
(2014)
Impact of passenger group dynamics on an airport evacuation process
using an agent-based model. In
Akhgar, Babak & Arabnia, Hamid R. (Eds.)
2014 International Conference on Computational Science and Computa-
tional Intelligence, IEEE Computer Society, Las Vegas, Nevada, USA, pp.
161-167.

This file was downloaded from: https://eprints.qut.edu.au/69071/

c Copyright 2014 IEEE

Personal use of this material is permitted. Permission from IEEE must be obtained for all
other uses, in any current or future media, including reprinting/republishing this material
for advertising or promotional purposes, creating new collective works, for resale or redis-
tribution to servers or lists, or reuse of any copyrighted component of this work in other
works.

Notice: Changes introduced as a result of publishing processes such as

copy-editing and formatting may not be reflected in this document. For a
definitive version of this work, please refer to the published source:

https://doi.org/10.1109/CSCI.2014.111
 Impact of Passenger Group Dynamics on an Airport Evacuation Process Using an
Agent-Based Model

Lin Cheng, Vikas Reddy, Clinton Fookes, Prasad K.D.V Yarlagadda

Queensland University of Technology
Brisbane, Australia
l7.cheng@student.qut.edu.au

AbstractThe safety of passengers is a major concern to expensive and time-consuming. As a result, many of the
airports. In the event of crises, having an effective and efficient phenomena and laws during an evacuation are only carried
evacuation process in place can significantly aid in enhancing out by model simulations [2, 3, 5]. Although it has been
passenger safety. Hence, it is necessary for airport operators to proven that social interactions greatly influence crowd
have an in-depth understanding of the evacuation process of behaviours and decision making, far too little attention has
their airport terminal. Although evacuation models have been been paid to group dynamics when developing passenger
used in studying pedestrian behaviour for decades, little flow models under both normal and emergency conditions
research has been done in considering the evacuees group [4, 6-9].
dynamics and the complexity of the environment. In this
paper, an agent-based model is presented to simulate Different from other building environments, an airport is
passenger evacuation process. Different exits were allocated to considered as a complex system that comprises multiple
passengers based on their location and security level. The stakeholders and social interactions [10]. For example,
simulation results show that the evacuation time can be before boarding the flight, passengers are required to pass
influenced by passenger group dynamics. This model also mandatory processes which include check-in, security
provides a convenient way to design airport evacuation process and Customs. Therefore, the security level varies in
strategy and examine its efficiency. The model was created the airport, which needs special consideration during the
using AnyLogic software and its parameters were initialised evacuation. To model pedestrian behaviour in such a
using recent research data published in the literature.
complex environment, the agent-based modelling technique
Keywordsgroup dynamics; agent-based model; airport;
is one of the best choices. In agent-based models, pedestrians
evacuation; simulation follow some pre-determined rules of behaviour, which allow
pedestrian-agents in the model system behave naturally and
autonomously. This unique characteristic makes agent-based
I. INTRODUCTION modelling particularly suitable for study of pedestrian
In recent years, air passenger traffic is growing at a behaviour in complex environments [11].
steady pace despite difficult global economic environment
This paper aims to evaluate the impact of group
[1]. Besides managing the needs and expectations of
dynamics on passenger flow during the evacuation process in
passengers, the airport operators also spend significant
an international departure terminal using the agent-based
resources and time in ensuring the safety of passengers who
modelling method. The remainder of the paper is organised
use their airports. In the light of recent terrorist incidents and
as follows. Section 2 reviews previous work related to
turmoil, this has become a top priority. In such emergency
pedestrian evacuation. Section 3 demonstrates the
situations, having an effective evacuation procedure could
construction and configuration of the agent-based evacuation
substantially mitigate the effect of threat and associated risk
model in the context of an international airport. Section 4
for the passengers and staff. Hence, an in-depth
provides the simulation results and analysis, while Section 5
understanding of the evacuation procedure, specific to each
summarises the major findings.
airport in the context of its existing infrastructure is of
utmost importance. Such knowledge also aids airport
operators to analyse the impact of the incident on other II. RELATED WORK
operations of the airport. Gwynne, et al. summarised 22 different evacuation
In previous studies of pedestrian evacuation, a vast models in their review [12]. Based on the nature of model
majority of researchers chose to use computer modelling application, those models are categorised into three different
techniques instead of the full-scale evacuation practice [2-5]. categories: optimization, simulation and risk assessment.
The major concerns of a real evacuation trial are: (1) the Optimization models try to find out the optimal evacuation
potential threat of injury to volunteers; (2) the lack of realism path, exit or flow characteristic, simulation models tend to
reaction during experiment (for example, the arising stress demonstrate the behaviour and movement observed in the
and panic); (3) the limit of reproducible results in repeat evacuation, while risk assessment models attempt to define
experiments; and (4) full-scale evacuation can be too potential hazards and bottlenecks in the evacuation process.
 Santos and Aguirre also presented a critical review of to the agent-based mindset, while the pedestrian movement
emergency evacuation simulation models [4]. They pointed is governed by the customised social force algorithm.
out that one common shortcoming of the reviewed models
In this paper, a new evacuation model which incorporates
lay in the absence of inclusion of social psychological
group dynamics will be introduced. The model is created
relevant group level characteristics. However, they also
using the agent-based model and simulates an evacuation
noticed that in some extreme situations where mass
event in an airport. It is assumed that the panic behaviour
behaviour exists, most potential evacuees do not have
does not exist in the evacuation so that the pedestrian group
enough opportunities to interact with their fellow group
behaviour can be preserved and analysed.
members, thus in those situations, the distinction between
group and individual level evacuation behaviour is less
meaningful. III. METHODOLOGY
Bonabeau [11] summarised the benefits of the agent- In an agent-based model, three key elements need to be
based model: (1) agent-based modelling (ABM) captures identified and modelled: agents, their environment, and their
emergent phenomena; (2) ABM provides a natural interactions with other agents and the environment [14]. In
description of the system; (3) ABM is flexible. These addition to this, Gwynne, et al. suggests that it is essential to
advantages make agent-based modelling ideal for simulating consider the configurational, environmental, procedural and
evacuation process. By using an ABM for fire escape, the behavioural aspects when proposing evacuation model [3].
author demonstrated how a column in front of the emergency In the real-world, the occurrence time of most
exit unexpectedly reduced the injury and increased the speed emergencies is unpredictable. Hence, in our model the point
of the pedestrian flow. The simulation result is verified by of triggering of emergency needs to be set externally by the
real-world experiments and indicated that the ABM can user rather than predefining it. It provides more flexibility to
capture the emergent phenomenon in a natural way. the model. This implies the model not only simulates the
Based on an enhanced cellular automation model, evacuation process, but also the normal airport operations as
Schultz, Lehmann [13] proposed a stochastic model to well. This section describes the setup of the evacuation
evaluate pedestrian dynamics under emergency cases in model by (1) introducing the airport environment and
airport terminals. They stated that airports are divided into passenger behaviour under non-emergency situation; (2) the
public and non-public area. Thus different security levels are configuration and procedure during the evacuation process;
required. They also suggested that a managed guidance and (3) passenger behaviour during evacuation.
system is necessary during the emergency situation, because
in a static guidance system, the pre-defined routes cannot be A. Airport Environment and Pedestrian Setting
guaranteed to be safe for evacuees. The model environment is an international airport
Zheng, Zhong [2] discussed the advantages and departure terminal, which is divided into landside and
disadvantages of seven evacuation modelling techniques. airside. The landside of the terminal is open to the public,
Those methods include cellular automata models, lattice gas while the airside of the terminal is only accessible for
models, social force models, fluid dynamic models, agent- passengers. Fig. 1 illustrates a high-level description of
based models, game theory models, and approaches based on passenger departure processes in the model. Passenger
experiments with animals. They concluded that all agent- activities are categorised into processing and discretionary
based models are microscopic. They are more activities [15]. Processing activities are mandatory for
computationally expensive compared to other models but passengers before boarding the plane. On the landside of the
have the ability to model heterogeneous humans. They terminal, passengers check-in for their flights, and pass
pointed out that a new trend in crowd evacuation models is through security check and Customs before entering airside
based on the combination of multiple approaches because of and boarding. Discretionary activities are considered as any
the complexity of pedestrian behaviour. In this paper, a other activities undertaken by passengers during non-
similar hybrid approach has been adopted in the proposed processing time [15, 16]. It can happen between two
model in which heterogeneous agents are created according

Landside Airside

Check-in Security Customs Boarding

Discretionary Activities

Fig. 1. Airport departure processes.

sequential mandatory activities as shown in Fig. 1. Examples In our simulations, we presume there are three security
of discretionary activities in the proposed model include levels (these could be adapted based on the operating
random walking, store browsing, having food and using conditions of the airport). Passengers who have not been
other airport services. Retail shops and airport services are examined by the security personnel are categorised as having
located at both landside and airside to emulate the real-world security level 1 status; passengers that passed security but
scenario. not the Customs have security status level 2; passengers that
pass both security and Customs possess security status level
Pedestrians in the model are categorised into passengers 3. In our model, it is assumed that only certain exits are
and wavers. Passengers are those who will board on the accessible to people depending on their security level status
plane, while wavers are fellow companions, who accompany as described below.
the passengers to the airport but do not board on the flight.
The model takes the departure flight schedule and passenger The landside of the terminal is the public area. The
number for each flight as inputs. The attributes of the agents crowds on the landside are treated to be on security level 1,
such as group size, speed, flight schedule and class, and along with all outgoing passengers who have not cleared the
shopping preference are initialised. The interactions between security check. They will choose one exit among the three
the agents and the environment are defined. Based on the located on level 4 that has the minimum walking distance
flight schedule, the agents are appropriately introduced in the while evacuating the airport. Situations are more complex on
environment. Detailed micro-activities in each process the airside of the terminal. On the airside, there are two
(check-in, security and Customs) is modelled based on mandatory processes: security and Customs, and different
observational data collected by the Human System team of security levels are imposed on them. Passengers belonging to
the Airports of the Future project [15]. security level 2 will evacuate through exit 2 on level 3, and
passengers with security level 3 will evacuate through exit 3
B. Configuration and Procedure During Evacuation on level 3.
Fig. 2 shows the layout of the airport departure terminal In the simulation, once the emergency ceases passengers
used in the simulation. As can be seen from the terminal returning into the terminal will keep their security level
layout, three emergency exits (marked as red circles) are status intact, so that they can continue to finish their
available on both landside (level 4) and airside (level 3) of remaining processes rather than doing them from the
the terminal. In the event of an evacuation, passengers will beginning. However, the simulation is flexible and different
be notified by an emergency alarm, and then they will make policies can easily be implemented and tested (e.g. the policy
their way to the nearest exit under the guidance of building that all passengers must be re-scanned on entry, regardless of
wardens and airport staff. Passengers will remain at their security levels when they are evacuated). TABLE I.
designated assembly points until it is safe for them to re- summarises the corresponding exits for pedestrian with
enter the building. different security levels.

(a) (b)
Fig. 2. Airport environment defined in our simulation. The exits are marked as red circles. (a) check in area and retail (landside); (b) Security, Customs,
Boarding and retail (airside).

TABLE I. EXITS ASSIGNED FOR PASSENGERS OF DIFFERENT SECURITY LEVELS.

Security Level 1 2 3
Security checked;
Domain Security unchecked Customs checked
Customs unchecked
Exit 1,2,3 on
Emergency Exits Exit 1 on Level 3 Exit 2 on Level 3 Exit 3 on Level 3
Level 4
 Fig. 3. Typical response followed by passengers during evacuation.

(a) (b)
Fig. 4. Similar to Fig. 2, but illustrates the environment in 3D, along with passengers.

C. Behaviour Responses to Emergency Evacuation IV. RESULT AND ANALYSIS

The likely behavioural response of the evacuees is The 3D simulation environment of an international
essential to the model. There are two levels of behavioural airport departure terminal is shown in Fig. 4. The model is
responses: global and local [17]. The global behaviour level built on AnyLogic 6.8 platform to simulate the daily
outlined the general escape strategy. At the start of the operation of the airport. Activities of each agent in the
evacuation alarm, passengers and airport staff will spend system were updated successively according to preset
some time to respond to the signal. After recognising the characteristics within a discrete-event structure of AnyLogic
situation, passengers need to decide the evacuation option, simulation software. The model is validated by using the face
for example, available exits and the closest distance to the validation and statistical validation methods. The face
exit. Passenger groups with different opinions may spend validation is based on the knowledge of domain experts. The
more time on discussion. This period of time is described as 2D/3D animation of the model not only gives an overview of
response time. After making the decision, passengers will the simulation, but also aids in face validation. In the
move towards the chosen exit, during this period, the animation, passenger behaviour such as walking, waiting and
movement time is recorded. Due to potential congestion in grouping can be directly seen. The interaction between
front of the exits, it is possible that passengers need to wait passengers and the airport environment in different areas like
before they make their way out. Another reason that could check-in, security and border processing can also be
lead to longer waiting times is that passengers travelling in analysed. By comparing the visualised crowd behaviour with
groups will wait for fellow passengers to regroup around the the experience of airport experts, the processes and structures
exit area. They generally ensure that all group members are of the model can be assured.
safe and would like to evacuate together.
Statistical validation is conducted by comparing key
The evacuation time for a pedestrian (passenger) is figures generated from the simulation model with the
defined as the time when the evacuation alarm set off to the observation data from the airports. Since the collection of
time that the pedestrian leaves the exit. Typical steps observation data from a real airport environment is extremely
passengers take during an evacuation can be seen from Fig. 3. labour intensive and time consuming [16], available data are
On the local behaviour level, based on individuals limited in the areas of interests such as average queuing time
personal attributes such as age and travel purpose, and total dwell time at each departure process. After the data
pedestrians have different degrees of knowledge about what comparison, parameters in the model are calibrated in order
to do when evacuating. Therefore, people in the model have to adjust the model output data to the observation data within
varying response times to the evacuation alarm. During the tolerable differences (differences between average
movement, passengers in the same group will compromise observation times and simulated times less than 2 minutes).
their speed to the slowest group member in order to travel at TABLE II. compares the actual and the simulated time
the same speed. obtained at each process. It shows that the simulation is
reflective of the actual situation.
TABLE II. COMPARISONS OF QUEUE AND DWELL TIMES AT CHECK-IN, Fig. 5 compares the time distribution of the evacuation event
SECURITY AND CUSTOMS BETWEEN THE ACTUAL TIME AND THE
SIMULATION.
on both levels between the setting of passengers travelling
(1) alone; (2) in groups. It is shown that the total evacuation
Queue times [min] Dwell times [min] time of passengers travelling in groups is longer than that of
Domain
Actual Simulation Actual Simulation passengers travelling alone. On both level 4 and level 3,
Min 0.58 0.48 1.95 3.60 passengers in groups spend 146.18 and 173.45 seconds to
Check-in Max 42.81 56.85 53.56 62.00 finish the evacuation. The figures are 93.68 and 146.76 for
passengers travelling alone, which are 36% and 15% shorter
Average 12.88 12.58 16.65 18.76
in comparison. Distinct time differences can be found on all
Min 1.23 0.74 1.90 3.28 sub-events of the evacuation process on level 4 as well as the
Security Max 17.09 8.39 21.06 20.02 response time and waiting time on level 3. This result of
Average 3.75 3.53 6.88 7.86
response time indicates that the initial response of passengers
who travelling in groups is slower than those travelling
Min 0.33 1.16 0.55 2.13 alone. For it can take longer time for passengers in a group to
Customs Max 15.46 30.22 18.58 36.40 communicate with each other and make decisions in
Average 4.80 5.57 6.00 7.50 response to the evacuation signal. The waiting time for
passengers who travel alone is mainly caused by the
congestion in front of the exit. While for group travellers, the
In order to demonstrate the general behaviour of waiting time is not only due to the congestion, but also the
passengers and ensure the reliability of the experimental time associated with regrouping. Therefore, the waiting
results, the evacuation event is set at 7:30 AM, one of the time for passengers travelling in groups is reasonably higher.
peak times of the day to collect more sample data. TABLE
On both level 4 and level 3, passengers spent the majority
III. summarised the distribution of passengers in the airport
of their evacuation time on the evacuation movement.
departure terminal when the evacuation starts. The results are
Because of the group dynamic, larger groups are supposed to
collected from five experiments. On average, there are
move slower than small groups or individual travellers [4,
approximately 1000 passengers in the simulation system at
18]. This behaviour is well illustrated on level 4. Passengers
7:30 AM. In the condition of passengers travelling alone, on
travelling in groups on level 4 spend approximately 20
average 218 passengers (22% of the total passengers) are
seconds longer in moving during the evacuation. However,
found on level 4, while 782.2 passengers (78%) are on level
movement times under the two different settings on level 3
3. These figures changed to 361.2 (36%) and 644.6 (64%)
are very close. A possible explanation for such a
under the condition that passengers are travelling in groups.
phenomenon is that the pathways to the exits on level 3 are
One of the most distinctive characteristics of an agent-based
narrower than those at level 4 (as can be seen from the
model is that agents are able to act autonomously in the
highlighted areas in the density map illustrated in Fig. 6).
simulation environment. This advanced feature strongly
Severe congestion was observed all along passageways
reflects the real-world human behaviour. As a result, even at
through to exit on level 3. Thus, passengers travelling alone
the exact time-point of several experiments, the agent
had to slow down due to congested passageways and their
number in the system, agents positions and their
speed was comparable to those travelling in groups. On the
undertaking activities can be different.
other hand, passengers on level 4 had more open space,
The differences in building layouts, passenger numbers which allowed those individual travellers to advance towards
and activities require the evacuation process to be analysed the exit quickly.
separately on level 4 (the landside) and level 3 (the airside).

TABLE III. DISTRIBUTION OF AGENTS IN THE AIRPORT TERMINAL UNDER THE SETTING OF PASSENGERS TRAVELLING: (1) ALONE; AND (2) IN GROUPS.

Number of agents in the experiment

Passenger travelling alone Passenger travelling in groups
Exp No. level4 level 3 Total level4 level 3 Total
1 219 771 990 343 654 997
2 222 768 990 378 626 1004
3 212 789 1001 328 678 1006
4 217 786 1003 381 629 1010
5 220 797 1017 376 636 1012
Avg. 218 782.2 1000.2 361.2 644.6 1005.8
 (a) (b)
Fig. 5. Average evacuation time of passengers on: (a) level 4, landside; and (b) level 3, airside for the two different settings.

(a) (b)
Fig. 6. Pedestrian density map during the evacuation process. (a) level4; (b) level 3.

V. CONCLUSIONS
However, a few limitations of this pilot study need to be
The simulation of the evacuation process in the airport acknowledged. First, the proposed model is not designed for
terminal shows that the agent-based model can be used to the extreme evacuation situation. Under extreme cases, there
analyse pedestrian group dynamics in a complex is no guarantee that the pre-defined escaping routes are safe
environment. Based on passengers locations in the airport, for the evacuees. Second, evacuation subjects in the model
three security levels are differentiated, which require are all passengers. However, in the real-world, there are large
passengers to evacuate through different exits. The numbers of airport staff that need to be considered as well.
simulation results suggested that passengers with group As part of future research, we would like to consider a
dynamics spend longer time in making decisions, moving to phased evacuation approach i.e. only areas directly
the exits and waiting for other group members during the threatened will be evacuated first and areas at lesser risk will
evacuation. be evacuated later. Furthermore, different exit strategies
This simulation technique prevents the potential risks in employed at various airports could also be trailed using this
real practical trials and reduces research expense. Moreover, framework. The dissemination of evacuation information
the simulation results provide valuable information such as among passengers and further addition of attributes to agents
how passengers react to an evacuation signal, which route to (such as age, gender, spatial cognition) will also be explored.
choose in the evacuation and the average time for passengers
to finish the evacuation. The simulation is also to report ACKNOWLEDGMENT
congestions through the 3D visual demonstration during the
evacuation. The evacuation model offers an expedient way The Airports of the Future research project is supported
for airport managers to propose and test evacuation plans. under the Australian Research Councils Linkage Projects
Given the information of flight schedule and passenger funding scheme (LP0990135). The authors would like to
number, the evacuation simulation can be run at any time of thank Philip Kirk who provided some of the model
day and the simulation results will provide valuable parameters used in the simulation. The authors also
information for them to respond proactively to any potential acknowledge the contribution of the industry stakeholders
congestion. involved in this project. More details on the project can be
found at: www.airportsofthefuture.qut.edu.au.
 REFERENCES
1. IATA, Annual Report 2013. 2013, International Air
Transport Association: Beijing. 11. Bonabeau, E., Agent-based modeling: Methods and
2. Zheng, X., T. Zhong, and M. Liu, Modeling crowd techniques for simulating human systems.
evacuation of a building based on seven Proceedings of the National Academy of Sciences
methodological approaches. Building and of the United States of America, 2002. 99(Suppl
Environment, 2009. 44(3): p. 437-445. 3): p. 7280-7287.
3. Gwynne, S., et al., A review of the methodologies 12. Gwynne, S., et al., A review of the methodologies
used in the computer simulation of evacuation from used in evacuation modelling. Fire and Materials,
the built environment. Building and Environment, 1999. 23(6): p. 383-388.
1999. 34(6): p. 741-749. 13. Schultz, M., S. Lehmann, and H. Fricke,
4. Santos, G. and B.E. Aguirre, A critical review of Pedestrian Dynamics in Airport Terminals
emergency evacuation simulation models. 2004. Considering Emergency Cases. Proceedings of
5. Schadschneider, A., et al., Evacuation Dynamics: Internation Council of the Aeronautical Sciences,
Empirical Results, Modeling and Applications, in 2006.
Extreme Environmental Events, R.A. Meyers, 14. Macal, C.M. and M.J. North, Tutorial on agent-
Editor. 2011, Springer New York. p. 517-550. based modelling and simulation. Journal of
6. Qiu, F. and X. Hu, Modeling group structures in Simulation, 2010. 4(3): p. 151-162.
pedestrian crowd simulation. Simulation 15. Kraal, B.J., V. Popovic, and P.J. Kirk, Passengers
Modelling Practice and Theory, 2010. 18(2): p. in the airport : artefacts and activities, in Design :
190-205. Open 24/7. 2009.
7. Singh, H., et al., Modelling subgroup behaviour in 16. Livingstone, A., et al., Understanding the airport
crowd dynamics DEM simulation. Applied passenger landside retail experience, in DRS 2012
Mathematical Modelling, 2009. 33(12): p. 4408- Bangkok Research: Uncertainty, Contradiction
4423. and Value, P. Israsena, J. Tangsantikul, and D.
8. Ma, W., et al., Modelling passenger flow at airport Durling, Editors. 2012, Department of Industrial
terminals : individual agent decision model for Design, Faculty of Architecture, Chulalongkorn
stochastic passenger behaviour, in SIMULTECH University: Chulalongkorn University, Bangkok.
2012. 2012: Rome. 17. Filippidis, L., et al., Representing the Influence of
9. Cheng, L., et al., Analysis of Passenger Group Signage on Evacuation Behavior within an
Behaviour and Its Impact on Passenger Flow Evacuation Model. Journal of Fire Protection
Using Agent-Based Model, in IC-MSQUARE. Engineering, 2006. 16(1): p. 37-73.
2013: Prague, Czech Republic. p. 4. 18. Schultz, M., C. Schulz, and H. Fricke, Passenger
10. Wu, P.P. and K. Mengersen, A review of models Dynamics at Airport Terminal Environment.
and model usage scenarios for an airport complex Pedestrian and Evacuation Dynamics 2008, 2010:
system. Transportation Research Part A: Policy and p. 381-396.
Practice, 2013. 47: p. 124-140.