Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Aishwarya Thipparthi
Supervised by
December 2016
ii
Dedication
Acknowledgments
Abstract
Agent based simulation of crowd evacuation during an emergency
Aishwarya Thipparthi
Concert venues such as auditoriums are usually crowded and have a limited number of
exits and escape pathways. The efficiency of evacuations in case of emergencies depends
on proper design of the auditorium.Planning and testing of safe safe designs and practices
is difficult due to the multiple scenarios to be tested and the high costs involved. Computer
simulation is an effective way to evaluate the safety of a building. This project aims to
simulate the crowd evacuation of an auditorium by varying the design parameters.The areas
that accumulate high densities of people are at risk of causing a stampedes. They are
identified by painting density maps. The different design layouts are compared with each
other and overall safety of the designs is evaluated.
v
Contents
Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Problem definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.5 Roadmap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1 AnyLogic software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Pedestrian Library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.1 Basic elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.2 Pedestrian Statistics collection . . . . . . . . . . . . . . . . . . . . 8
2.2.3 Space Markup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Analysis elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Layout Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4.1 Layout 1: Single aisle with single exit . . . . . . . . . . . . . . . . 10
2.4.2 Layout 2: Single aisle with two exits . . . . . . . . . . . . . . . . . 11
2.4.3 Layout 3: Double aisle with two exits . . . . . . . . . . . . . . . . 11
2.4.4 Layout 4: Triple aisle with two exits . . . . . . . . . . . . . . . . . 11
3 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1 Pedestrian Flow implementation . . . . . . . . . . . . . . . . . . . . . . . 14
3.1.1 Layout1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1.2 Layout 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
vi
3.1.3 Layout 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.1.4 Layout 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.0.1 Maximum densities in Layout 1 . . . . . . . . . . . . . . . . . . . 24
4.0.2 Maximum densities in Layout 2 . . . . . . . . . . . . . . . . . . . 25
4.0.3 Maximum densities in Layout 3 . . . . . . . . . . . . . . . . . . . 26
4.0.4 Maximum densities in Layout 4 . . . . . . . . . . . . . . . . . . . 26
5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.1 Current Status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.3 Lessons Learned . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
vii
List of Figures
Chapter 1
Introduction
1.1 Background
A disaster can be defined as any dangerous event which brings with it significant human
and financial losses. There can be natural disasters such as snow storms, earthquakes,
tornadoes or man-made disasters such as fires, air plane crashes, collapsing buildings.[6]
In the former case, there is no way to prevent one from occurring, the only thing that is
under human control is how we respond to emergency situations. In the latter case, the loss
can be significantly reduced by effective planning and preparedness.
One of the issues that arises with safety planning is the cost associated with it as it is very
difficult to test the effectiveness of a system during a crisis. There are many parameters that
have to be taken into consideration in each scenario. Simulation is an elegant solution that
addresses most of the problems associated with safety planning and design with minimum
cost attached with it. It offers a way to design different scenarios and test the effectiveness
of the system in each case along with a control on the parameters.
1.2 Motivation
The ultimate goal of emergency management is contain the loss of human lives and prop-
erty damage.Venues such as auditoriums and stadiums usually seat a large number of peo-
ple and carry a high risk of stampedes and crushes during an emergency situation. This
project focuses on the simulation of evacuation of people from auditorium and thereby
evaluating the safety of the auditorium. It is essential identify the weakness in the design
2
such as high density areas which can cause bottlenecks. There certain approaches that can
be followed to achieve this: Decreasing the probability of disastrous incidents occurring
by making safer design choices. Placing hazardous installations at safe distances from the
population thereby reducing the impact of disasters. Creating properly tested plans which
can be utilized to evacuate people from area of danger should an emergency arise. [16]
Concert venues such as auditoriums are usually crowded and have a limited number of
exits and escape pathways. The efficiency of evacuations in case of emergencies depends
on proper design of the auditorium.Planning and testing of safe safe designs and practices
is difficult due to the multiple scenarios to be tested and the high costs involved. Com-
puter simulation is an effective way to evaluate the safety of a building. The simulation
technique,agent based modeling is an approach which involves a collection of autonomous
entities called ”agents” which interact with each other in the environment.[2]. Using these
concepts,this project aims to simulate the crowd evacuation of an auditorium by varying the
3
design such as the number of aisles and exits. With this project, the areas that accumulate
high densities of people are at risk of causing a stampede are identified by painting density
maps. The designs are compared and the safety of the design is evaluated.
Helbing et al[11] presented a model that is based on physics and uses social forces to
represent human behavior in situations of panic. Each individual is represented as a particle
with a particular mass and velocity in a certain direction. In paper Braun et al[3] describes
an agent based model to simulate virtual human crowds in emergencies based on the model
proposed in Helbing et al [11]. Ha et al[9] also utilize social forces model to study the
crowd behavior in evacuation of multi-floor buildings. In this model[9], the relation of the
building architecture is studied in relation to crowd evacuations in emergencies. Factors
such as door size of a room, sizes of exits, speed and friction coefficient are investigated to
improve the efficiency of evacuations. Camillen et al [4], study pedestrian behavior while
visiting and evacuating closed spaces, specified to a museum scenario. A multi-agent simu-
lation was developed by varying the crowd size and arrival times. Okaya et al[15] propose
a model for crowd evacuations which is based on the Belief-Desire-Intention model and
4
Helbing’s social forces model [11]. The Belief-Desire-Intent model is affected by human
relationships, and it was found that due to these interactions the evacuation takes a longer
time. Cheng et al [5], present a model that studies behavior of groups within crowds and in-
vestigates how these dynamics affect the overall pedestrian behavior. The model simulates
an airport environment and compares the passenger behavior with respect to groups. Zia et
al [19] investigate the effect of multiple exits in an evacuation scenario where individuals
are more likely to take an exit that is visible directly to them rather than follow a directional
guide.
The second area of research has been the crowd evacuation on urban roadways. Ahn et al
[1] present a hybrid model for simulating road network that combines the approaches of
micro based models and macro based models by balancing the level of details. Handford et
al [10] propose a model to simulate the behavior of drivers during evacuations by employ-
ing the social forces crowd model and using route similarity as a metric. Lucas et al[14],
model the emergency response of police, fire and medical teams of during the evacuation
of an urban roadway during a terrorist attack.
All the models discussed above follow the agent based modeling. EVANCET4 is a
flow based model where the environment is simulated as a network of nodes that repre-
sent physical structures[17]. In flow based models, social interactions are not considered
during modeling the behaviors. Cellular automata are another approach of modeling evac-
uations in which the space is discretized in individual cells[17]. In the simulation program
EGRESS, this technique is followed[17].
The next significant area of current research has been in the simulation of crowd behavior
during evacuations [18]. Chu et al [7] have used social behavior studies and incorporated
them into their evacuation models. Liang et al[13] propose a system in which information
is embedded into the virtual environment which will influence the crowd behavior. Helio-
vaara et al[12] follow a different approach in evacuation behavior, by presenting a model
that simulates people trying to avoid colliding with oncoming people by observing their
5
walking directions. Hence, simulation and particularly its application to practical problems
such as evacuation is a very active area of research.
1.5 Roadmap
In the chapter Design,the software AnyLogic that was used to perform the simulations and
its pedestrian library will be described. The layouts of the each of the designs for the differ-
ent experiments performed are also discussed. In the chapter Implementation, the method
used to define the pedestrian flow via flowcharts is explained. The execution of each model
and the density maps painted during the simulation are presented and explained. In the
chapter Analysis, the results of the simulation will be presented. The effectiveness of each
layout will be analyzed using the time vs. density and the maximum density recorded in
each case. In the chapter Conclusion, the results of the experiments and the analysis will be
summarized in the section Current Status. The features that couldn’t be incorporated in this
project due to the time constraints and interesting areas of research that could be explored
to expand this project are discussed in Future Work. The learning achieved through this
project will be described in the section Lessons Learned.
6
Chapter 2
Design
AnyLogic has graphical user interface which makes it easily understandable for begin-
ners. The visual environment also quickens the overall process of model development. The
models developed via the visual elements are fully mapped to java code and these programs
can be extended to standalone applications which is a very useful functionality.
AnyLogic provides built-in libraries that can be used for domain specific models. The
main domains where AnyLogic is applicable are market and competition, health care, sup-
ply chains and logistics, pedestrian flows, transportation & warehousing etc. AnyLogic
consists of a graphical interface that allows the user to build simple models with ease.
[8][2].
7
The pedestrian library is made of basic elements that act as the foundational blocks with
which the model is designed. They are described in the following sections.
PedSource
PedSource generates pedestrians and is usually the starting point of the pedestrian move-
ment. the flow of the pedestrians can be customized based on requirements. The pedes-
trians can be generated in groups or individually, the rate of arrival and the speed of the
pedestrians can be set in the parameters. The exact location of the point of origin of the
pedestrians can also be set by using coordinate points or area or a specific target line.
PedSink
PedSink is used to discard pedestrians. This is the opposite of PedSource and is generally
the end point of the flow. It has only one connecting ”in” port.
PedGoTo
decided by the system. This mode has been used in the current project.The Follow Route
can be used for user-defined paths.
PedWait
PedWait makes the pedestrians wait for a specified period of time in a particular location.
The location can again be defined based on target line,coordinate points or area. In this
project, the area mode has been chosen to make the pedestrians wait in the seating area of
the auditorium. The exact wait location within the area can be specified using attractors.
The pedestrians wait until the specified time expires or until the functionfree() has been
called by the user explicitly.
PedSelectOutput
This element can be used when there is more than one route the pedestrian can go follow.
It has one input and several output ports each of which are to be connected to a separate
route. The routing distribution can be done based on Condition or Probability
PedEnter
This element takes in the pedestrians generated in another location and can be used to
organize the pedestrian groups and set their parameters.
PedExit
This element is used to direct pedestrian flow to another location and remove them from
their current space.
One of the most important and useful features of the pedestrian library is the ability to
collect statistics based on the pedestrian density. This can be collected in two ways. Using
a Density Map or Pedestrian flow statistics
9
Density Map
Pedestrian density map paints the environment space dynamically once the pedestrians start
moving. It uses a logarithmic scale for the color scheme. The color depends on the density
and the areas with critical densities are painted in red. The critical density for this project
has been set at 1.5 pedestrians per square meter.
This element is used to collect statistics of pedestrians crossing a particular line. The
statistics can be collected based on right or left flow. The functions, traffic() is used to
calculate the number of pedestrians during the last hour and the function intensity() is used
to calculate the traffic per length of the line.
In addition to the flow and statistics elements described above, a major part of the design
involves marking the space in the simulation model and setting up the environment.
Walls
Walls are used to demarcate the space and set up boundaries.Pedestrians cannot pass through
the walls. They can be used to draw exterior walls or a rectangular wall which is com-
pletely inaccessible by the pedestrians. This project uses walls to define the auditorium
boundary,seats and the stage.
Target Line
Target lines are an important element of space mark up as they can be used as the destination
points of the flow chart. The other pedestrian element such as PedGoTo use target line as
the location.
10
Area
Area is a space that can be used to specify the location pedestrians appear and wait. This
has been used to create the seating area in the auditorium.
Data Set
The data set stores data in two dimensional double format and can update dynamically. The
x value of the data is set to time in this project and the y values as the densities. This is
used to plot the trends in density of the specified point.
As seen in Fig 2.1 ,in this layout, the seating area is divided by a single aisle that runs
through the middle. Each block of the seating seats 50 people. The exit is placed at the left
end of the auditorium.
11
As seen in Fig 2.2,in this layout, the seating area is divided by a single aisle that runs
through the middle. Each block of the seating seats 50 people. There are two exits placed
on the left and right side.
As seen in Fig 2.3,in this layout, the seating area is in the middle, and the aisles are on the
either side of the area. The exits are located at the end of the aisles.
As seen in Fig 2.4,in this layout, the seating area is divided by a single aisle that runs
through the middle. There are also two other aisles running on the either side of the seating
area towards the wall. Each block of the seating seats 50 people. There are two exits placed
12
Chapter 3
Implementation
In this chapter, the implementation of the pedestrian flow and the execution of the model is
explained with the help of screen shots.
3.1.1 Layout1
The first layout, single aisle with single exit, is defined as shown in the fig 3.3. Since all the
pedestrians are generated in two blocks of seating area,there are to be two flowcharts similar
to each other. Both the flows have different PedSource elements to generate pedestrians and
different PedSink elements to dispose them. However,the PedGoTo elements both direct
them to same target location.
Once the simulation is run, fig 3.1 shows the generation of the pedestrians and the
crowd moving to wait in the seating area.
Fig 3.2 shows the evacuation of crowd from the auditorium via the central aisle toward
the exit. After the waiting shown in fig3.1 the evacuation is triggered by clicking the
evacuate button. The densities of each stage are shown by the Density Map element. The
15
areas of high density are painted red. The current and maximum density values of two
points,one near the bottle neck seen at the end of the aisle and the other near the exit are
dynamically displayed while the model is running.
3.1.2 Layout 2
The second layout, the auditorium has a single aisle but two exits. It is defined as shown
in the fig 3.4. This is similar to the layout 1, with both of them having two PedSource and
PedSink elements. The difference is that one of the PedGoto elements has a different target
location. This simulates the pedestrians for both the exits.
The pedestrians waiting will be similar to fig 3.1 and once the evacuate button is clicked,
it will trigger the evacuation of the crowd by calling the free() function.
Fig 3.5 shows the evacuation of the pedestrians from the auditorium. The bottleneck at
the end of the aisle still is severe.
16
3.1.3 Layout 3
The third layout, the auditorium contains two aisle and two exits. The aisle are on the either
side of the seating area, as shown in the previous chapter.
The pedestrian flow for this layout is shown in fig 3.6 Since all the pedestrians are gener-
ated in a single block of seating area,there is only one PedSource generating all the 100
pedestrians.
The PedWait element causes the wait until the evacuate button is clicked. The PedS-
electOutput element is used to make the passengers have two different routes of evacuate
chosen probabilistically. The target locations of each route is one of the two exits. Finally
the PedSink elements dispose the pedestrians.
Once the simulation is run, fig 3.7 shows the generation of the pedestrians and the
crowd moving to wait in the seating area.
17
As soon as the evacuate button is clicked, the crowd starts to disperse. Fig 3.8 shows the
evacuation of crowd from the auditorium via the two aisles on the either side of the seating
area. The densities are painted accordingly by the Density Map element and the current
and maximum densities are collected at two points dynamically during the run time.
3.1.4 Layout 4
The fourth layout, the auditorium contains three aisle and two exits. Two of the aisles are
on the either side of the seating area and one through the middle of it, as shown in the
previous chapter.
The pedestrian flow for this layout is shown in fig 3.9. The seating area consists of left
and right blocks, hence there are two different PedSource elements generating 50 pedestri-
ans per block. These are then made to wait using the PedWait elements. Once freed, the
18
pedSelectOutput element is used to direct the pedestrians towards one of the three exits.
These routes are assigned to individual pedestrians probabilistically and can introduce an
element of chaos in the movement.
The pedestrian density map is uniform prior to the evacuation when the pedestrians are
waiting, as show in fig 3.10
The pedestrian evacuation is triggered by clicking the button, and fig 3.11 shows the
pedestrians evacuating from three aisle towards the exits. The density of the pedestrians is
painted by the density map element. The current and maximum densities at three different
points are dynamically displayed during the run time.
19
Chapter 4
Analysis
In this chapter, the experiments and their results are explained and compared with each
other. For each layout designed to be tested in this project, the capacity of the auditorium
was kept constant at 100 pedestrians. The variables were the aisles or pathways that the
pedestrians took to evacuate the building. The single aisle layout also has two sub-scenarios
with varying exits.
At the end of these experiments, the safety and vulnerabilities of each layout will be ex-
amined. This will be done by looking at the areas that have a risk of causing bottlenecks.
These areas are identified by looking at the density map of the auditorium, with red being
the critical density areas.
The experiments were run initially to identify which areas were at risk and then the current
and maximum densities at those specific points were captured dynamically and displayed
while the model was executing. The maximum density data was also stored into a data set
to analyze the densities. In the following section, the results of each layout are presented
and explained with the help of graphs.
The layout 1 performed very badly in comparison to the other layouts. The single aisle
design proved to be very unsafe, with a bottle neck appearing at the end of the aisle. The
critical density of the system was set at 1.5 pedestrians per square meter, but the maximum
densities in the layout surpass even this,making it extremely unsafe in emergency situations.
25
As seen in fig 4.1, after the evacuation begins at roughly 340 model time units(seconds),
the densities steadily rise and the peak density is seen at 2.5 pedestrians per square meter
which is very high compared to the critical density.
The layout2 outperformed layout1 by a very small margin,but the overall safety of this
auditorium design remains low as the bottle neck still carries a risk of very high densities.
The density vs. time plot of the model is shown in fig 4.2
As can be seen in fig 4.2, the density of the area steadily rises and crosses critical the
density of 1.5 pedestrians per square meter. The maximum density recorded for this layout
is 2.46 pedestrians per square meter which is almost as much as the layout 1. So this model
also has shown that an auditorium with single exit is still unsafe during an emergency.
26
The layout 3 performs slightly better than layout 1 or layout 2 which only have a single
aisle. This layout still shows that the areas near the exit get very congested during the
course of evacuation. The plots for the left and right exits are shown in fig 4.3 and fig 4.4
respectively.
As seen in fig 4.3,once the evacuation begins, the densities see a sharp rise and the peak
density is seen at 1.9 pedestrian per square meters.
The densities for right exit exhibit a similar trend also peaking at 1.9 pedestrians per
square meter as seen in fig 4.4
The layout 4 outperforms all the other layout designs by a large margin. The auditorium
design with three aisles sufficiently distributes the crowd in the space as per the simulation.
27
Figure 4.3: Density vs. Time plot of point near left exit of layout 3
The aisles get slightly congested as seen by the orange density map, but this is still less
than the critical density. Points near each of the three exits were measured for density and
the graphs plotted for three.
As seen in fig 4.5, after the evacuation begins, the densities increase initially and then
decrease and start to increase steeply. The maximum density for the left exit is found to be
0.93 pedestrians per square meter.
The graph seen in the fig 4.6 exhibits a similar trend to the fig 4.5 with a slightly higher
peak. The maximum density was found to be 0.94 pedestrians per square meter.
The density vs. time graph for center aisle of the layout 4,as seen in fig 4.7 shows a
slightly higher maximum density compared to the the other two points of measure with
1.15 pedestrians per square meter. But it can be noted that this is still lower than the critical
density of 1.5 pedestrians per square meter.
28
Figure 4.4: Density vs. Time plot of point near right exit of layout 3
Figure 4.6: Density vs. Time plot of point in right aisle of layout 4
Figure 4.7: Density vs. Time plot of point in the center aisle of layout 4
30
Chapter 5
Conclusions
Bibliography
[1] Nguyen Thi Ngoc Anh, Zucker Jean Daniel, Nguyen Huu Du, Alexis Drogoul, and
Vo Duc An. A hybrid macro-micro pedestrians evacuation model to speed up sim-
ulation in road networks. In International Conference on Autonomous Agents and
Multiagent Systems, pages 371–383. Springer, 2011.
[2] Andrei Borshchev. The big book of simulation modeling: multimethod modeling with
AnyLogic 6. AnyLogic North America Chicago, 2013.
[3] Adriana Braun, Bardo EJ Bodmann, and Soraia R Musse. Simulating virtual crowds
in emergency situations. In Proceedings of the ACM symposium on Virtual reality
software and technology, pages 244–252. ACM, 2005.
[4] Francesca Camillen, Salvatore Caprı̀, Cesare Garofalo, Matteo Ignaccolo, Giuseppe
Inturri, Alessandro Pluchino, Andrea Rapisarda, and Salvatore Tudisco. Multi agent
simulation of pedestrian behavior in closed spatial environments. In Science and
Technology for Humanity (TIC-STH), 2009 IEEE Toronto International Conference,
pages 375–380. IEEE, 2009.
[5] Lin Cheng, Clinton Fookes, Vikas Reddy, and Prasad KDV Yarlagadda. Analysis
of passenger group behaviour and its impact on passenger flow using an agent-based
model. In Simulation and Modeling Methodologies, Technologies and Applications
(SIMULTECH), 2014 International Conference on, pages 733–738. IEEE, 2014.
[6] P Maria Joseph Christie and Reuven R Levary. The use of simulation in planning
the transportation of patients to hospitals following a disaster. Journal of medical
systems, 22(5):289–300, 1998.
[7] Mei Ling Chu, Xiaoshan Pan, and Kincho Law. Incorporating social behaviors in
egress simulation. In Proceedings of 2011 Computing in Civil Engineering Workshop,
pages 19–22, 2011.
33
[8] Ilya Grigoryev. Anylogic 7 in three days. A quick course in simulation modeling,,
2015.
[10] David Handford and Alex Rogers. Modelling driver interdependent behaviour in
agent-based traffic simulations for disaster management. In Advances on Practical
Applications of Agents and Multiagent Systems, pages 163–172. Springer, 2011.
[12] Simo Heliövaara, Timo Korhonen, Simo Hostikka, and Harri Ehtamo. Counterflow
model for agent-based simulation of crowd dynamics. Building and Environment,
48:89–100, 2012.
[13] Anson Yuanxi Liang, Malcolm Yoke Hean Low, Michael Harold Lees, Wentong Cai,
and Suiping Zhou. A framework of intelligent environment with smart-active objects
(iesao) for flexible and efficient crowd simulation. In Proceedings of the 2010 Spring
Simulation Multiconference, page 19. Society for Computer Simulation International,
2010.
[14] Thomas W Lucas, Susan M Sanchez, Lisa R Sickinger, Felix Martinez, and
Jonathan W Roginski. Defense and homeland security applications of multi-agent
simulations. In 2007 Winter Simulation Conference, pages 138–149. IEEE, 2007.
[15] Masaru Okaya and Tomoichi Takahashi. Human relationship modeling in agent-based
crowd evacuation simulation. In International Conference on Principles and Practice
of Multi-Agent Systems, pages 496–507. Springer, 2011.
[16] M Pidd, FN De Silva, and RW Eglese. A simulation model for emergency evacuation.
European Journal of Operational Research, 90(3):413–419, 1996.
[17] Gabriel Santos and Benigno E Aguirre. A critical review of emergency evacuation
simulation models. 2004.
34
[18] Neal Wagner and Vikas Agrawal. An agent-based simulation system for concert venue
crowd evacuation modeling in the presence of a fire disaster. Expert Systems with
Applications, 41(6):2807–2815, 2014.
[19] Kashif Zia and Alois Ferscha. A simulation study of exit choice based on effective
throughput of an exit area in a multi-exit evacuation situation. In Proceedings of the
2009 13th IEEE/ACM International Symposium on Distributed Simulation and Real
Time Applications, pages 235–238. IEEE Computer Society, 2009.