Evacuation Simulation under Different Conditions using a Safest

Path Routing Algorithm

Denis Shikhalev

, Renat Khabibulin

, Ulrich Kemloh

and Sergey Gudin

The State Fire Academy of EMERCOM of Russia, Moscow, Russia

Jülich Supercomputing Center, Forschungszentrum Jülich GmbH, Jülich, Germany

Ghent University, Ghent, Belgium

Keywords: Pedestrians Dynamics, Evacuation Strategies, Safest Evacuation Route.

Abstract: In this contribution we propose a safest path route algorithm for determination of the safest path directions

of pedestrians in case of fire. The model and the algorithms are implemented in an open source framework

(JuPedSim) which is a research platform to simulate pedestrian dynamics. We found that increasing the

importance of the obstruction criteria (responsible for people’s density) leads to a reduction of the total

evacuation time. The proposed algorithm allows the even distribution of the evacuees to all available

emergency exits, when there is an uneven distribution of people on the escape routes while avoiding a place

with fire hazards. We simulate the evacuation of a shopping centre and showed that the application of the

algorithm can reduce the total evacuation time up to 63% depending on the settings of the algorithm.

1 INTRODUCTION

In the last few decades, large fires in shopping malls

were the reason of many people's death. A few of

them are listed below:

 December 25, 2000. A fire occurred in a

central shopping Centre (Luoyang, China). The

fire killed 309 people;

 August 01, 2004. A fire occurred in a

supermarket (Asunción, Paraguay). The fire

killed 464 people;

 May 28, 2012. A fire occurred in a Villagio

Mall (Doha, Qatar). The fire killed 19 people,

including 13 children.

One of distinguishing features of shopping malls

is the uneven distribution of people in the building.

It can influence an organization’s evacuation process

and leads to an unbalanced use of emergency and

exits routes. A significant number of people are

accumulated in supermarkets and shops of home

appliances compared to other shops of a shopping

mall.

An analysis of some existing escapes route

systems from different countries (Shikhalev and

Khabibulin, 2013) showed that only a third of the

systems were able to determine the direction of the

escape routes using a scientifically founded method.

The studies of many authors (Carattin, 2011;

Kobes, Helsloot et al., 2010; Samochine, 2004;

Sandberg, 1997) indicate the following problems in

the area of evacuation management in shopping

malls:

 Uneven distribution of people inside shopping

malls ;

 Organisation problems in the evacuation

process, done by the staff of shopping malls;

 Lack of information about possible (available)

evacuation directions.

Therefore, the lack of both models and

algorithms of information and analytical support for

evacuation managements leads to the fact that a

decision maker cannot objectively evaluate the

whole range of hazards and determine safe routes for

people during an emergency evacuation.

To solve these problems, a mathematical model

of a safest path route algorithm was developed. The

algorithms are used to calculate the safest path for

people in a danger zone, and to direct them to a safer

area (Shikhalev, Khabibulin et al, 2014).

In a first estimation the model showed a positive

impact on the evacuation time and overall on the

people’s safety during evacuation simulations

(Shikhalev, Khabibulin et al, 2014). Nevertheless, it

is needed to complete a full estimation of all features

of the model as well as determine the best

combination of properties for evacuation

simulations.

Shikhalev, D., Khabibulin, R., Kemloh, U. and Gudin, S.

Evacuation Simulation under Different Conditions using a Safest Path Routing Algorithm.

In Proceedings of the 18th International Conference on Enterprise Information Systems (ICEIS 2016) - Volume 2, pages 62-69

ISBN: 978-989-758-187-8

In this paper we consider the results of

evacuation simulations using the algorithm.

Simulations were performed using the Jülich

Pedestrian Simulator, JuPedSim (Kemloh, Chraibi et

al, 2015) with various numbers of people and

different objects.

This work is structured as follow:

In the second section we introduce the model and the

algorithm. In the third section, we present computer

simulations and analyses. Some concluding remarks

are given in the last section.

2 DESCRIPTION OF THE

SAFEST PATH ROUTE

ALGORITHM

The safest path route algorithm is applied for

calculation of the safest path for people from

different points of a building to the exterior of the

building. Originally the safest path route algorithm

was created for a shopping mall. The main tasks of

the algorithm are to calculate a safest route and

direct people by a path (current or newly defined).

The algorithm of Floyd-Warschall (Floyd, 1962)

was applied for calculating the safest path to the

nearest gate. Normally, Floyd-Warschall’s algorithm

finds the shortest path between all pairs of edges in a

graph. A physical distance is used as the weight of

the edges. For our task, we used a complex criterion

φ as the weight of the edges. φ is calculated using

Equation 1:

222

)()()(

iii

lba ⋅+⋅+⋅=

γβαϕ

(1)

at:

a → min, i=1,…,n

b → min, i=1,…,n

l → min, i=1,…,n

where:

a – an obstruction criterion;

b – a timeliness criterion;

l – a length criterion.

α, β, γ – the weight coefficient at a, b, l .

The obstruction criterion is determined by the

ratio of the people’s density on a section of the

escape route network, to the maximum people’s

density that does not cause adverse effects to

humans. The timeliness criterion is directly linked to

fire hazards (high temperature, a large amount of

smoke, low visibility, toxic products of combustion

etc.). The length criterion is the relative length of the

current section. It is calculated as the ratio of current

escape route length, to the maximal escape route

length in a building. The coefficients (α, β, γ) are

added to regulate the importance of the individual

criteria. More details about the criterion and

manners of its computing are found in previous

work (Shikhalev, Khabibulin et al, 2014).

Under sections of escape route, we consider the

crossing of two (or more) escape routes in the

corridors of a shopping mall. Hence a section of an

escape route corresponds to an edge in the graph of a

shopping mall, and a place of cross of two (or more)

escape routes corresponds to a vertex.

We used the JuPedSim simulator for computer

implementation of the algorithm. The Generalized

Centrifugal Force Model (GCFM) is applied into the

simulator to simulate an evacuation process

(Chraibi, Seyfried et al, 2010). GCFM belongs to the

class of forces based models (Helbing and Molnar,

1995) and describes the movement of people at the

operational level (Hoogendorn, Bovy et al, 2002) i.e.

defines basic rules for the pedestrians such as

acceleration, braking and stop. Motion of the

pedestrians is determined by a so-called "social

power" (Helbing and Molnar, 1995). Calibration of

the basic parameters of GCFM (attractive and

repulsive forces, the size of the semi-axes of the

ellipse depending on the density and velocity of the

people flow etc.) were performed in (Burghardt,

2009; Meunders, 2011). Verification and validation

of the GCFM, as well as a more detailed description

is given in (Chraibi, Seyfried et al, 2010; Chraibi,

2012; Kemloh, 2012).

At each step of the simulation, the evacuees are

sent through the shortest path, to the nearest

emergency exit in the building i.e. a shortest path

route algorithm (ShPA) is used to determine the

shortest escape route (Fig. 1).

Figure 1: The shortest path route algorithm (ShPA).

However, there is a need to change the ShPA

with regards to the problem of determination of the

Evacuation Simulation under Different Conditions using a Safest Path Routing Algorithm

Figure 2: The safest path route algorithm (SaPA).

safe evacuation routes. For this purpose, a safest

path route algorithm (SaPA) was developed (Fig. 2)

An update frequency (UF) was added into the

SaPA for the possibility of regulating how often the

algorithm will be refreshed. Thus, a UF value of 5

means that the safest path will be calculated once for

every 5 seconds and the direction of movement will

also be updated (in decision areas) after every 5

seconds.

The shortest path is given for all evacuees in an

initial stage (pre-evacuation). Then the safest path is

computed for each node (decision area) according to

the UF. In other words, the decision area is a place

where two or more routes crossed. It is possible

from this place (decision area), to direct the

pedestrians by a new path, for example, by applying

dynamic indicator (Shikhalev and Khabibulin,

2013). If a current path is not a safest path anymore,

a re-routing will happen in the decision area.

The safest path route algorithm was implemented

into JuPedSim as a separate module and it can be

chosen from other routing algorithms such as a

quickest or shortest path algorithms.

The main purpose of simulation is to evaluate the

effectiveness of the safest path route algorithm. This

evaluation is done by comparing the performance of

the shortest path route algorithm and the safest path

route algorithm.

Thus, the following research problems should be

answered during simulation:

 How are the a-criteria and b-criteria changed in

the process of evacuation and under what

quantitative values of criteria does the process

of re-routing happens?

 Which are the effects of re-routing pedestrians?

 How do weight coefficients affect the course of

the evacuation process?

 How does the update frequency affect the

course of the evacuation process?

 When it is advisable to apply the safest path

route algorithm?

To answer these questions, it is necessary to

conduct a preliminary assessment of the adequacy of

the developed algorithm. From there, we perform a

computer simulation of the evacuation process on

the topology of an existing shopping mall, as an

example.

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

3 SIMULATION AND ANALYSIS

In this section we provide simulation results and its

analysis.

3.1 Preliminary Assessment

Several series of simulations at the T-junction of

escape routes (Fig. 3) were carried out within a

preliminary assessment of the SaPA as well as on the

abstract model of the building (Fig. 4).

Figure 3: Objects of simulation within a preliminary

assessment – T-junction.

Simulation results at the T-junction identified

one of the features of the SaPA: the safest path route

algorithm behaves itself as the shortest path route

algorithm when there are no congestions or high-

density of pedestrians. It was also found that the

usage of weight coefficients has an impact on the

time of re-routing during evacuation. Weight

coefficients (0,9-0-0,1) lead to the re-routing that

occurs both with few and many pedestrians. In its

turn, weight coefficients of 0,6-0-0,4 allow to re-

route flows when we have a high number of

pedestrians. Moreover, re-routing occurs only at the

maximum configured number of people in

simulation. (250 per., 6.25 pers./m

in a case when

the weights are not applied). Re-routing moments

and duration of re-routing become longer when the

importance of a-criterion is increased.

The results obtained in the T-junction

simulations led us to several conclusions:

 The safest path route algorithm behaves as the

shortest path route algorithm in the case where

there are no congestions or high-density of

pedestrians;

 The application of weight coefficients

influences the course of the evacuation process

where the escape routes sections are of

different geometrical size.

 In order to achieve the minimal evacuation

time and prevent pedestrians’ congestions, it is

necessary to increase the importance of a-

criteria. Reducing the importance of a-criteria

leads to an increase in evacuation time;

 It is possible to regulate (zoom or zoom out) a

moment of re-routing of evacuation flows by

applying different weight coefficients.

Figure 4: Objects of simulation within a preliminary

assessment – Abstract model of the building.

After simulations at T-junction we continue in

the abstract model (fig. 4). Simulation results in the

abstract model allow us to state the following facts.

Firstly, for uneven distribution of people during

evacuation, the SaPA can immediately distribute

pedestrians evenly to the emergency exits. This in

turn significantly affects the evacuations time.

Efficiency of the SaPA (fig. 5) reduces when there is

an uneven distribution of people. Under efficiency

we understand the ratio of evacuation time with the

SaPA to evacuation time with the ShPA.

Figure 5: Efficiency Distribution of the SaPA depending

on terms of people distributions. (0-150-150-0 value

corresponds to the people distribution in rooms 1,2,3 and

4, respectively (Fig. 4)). White – scenario 1, black -

scenario 2.

Secondly, there is a negative efficiency of the

SaPA at 4%. This happened in the case where we

Evacuation Simulation under Different Conditions using a Safest Path Routing Algorithm

had an uneven distribution of a small number of

people (up to 50 people). Having reviewed the

evacuation process in decision area, a reason of the

negative efficiency of the SaPA was found. This is

due to the update frequency of the SaPA which was

equal to 1. There were many re-routing of

pedestrians while they followed the decision areas

(geometric size of decision area is 2 meters by 2

meters). Pedestrians were sometimes directed to

different exits. This in turn, had to slow down the

speed of pedestrians and as a result, increase the

evacuation time. Thus, an optimal value of the

update frequency should be investigated and

determined. The simulation results in the abstract

model led us to conclude these facts:

 Weight coefficients do not play any role when

there are two identical routes (by both

geometric characteristics and number) from

decision area to the exit;

 A SaPA update frequency of 1 has a negative

impact on evacuation process given a small

number of pedestrians;

 The SaPA allows evenly "download" sections

of evacuation routes when there is an uneven

distribution of pedestrians. Other words the

SaPA directs pedestrians by routes which are

not using during evacuation but can be.

3.2 Simulation of a Shopping Mall

After preliminary assessment we performed

simulations in a shopping mall. The plan is shown in

figure 5. The color represents the decision areas.

Some geometric characteristics of the evacuation

exits and evacuation route sections in the front of

evacuation exits are shown in Table 1.

Table 1: Geometric characteristics of the evacuation exits.

№

Parameter

Evacuation exit

1 2 3 4 5 6 7

1 Width, m. 3,0 3,0 1,5 4,0 2,0 2,0 2,0

Width of evacuation

route section in the

front of evacuation

exit, m.

6,4 6,4 2,0 10,1 2,2 2,2 2,2

The number of people in evacuation simulation

was chosen in the rate of 1 person per 1 m

of retail

premises (total number of evacuees in 2609 person).

The influence of the update frequency on the

evacuation process was considered in the first series

of simulation. The simulation results are shown in

figure 7.

Analysis of figure 7 shows that for many people,

the closest emergency exit is №7, but based on its

geometrical characteristics, it is not preferable

because of its small width (See Table. 1). However,

most of the pedestrians were distributed between

exits 1, 2, 4 which are preferable due to their

geometrical dimensions (exit width). Thus, the

direction of all pedestrians to the shortest emergency

exits is not always justified and often leads to a

significant increase of total evacuation time.

Figure 6: Layout of shopping mall.

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

Figure: 7: People distribution to emergency exits. A -

ShPA; B - SaPA with update frequency equal 5 (result of

minimal evacuation time); C - SaPA with update

frequency equal 13 (result of maximal evacuation time).

Application of the SaPA allows to reduce evacuation

time up to 63% depending on the update frequency

of the algorithm.

Assessment of the update frequency of the

algorithm showed that the preferred frequency is 5.

That is why the frequency used for further studies

will be 5.

The simulation results in the T-junction suggest

that using different weight coefficients can reduce the

evacuation time. An analysis of the effect of weight

coefficients on pedestrian’s distribution to emergency

exits was conducted in the next stage of simulation.

The simulation results are shown in Figure 8.

Results confirmed previous findings about the

effect of the weight coefficients on evacuation process.

It should be noted that using weight coefficients of 0.7-

0-0.3 or 0,6-0-0,4 leads to the same results as not using

weight coefficients at all. Nevertheless, these

conditions (weights: 0,7-0-0,3; 0,6-0-0,4; without

weight coefficients) contribute to reducing evacuation

time in comparison with ShPA by 21%.

The main difference between the weights of 0,7-

0-0,3 or 0,6-0-0,4, however, as between all the

weight coefficients is the people’s distribution

according to emergency exits.

Figure 8: Weight coefficients vs. evacuation time.

Figure 9 presents the data with more details on

the distribution of people to emergency exits.

Figure: 9. The pedestrian distribution to emergency exits

depending on the weights. A – Weight coefficients are

0,9-0-0,1. B - weight coefficients are 0,8-0-0,2. C - weight

coefficients are 0,7-0-0,3. D - weight coefficients are 0,6-

0-0,4. E – without weight coefficients.

Figure 9 shows that the largest reduction of the

evacuation time was achieved when pedestrians

were directed to wider exits and in contrast the

maximum evacuation time was achieved by

"loading" narrow exits.

It was also interesting to consider the fairly

frequent assertion of researchers in the field of

human behaviour, that people in a fire will follow

the escape routes they used to get into the building

(Kobes, Helsloot et al, 2010; Samoshin, 2004;

Sandberg, 1997). It is likely that visitors enter a

building on the gate leading from the metro stations,

parking places, etc. Corresponding exits are 1, 2 and

4 in figure 6.

To carry out the simulation exits (3, 5, 6 and 7)

are blocked, because it is unlikely that they can be

used by most pedestrians entering the building.

Different cases were simulated and investigated

particularly when all of the exits (1, 2, and 4) are

opened and then when one of the exits is blocked.

The simulation results for different positions of the

emergency exits are shown in figure 10.

Figure 10: Dependence of the evacuation time to an

algorithm (SaPA vs. ShPA). White – SaPA. Black – ShPA.

Evacuation Simulation under Different Conditions using a Safest Path Routing Algorithm

The simulation results show that the direction of

all pedestrians only through the main evacuation

exits can significantly reduce the evacuation. For the

last part of simulation we elaborated four evacuation

strategies:

 Strategy 1. Applying the SaPA with weight

coefficients equal 0,9-0-0,1 provided that all

exits are opened;

 Strategy 2. Applying the SaPA with weight

coefficients equal 0,9-0-0,1 provided that only

the main exits are opened;

 Strategy 3. Applying the ShPA provided that all

exits are opened;

 Strategy 4. Applying the ShPA provided that

only the main exits are opened.

Figure 11 shows the simulation results with

mentioned strategies. Minimal evacuation time was

achieved when the strategy 1 was chosen. The SaPA

is still preferable than the ShPA only if main

emergency exits are available. However, for cases

where the only possibility is to direct people through

the shortest path, it is necessary to use strategy 4.

Figure 11: Ratio of evacuation time to an evacuation

strategy.

4 CONCLUSIONS AND FUTURE

WORK

In this paper we presented the results of full

assessment of the safest path route algorithm in the

framework of evacuation simulations. It was found

that the weights of 0,9-0-0,1 should be applied to

prevent congestions during evacuations, when

people are of high density. For the algorithm, an

update frequency of 5 should be chosen to timely

direct the pedestrians to safe evacuation paths. The

algorithm is suitable for cases when there are no

widely dispersed emergency exits, uneven

distribution of evacuation flows to the exits as well

as to prevent congestions of high density of people

in evacuation.

Obtained results allow as talk about effectiveness

of proposed algorithm. However an experimental

assessment is required for its application in a real

evacuation process.

The following phenomena should be investigated

within the frame of an experimental assessment:

 people's reaction to dynamic indicators;

 do pedestrians follow the routes which would

be offered;

 how staff responsible for evacuation

organization will operate with dynamic

indicators;

A plan of future research is to create a decision

support system for emergency evacuation in a

shopping mall based on obtained results.

ACKNOWLEDGEMENTS

The simulation results of this paper would never

have appeared without sincere assistance of member

of the «Civil security and traffic» division of Jülich

Supercomputing Centre at the Forschungszetrum

Jülich GmbH.

REFERENCES

Burghardt, S., 2009. Analyse und vergleichende

Untersuchung zum Fundamentaldiagramm a Treppen.

Masterthesis. Bergische Universität Wuppertal.

Carattin, E., 2011. Wayfinding architectural criteria for the

design of complex environments in emergency

scenarios. In Advanced research workshop

proceedings. Santander, Universitad de Cantabria.,

Spain.

Chraibi, M., Seyfrid, А. and Schadschneider, А., 2010.

Generalized Centrifugal Force Model for pedestrian

dynamics. Physical review E.

Chraibi, M., 2012. Validated force-based modeling of

pedestrian dynamics. PhD thesis. Forschungszentrum

Jülich, Jülich, Germany.

Floyd, R., 1962. Algorithm 97: Shortest Path.

Communications of the ACM 5.

Helbing, D., Molnar, P., 1995. Social force model for

pedestrian dynamics. Physical review E.

Hoogendorn, S. P., Bovy, P. and Daamen, W., 2002.

Microscopic pedestrian wayfinding and dynamics

modelling. Pedestrian and Evacuation Dynamics.

Kemloh, U., 2013. Route choice modeling and runtime

optimization for simulation of building evacuation.

PhD thesis. Forschungszentrum Jülich, Jülich,

Germany.

Kemloh Wagoum, A.U., Chraibi, M. & Zhang, J. 2015

JuPedSim: An open framework for simulating and

analyzing the dynamics of pedestrians. In 3rd

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

Conference of Transportation Research Group of

India.

Kobes, M., Helsloot, I., et al., 2010. Building safety and

human behavior in fire: A literature review. Fire

Safety Journal.

Meunders, A., 2011. Kalibrierung eines Mikroskopischen

Models für Personenströme zur Anwendung im

Project Hermes. Masterthesis. Bergische Universität

Wuppertal, Wuppertal, Germany.

Molnar, P., 1995. Modellierung und silmulation der

dynamik von Flussgagerstromen. Dissertation.

Universität Stuttgart, Stuttgart, Germany.

Samochine D.A., 2004. Towards an understanding of the

concept of occupancy in relation to staff behaviour in

fire emergency evacuation of retail store. PhD thesis.

University of Ulster, Belfast, UK.

Sandberg, А., 1997. Unannounced evacuation of large

retail-stores. An evaluation of human behavior and the

computer model Simulex.

Shikhalev, D., Khabibulin, R., 2013. Escape route systems

at shopping malls. Fire and explosion safety journal.

Shikhalev, D., Khabibulin R., 2013. Patent (in Russia) on

December 27, 2013 №136212 «Dynamic indicator».

Shikhalev, D., Khabibulin, R., Kemloh U., 2014.

Proceedings of conferences. Development of a Safest

Path Algorithm for Evacuation Simulation in Case of

fire. Proceedings of IV international conferences on

agents and artificial intelligence – ICAART 2014.

Evacuation Simulation under Different Conditions using a Safest Path Routing Algorithm