Proceedings: Building Simulation 2007

A NEW COMPOSITE FIRE EVACUATION (CFE) MODEL

BASED ON HUMAN BEHAVIOR
Chen Tao1, 2, Yang Rui1, 2, Sun Zhanhui1, 2, Weng Wenguo1, 2
Email: chentao.a@tsinghua.edu.cn
1
2

Center for Public Safety Research, Tsinghua University, Beijing, 100084, China
Department of Engineering Physics, Tsinghua University, Beijing, 100084, China

ABSTRACT
Occupant evacuation dynamics is a kind of manybody system of strongly interacting persons. A great
deal of factors should be taken into account in the
study of fire evacuation, such as human behavior, fire
products and architecture. To describe human
behavior, there are two different evacuation models,
the social force model and the lattice-gas model.
Considering the impact of fire product upon people's
physiology and psychology, the CFE (Composite
Fire Evacuation) model is presented in this paper,
which includes three sub models: lattice-gas model,
social force model and pre-movement time model.
And a direction potential field concept is introduced
in the CFE model to solve the moving directionality
of people. In the end, the fire case of Yiyuan dancing
hall in Fuxin is simulated by the CFE model and the
lattice-gas model based on the human characteristics
of China.

KEYWORDS
Evacuation Model, Fire, Human Behavior, Case
Study

The key to guarantee the human safety in building is

that the RSET must be smaller than the Available
Safe Egress Time (ASET, viz. the time period from
fire breaking out to endangering human safety). The
universal criterion of safe evacuation time-line is
shown in Figure 1.

Available Safe Egress Time (ASET)

Required Safe Egress Time (RSET)
alarm time

detect

pre-movement time

alarm

moving time

finish evacuating building limitation

Figure 1 Universal criterion of safe evacuation time-line

The RSET means the time period from the moment
of fire breaking out to the moment that all the people
reach the safe area. The RSET includes the detecting
and alarm time talarm, the pre-movement time tpre and
the movement time tmove, where the tpre consist of the
recognition time treg and the response time tresp,
expressed as Eq. (1).
RSET = talarm + t pre + tmove = talarm + (treg + tresp ) + tmove

INTRODUCTION
In recent years, the study of performance-based fire
protection design and fire risk assessment develops
rapidly in the world and it has been one of key issues
in the research field of fire safety. One of the most
important goals of these researches is to ensure the
safety of life. To realize the rational and optimal fire
protection design, proper calculation of the Required
Safe Egress Time (RSET) is required.
It is thought that evacuation and fire development go
irreversibly and synchronously along the same timeline. The fire development can be divided into
several periods, such as breaking out, fire increasing,
stable burning, fire decay and fire extinguishing. For
human safety, the first two periods should be cared
about mainly. Generally, an evacuation will go
through the stages of perceiving a fire, preparing,
evacuating, reaching safe place, etc. During the
evacuation, there are two significant moments, the
fire detecting time and the time of fire endangering
human safety.

(1)

The talarm can be estimated by the smoke spreading

simulation and the characteristic of detection system,
the tpre is generally acquired from statistics, and the
tmove can be predicted by the simple empirical formula
or the evacuation model which mainly depends on
crowd density, walking speed, width of exits, etc.

CFE MODEL
According to the representation methods of the
enclosure (Gwynne et al. 1999), models can roughly
be classified as coarse network models, fine net
models or continuity models. In the coarse network
model, each node may represent a room or corridor
irrespective of its physical size. Nodes are connected
by arcs representing actual connectivity (such as door,
exit) within the structure. This model presents
difficulties when incorporating local movement and
navigation including overtaking, the resolution of
local conflicts, and obstacle avoidance. Using the
fine net approach the entire floor space of the
enclosure is usually covered in a collection of tiles or

- 703 -

Proceedings: Building Simulation 2007

nodes. It can show geometry form of the floor space,

position of the inside barrier and peoples accurate
position, but it is difficult to simulate the crowded
situation. Based on the systematic frame of multiparticles and self-driven, the continuity model (also
called social force model) uses mechanics equations
to simulate crowd dynamics. The individual has the
ability of independent movement and making a
response to surrounding environment in this model.
This simulation can subtly realize pushing, panic and
visual range (Helbing et al. 2000) at great cost of
computing time, which will result in low
computational efficiency.
Based on the human behavior characteristic of people
in China, CFE (Composite Fire Evacuation) model
has been described in this paper, which includes three
sub models: lattice-gas model, social force model and
pre-movement time model. The impact of fire
products upon people's physiology and psychology
has also been considered in this model. The CFE
model comprehends the advantage of the lattice-gas
model which is high computational efficiency, and
the social force model, which is fine simulation of
crowd dynamics. It adopts the lattice-gas approach in
low crowd density areas and uses the social force
model to simulate the panic and crowded behavior in
high crowd density areas.
Lattice-gas sub-model
Traditional lattice-gas model is a kind of discrete
non-linear kinetics system based on the fluid
molecular kinetic theory (Li et al. 1994), and it is the
embodiment and application of cellular automaton in
statistical physics and fluid mechanics. In a
microscopic scale, the model describes a large
number of particles with the same quality moving
and interacting in the regular discrete nets following
a certain collision rule, and it will simulate the
complicated
physical
phenomenon
in
the
macroscopic scale. In a word, it is a physical model
with discrete time, space and target.
Each human individual have a certain body size
which can be expressed by the area of peoples
horizontal projection. In this model, the size of lattice
is fixed through the area of peoples horizontal
projection and the statistics value of shoulder width.
It is indicated that the average area of pedestrian
horizontal projection is 0.113m2 in Russia
(Predtechenskii et al. 1971), 0.1458-0.1862m2 in
Austria (Kendik 1985) and 0.12-0.19m2 in Germany
(Schneider 2001). According to "Chinese adult's
physique determines standard", the statistical
averages of adult's height and weight are shown in
Table 1 (Yu 1997). About the proportionate
relationship of the human body, it is indicated that
the shoulder width is 1/4 of the height. So, the
average shoulder width of Chinese can be calculated,
shown in Table 1. According to the above statement,

each grid of the lattice is set as 0.5m0.5m. The grid

area is 0.25m2 which is sufficient to contain one
person.
Table 1 Mean values of Chinese adult's height,
weight and shoulder width
Age bracket
M
Height
(cm)
F
M
Weight
(kg)
F
Shoulder M
width (cm) F
Age bracket
M
Height
(cm)
F
M
Weight
(kg)
F
Shoulder M
width (cm) F

18-20
169.9
158.8
59.2
51.1
42.5
39.7
41-45
168.9
158.2
66.2
58.1
42.2
39.6

21-25
170.2
159.2
61.2
51.3
42.6
39.8
46-50
167.8
157.7
66.0
58.8
42.0
39.4

26-30
170.1
159.4
63.2
53.5
42.5
39.9
51-55
167.8
157.4
66.2
58.5
42.0
39.4

31-35 36-40
169.8 169.6
159.0 158.9
64.6 65.7
54.9 56.5
42.5 42.4
39.8 39.7
56-60
167.6
66.3
41.9

Except for certain quality and body size, a human

individual is a special particle with ideology.
Especially in fire emergency, peoples actions have
obvious directionality, viz. fleeing from the
hazardous area and evacuating to the safe zone or
outdoors. The traditional collision rule based on
momentum conservation can not describe peoples
action characteristics accurately. To solve the moving
directionality of people, the lattice-gas model in this
paper introduces a direction potential field concept. It
means that each lattice has a direction potential. The
individual in the CFE model will move from the
lattice with higher potential to that with lower
potential.
The direction potential field can not only express the
relative distance to the exit but also reflect familiar
degree of evacuee to the structure of the building, the
escape route and the safety exit. It can even show the
trend of far away from dangerous area. Figure 2
gives a demonstration of net dividing and direction
potential field initializing for the 5m5m square
room with two exits. Figure 2(a) is the sketch map of
net dividing, position of exits and fire point. In
Figure 2(b), only relative distance to the exit is
considered in the direction potential field. The
direction tendencies of exit 1 and 2 are set as 0, and
the highest direction potential is 9 in this room.
Figure 2(c) shows different familiar degree of
evacuee to different exits, which results in the change
of direction potential field. Here, the direction
potential of exit 1 is 0 and that of exit 2 is 5, which
means exit 1 is in common use. And the highest
value of direction potential has changed into 11. The
influence of fire region is expressed by Figure 2(d).
The direction potential of fire region in the CFE
model is usually set as the highest value, which

- 704 -

Proceedings: Building Simulation 2007

represents the tendency of keeping away from

hazardous area.
Exit1

10

10

Fire
region

8
7

10.00

f =

9.000

8.000

Y Axis Title

6
5
4

7.000

4.000
3.000

4
3

Exit2
2

10

1.000
0

1
1

(a)

10

12.00

10

(b)

10

12.00

10.00
9.000

10.00

7.000

8.000

5.000

6.000

5.000

3.000
2.000

3.000
2.000

1.000

1.000

1
2

10

(c)

(d)

10

Figure 2 Net dividing and direction potential field

initializing in the 5m5m square room

Table 2 Values of dimensionless parameters in premovement time sub-model

Social force sub-model

During fire emergency, people's behavior is generally
complicated and unordered. But given relatively
simple and loose environment, peoples behavior can
be described by probability, which lead to the
development of gas-kinetic model (Helbing 1992).
Lewin has proposed another method to simulate
people's behavior (Lewin 1992). It is thought that the
behavior should be changed by the influence of
"social force". According to the concept of social
force, Helbing, et al. have proposed a dynamics
model to describe peoples psychology and crowded
behavior (Helbing et al. 2000). In this model, the
behaviour in a crowd will be influenced by two
different forces, socio-psychological and physical
force.
It is supposed that individual i, who is in dangerous
area at a certain moment t, expects to evacuate to the
v
safe area with speed vi0 (t ) and direction ei0 (t ) . His
social psychology force can be described by Eq. (2),
v
where vi (t ) means the actual speed, mi is the weight
and i is the characteristic acceleration time to
describe his accelerating ability.
v
v
v
v 0 (t )ei0 (t ) vi (t )
fi 0 (t ) = mi i

(2)

v
The physics environment force
includes f ij (the
v
force among people) and fiw (the force between
people and barrier). The kinetic equation to describe
evacuating velocity in the social force sub-model can
be expressed as:
mi

(5),

The five factors can be expressed by dimensionless

parameters a, b, c, d and e separately, whose values
are shown in Table 2.

1
1

(4).

variance of tpre,.

4.000

4.000

where t pre and tpre, are mean value and mean

7.000

6.000

2t pre , 2

9.000

8.000

(t pre t pre , ) 2

t pre , = t pre b(a + c + d + e)

11.00

11.00

2 t pre ,

exp[

The five main factors which influence tpre, (standard

pre-movement time) are: peoples state, building
characteristic, present position of people, fire
intensity, fire detecting and warning facility. The
calculation formula of tpre, is

5.000

6.000

2.000

Pre-movement time sub-model

In the pre-movement time sub model, a Gaussian
distribution function is used to describe the premovement time tpre (Jia et al. 2002), expressed by

v
v
v
v
v
dvi
v 0 (t )ei0 (t ) vi (t )
= mi i
+ f ij + f iw
i
dt
j i
w

(3).

Dimensionless
Parameter
People's
a
state

b
c
d
e

Characteristic and Value

Waking
Resting
Sleeping
1
1.2
1.5
Market or Office or House or
Hotel or Hospital or
Building amusement Factory
School
Apartment Sanatorium
place
building dormitory
characteristic
0.5
1.0 1.2~1.5 1.6~1.8 2.0
Position of
The number of rooms to the fire room / 10
people
large
middle
small
Fire intensity
-0.1
0
0.1
Pronunciatio
Alarm bell
Recording
Fire detecting
None
n
alarm
and warning
Accurate Inaccurate
broadcasting
facility
-0.2
-0.1
-0.1
0.2
0.1

CASE STUDY
Since 1991 in China, the twelve extraordinarily
serious fire tragedies have caused about 1,552 deaths,
each of which devitalized more than 50 people's lives.
Among them, the fire of Yiyuan dancing hall in
Fuxin resulted in 233 deaths. Taking this fire as an
example, FDS is used to simulate the fire scene and
to calculate and confirm ASET, and RSET is
calculated by the CFE model. The number of death
can be calculated by comparing of ASET and RSET.
Fire scene
The fire accident of Yiyuan dancing hall in Fuxin
took place at 13:30 on November 27, 1994. There
were 304 people in the dance hall when the fire broke
out. It caused 233 deaths, 133 men and 100 women.
Four people were severely injured and sixteen people
were slightly injured. Among the 231 deceased

- 705 -

Proceedings: Building Simulation 2007

whose identities were verified, there were 84 workers,

75 students and 61 job-waiting youth. According to
the statistic, most of the deceased were young people,
61 people between 14 and 17 years old and 159
people between 18 and 25 years old (Anno., Anno.
A).
Box

Box

Box

Front
corridor
North gate
Dance hall

2m

Sofa

Street of north

Box

0.8m

Box

1.8m

13m

Courtyard outside the south

Box

fire in this paper is relatively conservative. Figure 4

shows the calculation result of FDS at 6 minutes after
the start of the fire: the value of temperature and CO
density distribution map at the horizontal height of
1.6m, which is the height from the floor to the mouth
of person of average length.

3m
South gate
North

18m

Figure 3 The plan sketch of Yiyuan dancing hall in Fuxin

(a) Temperature ()
The area of Yiyuan dancing hall is 280m2. Figure 3
shows the Yiyuan dancing hall. The hall is 18 meters
long, 13 meters wide and 5 meters high, with six
boxes in the west which are 2.76 meters long, 2
meters wide and 2.4 meters high. The width of the
boxes doors is 0.8m, and the north gate is 0.8m, the
south gate is 1.8m (unfortunately, the south gate was
closed during the fire). All the exits are 2 meters high.
The building wall is fitted with the woven Dacron
and cotton with both upper and lower ends nailing
and pigeonholing. The top of the building is
decorated by wooden-ply board. There are 56 sofas
against the wall with synthetic leather as surface
material and polyurethane as filling.
(b) CO density (ppm)

ASET calculated by FDS

To simulate the ventilation effect of the window
break in FDS model, 6 vents with 0.08m2 area are set
up evenly in both sides of the hall, which are 2.0m
height to the floor. According to the experiment
database of NIST (National Institute of Standard &
Technology), U.S.A., the largest fire power of one
armchair is 2.25MW (Anno. B). It is supposed that
the fire taking place in the dance hall is t2fire with
0.046889kW/s2 of fire growth rate and there are
always two sofas burning steadily in the largest fire
power at any moment of spreading course after the
fire is steady. Then, in the simulation of FDS, the
area of fire is set as 2m2 to satisfy the horizontal area
of two armchairs. The fire material was adopted
polyurethane, and the power of fire after burning
steadily is 2.25MW/m2. It can be found from the
hypothesis that the fire source in FDS doesnt include
the woven Dacron and cotton hanging on the wall.
Since the burning rate of these chemical upholstery
fabrics is very fast, it is considered that the supposed

Figure 4 Temperature and CO density distribution map at

1.6m height (240s)
It is supposed in this paper that the temperature
higher than 80 or the concentration of CO more
than 800ppm will cause injury to the human body.
The simulation results of FDS show that the average
temperature at the horizontal height of 1.6m will be
higher than 80 in about 220s, and CO density will
reach 800ppm in about 280s. Thus, ASET is
confirmed as 220s by the simulation of FDS.
Evacuation simulation and analysis
It is described by Helbing et al. (2000) that the
influence radius among human individual is 5m in
the social force model. Consider that the high crowd
density area generally appears near the bottleneck
such as the exit and stair (Chen et al. 2003), in this
case study, the social force sub-model is used to
describe the panic and crowded behavior within 5m

- 706 -

Proceedings: Building Simulation 2007

range near the north door and the lattice-gas submodel will simulate evacuation in other low density
areas.
In the simulating of the CFE model, peoples weight,
shoulder width and speed are all regarded as normal
distribution. According to the statistical information
about the people in the dancing hall and the data of
Table 1, the choice of peoples parameters in the CFE
model is selected as described in Table 3.
Table 3 People's parameters in the CFE model
SEX
Number
Mean
Variance
Mean
Variance
Normal
Expect
Variance

Weight (kg)
Shoulder Width
(cm)
Speed (m/s)
Pre-movement
time(s)

174
60
1
47.5
3
1.0
5.0
0.2

130
51
1
44.7
3
1.0
4.0
0.2

t pre

240

tpre,

10

warning facilities, and the fire is assumed to be large.

According to Table 2, the value of the dimensionless
parameter a, b, d and e is 1, 0.5, -0.1, 0.1 respectively.
The value of c is calculated based on the topological
structure of building. Thus, the pre-movement time
of people in this building is calculated by the premovement time sub-model. Figure 5(a) presents premovement time of different persons, and Figure 5(b)
is the distribution of pre-movement time, which is a
normal distribution. From Figure 5, the biggest and
smallest pre-movement time are 170s and 85s.
Based on the parameter selection, the simulatiom
result of the CFE model is presented by the solid line
in Figure 6. For comparison, the analog result is also
obtained by the simple lattice-gas model, which is
expressed by dotted line in Figure 6. It is shown that
the evacuation efficiency calculated by the CFE
model is relatively low. It is probably contributes to
this models ability of describing the phenomena of
pushing and shoving among individuals near the
north gate, which reflects that people are eager to
evacuate to the safe area during a fire.
350

CFE model
Lattice-Gas model

ASET
300

Number of people not evacuated

170
160

Pre-movement time (s)

150
140
130
120
110

250
200
150
100
50

100

0
90

50

100

150

200

250

300

100

150

200

250

300

350

400

450

500

550

350

Figure 6 The simulation result of evacuating from north

gate during fire

No. of personnel

(a) Pre-movement time of different person

Regarding 220s as dangerous moment which is the

ASET calculated by the simulation of FDS, 162
persons will die from the simulation of simple latticegas model and the number of death will be 197 from
the CFE model. It seems that the simulation result of
the CFE model is closer to the real case situation.
One of the most probable reason is the difference in
simulation ability about crowd situations of these two
kinds of models. The ability of the simple lattice-gas
model to describe the interacting among individual is
not strong enough to simulate the evacuating
situation in the high crowd density area accurately,
especially in case where the exit is small and
congestion is likely to occur.

80
70
60

Number of people

50

RSET(s)

80

50
40
30
20
10
0
70

80

90

100

110

120

130

140

150

160

170

180

Pre-movement time(s)

(b) Distribution
Figure 5 Pre-movement time calculated by pre-movement
time sub-model
The dancing hall is a public place of entertainment
for people who are awake without fire detecting and

The number of 197 deaths in accident of Yiyuan

dancing hall simulated by the CFE model is smaller
than the real death number of 233. The authors
consider that it is mainly due to the neglect of the
chemical fiber cloth decorating to the wall in the

- 707 -

Proceedings: Building Simulation 2007

assumption of the fire scene. Thus, the fire situation

calculated in FDS simulation is not as dangerous as
the actual conditions. The neglect leads to the
deviation of ASET and influences the simulation
result of peoples death finally.

CONCLUSION
The CFE model is built based on the analysis of
Chinese peoples behavior characteristic, which
includes three sub models: lattice-gas model, social
force model and pre-movement time model. The
impact of fire products upon peoples physiology and
psychology has been considered in this model.
Evacuation during fire can be simulated by the CFE
model, and the result can predict RSET. With the
help of FDS model to calculate ASET, it is possible
to perform risk assessment or analyze the fire case
for the certain building or environment.
After the simulation and analysis of the CFE model
and simple lattice-gas model, it seems that the
simulation result of the CFE model is closer to the
real case situation. The reason is that the CFE model
can simulate the condition of pushing and shoving
among individual accurately. This condition reflects
the psychology that people in a fire are eager to
evacuate to the safe area as quickly as possible, and it
usually causes congestion near the exit and reduces
the efficiency of evacuating, which is described as
the fast-is-slow effect by Helbing (2000).
In the future, video material on escape panics will be
used to validate the CFE model. The authors are now
studying the impact of CO and high-temperature of
fire exhaust gases on people, the interpersonal
interactions which include direction, velocity,
acoustic information exchange, etc.

ACKNOWLEDGEMENT

methodologies used in evacuation modeling,

Fire & Materials. 23: 383-388.
Helbing D. 1992. A fluid-dynamic model for the
movement of pedestrians, Complex Systems 6:
391-415.
Helbing D., Farkas I., Vicsek T. 2000. Simulating
dynamical features of escape panic, Nature. 407:
487-490.
Jia C.Q., Liu C., Xiang T.H., Zeng D.L. 2002. The
Study of Evacuation Start Time in Fires, Fire
Safety Science (In Chinese) 11(3): 176-179.
Kendik E. 1985. SFPE Technology Report 85.4:
Assessment of Escape Routes in Buildings and
a Design Method for Calculating Pedestrian
Movement, Society of Fire Protection
Engineers. Boston, Massachusetts.
Lewin K., 1992. Field Theory in Social Science,
Harper, New York.
Li Y.X., Kang L.S., Chen Y.P. 1994. Lattice-Gas
Automaton, Tsinghua University Press,
Guangxi Science and Technology Press (In
Chinese).
Predtechenskii V.M., Milinskii A.I. 1971. Planning
for Foot Traffic Flow in Buildings, Amerind
Publishing Co. Pvt. Ltd., New Delhi.
Schneider V., 2001. Application of the individualbased evacuation model ASERI in designing
safety concepts, in: Proceedings of 2nd
International Symposium on Human Behavior in
Fire, March, Boston, USA, , 41-51.
Yu D.Z., 1997. Report on the investigation and
research of the physique of Chinese staff
members and workers, Fujian Sports Science
and Technology (In Chinese) 16(4): 6-13.

The authors appreciate the project 70503017

supported
by
NSFC
and
the
project
212006BAK01A02 support by MOST.

REFERENCES
Anonymity http://info.fire.sinobnet.com/
Anonymity A http://www.119.sd.cn/infocenter/
Anonymity B http://202.38.89.99/software_index/fastdata_index.htm
Chen T., Song W.G., Fan W.C., Lu S.X., Yao B.
2003. Pedestrian evacuation flow from hallway
to stairs, in: Faridah Shafii, Richard Bukowski,
Ron Klemencic(eds.), The CIB-CTBUH
Conference on Tall Buildings: Strategies for
Performance in the Aftermath of the World
Trade Centre, CIB TG50, Malaysia, 79-86.
Gwynne S., Galea E.R., Owen M., Lawrence P.J.,
Filippidis L. 1999. A review of the

- 708 -