Applied Mathematics and Computation 298 (2017) 57–64

Contents lists available at ScienceDirect

Applied Mathematics and Computation

A weighted evolving network model for pedestrian evacuation

a r t i c l e i n f o a b s t r a c t

Keywords: We examine how the dynamics of dependency relationship between evacuees affect the
Pedestrian evacuation evacuation time and cooperative behavior. First, we place the population in a weighted
Dependency relationship
network, and then model the dynamics of dependency relationship between evacuees by
Local behavior
letting link weights to evolve. By careful examination of the parameter space, the simu-
Cooperation
Evacuation Time lation results indicated that a weaker link weight evolution weakens the local behavioral
factor in the evacuation process, which further enhance the efficiency of crowd evacuation.
On the contrary, a stronger link weight evolution promotes heterogeneity of the crowd,
thus strengthening the local behavioral factor in the evacuation process and prolonging
the evacuation time. In addition, we conduct a simple experiment: evacuation of a room
with two exits. By adjusting the length of two exits, we can guide the gathering behavior
and regulate the crowd evacuation. This experiment can also provide an explanation for
why the exits should located symmetrical in the middle of a wall.
© 2016 Elsevier Inc. All rights reserved.

1. Introduction

The process of pedestrian evacuation is a typical decision-making process. In this process, pedestrians update their posi-
tion and behavior with a desire for safety and efficiency. Pedestrian decision-making is usually very complex in real-world
situations. Factors that affect decision-making include environmental characteristics and local interactive behavior. In the
process of evacuation, the decision of evacuees covers not only the choice of position and behavior, but also the dynamics
of dependency relationship. Unfortunately, the dynamics of dependency relationship is often omitted. We thus examine how
the dynamics of dependency relationship between evacuees affect the process of evacuation.
Crowd activities in various public places are becoming increasingly common, e.g., sports meeting, big concert and aca-
demic exchange. Safe evacuation is a significant challenge for both participants and organizers. In case of an accident, people
expect to escape from the dangerous area as soon as possible. Therefore, understanding the dynamics of evacuation is very
important. In general, evacuation refers to egress from an area due to a potential or actual threat. The dynamics of the evac-
uation processes is quite complex due to a large number of interacting people affected by external factors as well. Recently,
the investigation of evacuation dynamics is being studied by scholars from different disciplines, and the evacuation process
was described and understood on different levels: physical, physiological, psychological, and social [1–4]. In fact, safe evacu-
ation depends on many factors including accident prediction, emergency response, information transfer and crowd control.
In recent years, the pedestrian evacuation problem has been investigated from different perspectives, including macroscopic
models and microscopic models. Macroscopic models fail to consider the detailed interactions among pedestrians and the
characteristics of the environment (e.g., obstacles), thus bringing into question the suitability of these models in studying

∗
Corresponding author. Fax: +861062781475.
E-mail address: [email protected] (X. Zheng).

http://dx.doi.org/10.1016/j.amc.2016.10.040
0 096-30 03/© 2016 Elsevier Inc. All rights reserved.
58 K. Huang, X. Zheng / Applied Mathematics and Computation 298 (2017) 57–64

Fig. 1. Cellular automata model for pedestrian evacuation based on discrete choice principle. All pedestrians update their position and behavior asyn-
chronously in a random sequence. The position of a chosen pedestrian update according to Logit model, and the behavior of the chosen pedestrian update
following by the “best takes over” rule.

the real-world evacuation mechanisms [5,6]. Because microscopic models overcome the above-mentioned shortcomings by
accounting for the local interactive behavior of evacuees, these latter models have drawn more attention. The most popular
microscopic models usually include cellular automata model [7–9] and social force model [10–12]. Especially, the cellular
automata model is a rule-based dynamical model, discrete in space, time, and state variable, and hence well suited for
large-scale simulation and evacuation design. Varas et al [13] and Alizadeh [14] studied the evacuation space with obstacles
by a dynamic cellular automaton model, Yang et al [15] and Zheng et al [16] examined the evacuation process based on
cellular automaton when taking the fire accident into account, Huang et al [17] put forward a static floor field in cellular
automaton model for rooms with internal obstacles and multiple exits.
Intuitively, other than the environmental characteristics, the behavior of pedestrians changes with their movement;
namely, the process of evacuation is also a behavioral evolution process [18]. In [19], we presented initial results that pedes-
trians will adjust their evacuation decision on the basis of the environmental characteristics and local interactive behavior.
Ref. [20] also pointed out that in the process of evacuation, the people do not automatically follow the advice from the
officials, but tend to seek information, assess their own risk, and make independent decisions. It is essential that dynamic
evacuation models incorporate the important role of pedestrian behavior. Brown et al [21] described a hurricane evacu-
ation by a dynamic traffic assignment model which integrate the decision-making process. In fact, besides the dynamics
of decision-making for updating position and behavior, the dependency relationship between evacuees also varied in time.
Here, the dependency relationship represents for the interaction strength of agents linked by an edge in the network. If
the dynamics of dependency relationship is introduced into pedestrian evacuation, how does it affect the movement of
pedestrians?
Complex network is a large set of interconnected nodes connected by links, in which nodes and links are two fundamen-
tal units, that can have different meanings in different situations. In order to better understand the dynamics of different
aspects of complex networks, one may extend the existing network models from static to dynamic by introducing dynamical
factors into the network nodes or links. Weighted evolving network model is a typical dynamical network model. In this
type of network model, the link weight will change accompanying with the dynamics of the linked nodes. Therefore, we
investigate the impact of dynamics of dependency relationship between evacuees on the crowd evacuation by a weighted
evolving network model. Firstly, we place the population in a weighted network, and the dynamics of dependency relation-
ship between evacuees is modeled by the evolution of link weight. Then we study the influence of evolution of link weight
on the evacuation process. Namely, we investigate the effect of evolution of link weight on the evacuation time and coop-
erative behavior. Lastly, through the setting of exits of room, we regulate the behavior of evacuees to boost the efficiency of
evacuation.

2. Methods

First, the evacuation space is discretized into small grids, which can either be empty or occupied by only one pedes-
trian. So all the pedestrians can be described by an undirected network G = (υ , ε ). Here, υ = {1, ..., N} is the pedestri-
ans set, and ε ⊆ υ × υ is the link set. All pedestrians update their position and behavior asynchronously in a random
sequence. In each step, a chosen pedestrian can move to one of their unoccupied positions in the Moore neighborhood
according to a certain probability (See Fig. 1). It is noteworthy that when the number of unoccupied grids of a pedes-
trian is larger than one, the pedestrian has many grids to choose, which is a typical discrete choice problem. In this
article, we adopt the Logit model which had been introduced in detail in the Ref. [19]. In Fig. 1, the pedestrians num-
bered 1–3 represented the update order. With pedestrian 1, for example, they calculate the value of Mxy = kE Exy + kB Bxy for
their five unoccupied neighbors, here Exy and Bxy represent the measurement of environmental characteristics and local
 K. Huang, X. Zheng / Applied Mathematics and Computation 298 (2017) 57–64 59

interactive behavior, kE and kB represent the weight of Exy and Bxy The measurement of environmental characteristics Exy
is computed by the relative distance of the target position (x, y) and real-time position (x , y ), namely, Exy = Dx y − Dxy .

Here, Dxy = min(xe ,ye ) {(xe − x )2 + (ye − y )2 } is the minimal distance between grid (x, y) and all exits, and (xe , ye ) is the

position of exit e. Similarly, Dx y = min(xe ,ye ) { (xe − x ) + (ye − y ) } is the minimal distance between grid (x , y ) and all
2 2

exits. Because of the lack of evacuation resource, we model their interactions by the well-established social dilemmas. So,
the measurement of behavioral interaction Bxy is computed by the difference between target fictional utility and real-time
utility, namely, Bxy = Uxy − Ux y . Here, Ux y is the utility of the pedestrian in real-time position, Uxy is the fiction utility of
 R S
the same pedestrian in target position (x, y). Here, Uxy = j∈ wi j sTi Ps j , and the matrix P = [ ] stands for the payoff
i T P
matrix between cooperative (si = [1, 0]T ) and non-cooperative (si = [0, 1]T ) behavior. For the sake of simplicity, we use R = 1,

S = 0.15, T = b, P = 0.16. Likewise, Ux y = j∈ wi j sTi Ps j . It’s important to note that when a pedestrian in target position (x, y)
i
and real-time position (x , y ), the set of pedestrian i’s neighbors i is not identical. According to the calculation of Mxy , the
pedestrian move to the grid (x, y) according a certain probability pxy calculated by the Logit model, namely,
exp (Mxy )
pxy =  , (1)
x,y exp (Mxy )

After a chosen pedestrian updates his position, he adjusts the dependency relationship with his neighbors. Initially, we
assume that all the links between two pedestrians equal to 1, namely, wxy (0) = 1, (x, y) ∈ ε . Suppose that the update rule of
dependency relationship with Markov property. In detail, if a neighbor y of the chosen pedestrian x in the new position is
not a neighbor of x in the last step, then the link weight of these two pedestrians set as 1, namely wxy (t + 1) = wyx (t + 1) = 1.
On the contrary, if a neighbor y of the chosen pedestrian x in the new position is just one of the neighbor of x in the last
step, then the link weight of these two pedestrians evolve according to the “Win-stay, Lose-shift” principle [22],
 
wxy (t + 1 ) = wxy (t ) + sgn Pxy (t ) − Ūx (t ) × 
(2)
wxy (t + 1 )=wyx (t + 1 )

Here, Pxy (t ) = wxy sTx P sy , Ūx (t ) = j∈ wi j sTi Ps j /|i |,  is the evolution strength of link weight. It is noteworthy that 
i
is a factor of δ . The sgn is a function defined as follow:

1, x > 0
sgn(x ) = 0, x = 0 (3)
−1, x < 0
In order to guarantee the boundedness of the link weight, we assume that the range of link weight falls into wxy ∈ [1 − δ ,
1 + δ ] ∀x, y. Certainly that when a pedestrian leaves the position in the last step, the link weight between the pedestrian
and their unconnected neighbors sets as 0. Without much loss of generality, we use δ =0.4 [23]. In addition to the update of
position and weight link, the behavior of the chosen pedestrian will update following by the “best takes over” rule [24–29],
namely, the focal pedestrian will select the behavior of a player in their self-inclusive Moore neighborhood who received
the greatest total payoff in the previous round.

3. Results

We first describe the results of a typical situation, i.e., the evacuation of a large room sized L × L with one exit in the
middle of the wall. Initially, we assumed that the density of the crowd is ρ = N/L2 , and each pedestrian selected cooperative
or non-cooperative behavior randomly. In Section 3.1, we focus on the influence of the parameters b and ρ by scanning the
range of these parameter, and observe interesting collective phenomena in the evacuation process exhibited by the model.
In Section 3.2, we investigate the underlying mechanism of the collective phenomena by carefully examining the equilibrium
state of the evolution of behavior. Section 3.3 is devoted to a simple optimization of the crowd evacuation, namely, by setting
the length of two exits to regulate the behavior between evacuees, which is further boosting the efficiency of evacuation.

3.1. The impact of parameters on the crowd evacuation

First, we study the impact of parameter b on the evacuation time and cooperative behavioral evolution when the link
weight evolution strength  varied. Here, the value of parameter b can be viewed as a measure of the degree of emergency.
According to the value of parameter b, the degree of emergency can divided into three range: low emergency range, medium
emergency range, high emergency range. As the degree of emergency increases, the evacuation time as well as the fraction
of cooperative behavior of the crowd decreases, which is in line with the results in [19]. For low emergency range, the evac-
uees will perform a random walk and just leave the exit by chance, which enhance the probability of interaction between
evacuees and their neighbors. Therefore, the fraction of cooperative behavior of the crowd is in a high level. Simultaneously,
the link weight evolution will promote the cooperative behavior though introduce the heterogeneous of the crowd. From
Fig. 2, we can see that the larger the  is, the larger the evacuation time as well as the fraction of cooperative behavior of
60 K. Huang, X. Zheng / Applied Mathematics and Computation 298 (2017) 57–64

Fig. 2. The impact of parameter b on the evacuation time and fraction of cooperative behavior when the link weight evolution strength  varied. Here,
the size of the room is 63, the density of the crowd is 0.6, and kE = 0.5, kB = 1.

Fig. 3. Average evacuation time and fraction of cooperative behavior for a large room. Here, the size of the room is 63, the value of b is 1.1, and kE = 0.5,
kB = 1.

the crowd will be. For medium emergency range, the evacuees will perform an orientation walk, which broken the order
of crowd. When =0.05, the heterogeneous of the crowd is not enough to resist the negative effect of orientation walk, so
the fraction of cooperative behavior decreased, which further weaken the local interactive behavioral factor for the process
of evacuation, thereby, the efficiency of the crowd evacuation is increased. However, when  increased, the heterogeneous
of the crowd is increased, which strengthen the local interactive behavioral factor for the process of evacuation, and the
heterogeneous of the crowd is enough to resist the negative effect of orientation walk, so the evacuation time as well as the
fraction of cooperative behavior of the crowd is increased. For high emergency range, all the evacuees rushed to the exits
without hesitate, which further broken the order of crowd, and the neighbors of a pedestrian change frequently, the crowd
in a disorder state. At this time, the local interactive behavioral factor for the process of evacuation can be omitted, and all
the evacuees will choose the non-cooperative behavior eventually.
 K. Huang, X. Zheng / Applied Mathematics and Computation 298 (2017) 57–64 61

Fig. 4. Typical snapshots of the evacuation process. From top panel to bottom panel, the parameter  is 0, 0.05 and 0.2, respectively. The time step from
left to right is 0, 40, 250 and 10 0 0, respectively. Here, the size of the room is 63, the value of b is 1.1, and kE = 0.5, kB = 1.

In the following, we study the impact of crowd density ρ on the evacuation time and cooperative behavioral evolution
when the link weight evolution strength  varied (Fig. 3). Here, we mainly focus on the medium emergency range. Gen-
erally, as the crowd density ρ increasing, the evacuation time of the crowd is increased, and the evacuation time is nearly
proportional to the crowd density, which is in agreement with the conclusion in Ref. [30]. Ref. [23] claimed that the dy-
namics of dependency relationship can introduce the heterogeneous of crowd, which can promote the cooperative behavior
when the topology of crowd is static. However, when the link weight evolution strength is low (=0.05), the cooperative
behavior promotion effect introduced by the dynamics of dependency relationship is not enough to resist the negative effect
of orientation walk, the fraction of cooperative behavior in a low level, which further weaken the local interactive behavioral
factor for the process of evacuation, thus the time of the crowd evacuation is decreased. Particularly, when the  increased,
the heterogeneous of the crowd is increased, which enlarger the evacuation time. Obviously, this is in consistence with Fig.
2. Therefore, we can conclude that a weaker link weight evolution weakens the local behavioral factor in the evacuation
process, which further enhance the efficiency of crowd evacuation. On the contrary, a stronger link weight evolution pro-
motes heterogeneity of the crowd, thus strengthening the local behavioral factor in the evacuation process and prolonging
the evacuation time.
In order to observe the evacuation process clearly, we present some typical snapshots in the evacuation process for
different link weight evolution strength  (Fig. 4). Initially, we assume that all pedestrians are distributed in the middle
of the room, and each pedestrian chooses cooperative or non-cooperative behavior randomly. There exist some results that
mobility of individual in the lattice is not favorable for the cooperation [31,32]. Comparing to a traditional case, i.e., the
link weight of connected pedestrian is 1 (=0), smaller (=0.05) is not favorable for the pedestrian gathering. From the
middle panel of the snapshots, we can see that the pedestrian walk randomly, and the crowd in disorder. Thus, the influence
of local interactive behavioral factor is declined. But for the bottom panel of the snapshots (=0.2), we can see that the
distribution of pedestrians was tight and the pedestrians move synchronously, which further enhance the influence of local
62 K. Huang, X. Zheng / Applied Mathematics and Computation 298 (2017) 57–64

Fig. 5. Equilibrium state of the evolution of behavior when pedestrians walk in a closed space. The value of  equal to 0, and 0.4 for the top and bottom
panel, respectively. The curves in the left subfigure are the time courses of fraction of cooperative behavior, cooperative clusters and behavioral clusters.
Here, the size of the room is 63, the density of the crowd is 0.6, and b = 1.15, kE = 0.5, kB = 1.

interactive behavioral factor. So we can claim the results that a weaker link weight evolution strength is not favorable for
the promotion of cooperative behavior but in favor of the evacuation efficiency, on the contrary, a stronger link weight
evolution strength is favorable for the promotion of cooperative behavior but not in favor of the evacuation efficiency.

3.2. Equilibrium state of the evolution of behavior

According to Section 3.1, we can see that the dynamics of dependency relationship between evacuees have a great ef-
fect on the evacuation process. However, due to the presence of an exit, the behavior of pedestrians cannot evolve to the
equilibrium state. In [23], we have analyzed the dynamics of dependency relationship for the evolution of cooperative be-
havior when the topology of the individual is static. In the following, we investigate the equilibrium state of the evolution
of behavior when pedestrians walk in a closed space, which can help us understanding the underlying mechanism of the
collective phenomena in the evacuation process.
Suppose that the pedestrians distribute in a closed space, and the update rule of position and behavior is similar with
the case in Section 3.1 except for the exit (omit the exit factor for the walking of pedestrian). The top and bottom panel of
Fig. 5 show the collective behavior and behavioral evolution when =0 and =0.4, respectively. The larger the value of 
is, the larger the fraction of cooperative behavior will be. In addition, the difference of behavior will adjust the direction of
the pedestrian, which is a positive feedback for the promotion of cooperative behavior. In the first stage, all the pedestrians
distribute in the lattice randomly. As the time progresses, cooperative pedestrians come together and the number of smaller
clusters declines sharply. Comparing the top and the bottom panel of Fig. 5, we can see that when =0.4, the number
of behavioral clusters increase at first and then descend, which means larger  is more appropriate for the gathering of
 K. Huang, X. Zheng / Applied Mathematics and Computation 298 (2017) 57–64 63

Fig. 6. Evacuation of a room with two exits. The left subfigure is a schematic of a room with two exits, each exit can pass only one pedestrian, and the
length between two exits is l. The right subfigure is evacuation time as a function of the length between two exits. Here, the size of the room is 64, the
density of the crowd is 0.8, and b = 1.1, kE = 0.5, kB = 1.

pedestrian. In addition, in the behavioral clusters, the cooperators will increase the link weight with other cooperators,
which helps stabilize the existing cooperative behavior and makes all pedestrians choose cooperative behavior in the end.

3.3. Regulation of the crowd evacuation

From the Sections 3.1 to 3.2, it can be concluded that the local interactive behavioral factor affects the evolution of
cooperation though the typical collective phenomenon, namely, gathering of pedestrians. Gathering of pedestrians has a
great effect on the evacuation process. Pedestrians form clusters by coming together, whereupon those pedestrians who
belong to the same cluster move synchronously. However, the gathering of pedestrians has the potential negative effect on
the efficiency of evacuation. For this reason, we will introduce a simple experiment for safety evacuation in the following
section.
The thought of this experiment is that if we can guide the pedestrians to reduce to gathering behavior, did it work for
the increasing of efficiency of evacuation? In the following, we investigate how the evacuation time change if the length
between two exits varied. Apparently, an appropriate length between two exits can reduce the gathering of pedestrians.
The left subfigure of Fig. 6 is a schematic of a room with two exits, each exit can pass only one pedestrian, and the length
between two exits is l. Particularly, when l = 0, the two exits combined into one exit with the size of 2. From the right
subfigure of Fig. 6, we can see that no matter the value of  is, the efficiency of evacuation first increase then descend,
which is similar with Ref. [30]. However, when =0.05, which is not favorable for gathering, but it is the most optimal for
the evacuation. In addition, we can see that when l ≈ 2L , the evacuation time is the lowest, which means that this is the
best setting of two exits to reducing to gathering behavior. So, it can be concluded that guiding of gathering behavior is
beneficial for the increasing of efficiency of evacuation.

4. Conclusion

The process of pedestrian evacuation is a decision-making process. In this process, pedestrians update their position
and behavior with a desire for safety and efficiency. The factors which affect the decision-making including environmental
characteristics and local interactive behavior. For the local interactive behavior, the dynamics of dependency relationship is
often omitted. So, we examine how the dynamics of dependency relationship between evacuees affect the evacuation time
and cooperative behavior. We place the population in a weighted network, and then model the dynamics of dependency
relationship between evacuees by letting link weights to evolve. By careful examination of the parameter space, we can
conclude that a weaker link weight evolution weakens the local behavioral factor for the evacuation process, which further
enhance the efficiency of crowd evacuation. On the contrary, a stronger link weight evolution promotes heterogeneity of
crowd, thus strengthening the local behavioral factor in the evacuation process and prolonging the evacuation time. In
addition, we observed that the gathering of pedestrians has the potential negative effect on the efficiency of evacuation, so
we conduct a simple experiment: evacuation of a room with two exits. By adjusting the length of two exits, we can guide
the gathering behavior and regulate the crowd evacuation. This experiment can also provide an explanation for why the
exits should located symmetrical in the middle of a wall.
64 K. Huang, X. Zheng / Applied Mathematics and Computation 298 (2017) 57–64

Acknowledgments

The authors are grateful to the financial support by the National Key Research and Development Program of China (Grant
nos. 2016YFC0801200 and 2016YFC0801300), and by the National Natural Science Foundation of China (Grant nos. 71225007
and 61575103). We also like to thank the anonymous reviewers for their helpful comments.

References

[1] A. Schadschneider, W. Klingsch, H. Kluepfel, T. Kretz, C. Rogsch, A. Seyfried, Evacuation Dynamics: Empirical Results, Modeling And Applications,
Encyclopedia of Complexity And Systems Science, Springer, New York, 2009, pp. 3142–3176.
[2] S. Gwynne, E. Galea, M. Owen, P. Lawrence, L. Filippidis, A review of the methodologies used in the computer simulation of evacuation from the built
environment, Build. Environ. 34 (1999) 741–749.
[3] D. Helbing, I. Farkas, P. Molnar, T. Vicsek, Simulation of pedestrian crowds in normal and evacuation situations 21 (2) (20 0 0) 21–58.
[4] X. Zheng, T. Zhong, M. Liu, Modeling crowd evacuation of a building based on seven methodological approaches, Build. Environ. 44 (2009) 437–445.
[5] W. Yu, R. Chen, L. Dong, S. Dai, Centrifugal force model for pedestrian dynamics, Phys. Rev. E 72 (2005) 1–7.
[6] M. Twarogowska, P. Goatin, R. Duvigneau, Macroscopic modeling and simulations of room evacuation, Appl. Math. Model. 38 (2014) 5781–5795.
[7] G. Sun, Z. Jin, L. Song, A. Chakraborty, B.L. Li, Phase transition in spatial epidemics using cellular automata with noise, Ecol. Res. 26 (2011) 333–340.
[8] J. Tanimoto, A. Hagishima, Y. Tanaka, Study of bottleneck effect at an emergency evacuation exit using cellular automata model, mean field approxi-
mation analysis, and game theory, Phys. A: Stat. Mech. Appl. 389 (2010) 5611–5618.
[9] W.F. Yuan, K.H. Tan, An evacuation model using cellular automata, Phys. A: Stat. Mech. Appl. 384 (2007) 549–566.
[10] D. Helbing, P. Molnar, Social force model for pedestrian dynamics, Phys. Rev. E 51 (1995) 4282–4286.
[11] D. Helbing, I. Farkas, T. Vicsek, Simulating dynamical features of escape panic, Nature 407 (20 0 0) 487–490.
[12] F. Zanlungo, T. Ikeda, T. Kanda, Social force model with explicit collision prediction, Europhys. Lett. (Epl) 93 (2011) 1–6.
[13] A. Varas, M.D. Cornejo, D. Mainemer, B. Toledo, J. Rogan, V. Munoz, J.A. Valdivia, Cellular automaton model for evacuation process with obstacles, Phys.
A: Stat. Mech. Appl. 382 (2007) 631–642.
[14] R. Alizadeh, A dynamic cellular automaton model for evacuation process with obstacles, Saf. Sci. 49 (2011) 315–323.
[15] L. Yang, W. Fang, R. Huang, Z. Deng, Occupant evacuation model based on cellular automata in fire, Chin. Sci. Bull. 47 (2002) 1484–1488.
[16] Y. Zheng, B. Jia, X. Li, N. Zhu, Evacuation dynamics with fire spreading based on cellular automaton, Phys. A: Stat. Mech. Appl. 390 (2011) 3147–3156.
[17] H. Huang, R. Guo, Static floor field and exit choice for pedestrian evacuation in rooms with internal obstacles and multiple exits, Phys. Rev. E 78
(2008) 1–6.
[18] K. Huang, X. Zheng, Y. Yang, T. Wang, Behavioral evolution in evacuation crowd based on heterogeneous rationality of small groups, Appl. Math.
Comput. 266 (2015) 501–506.
[19] K. Huang, X. Zheng, Y. Cheng, Y. Yang, Behavior-based cellular automaton model for pedestrian dynamics, Appl. Math. Comput. 292 (2017) 417–424.
[20] A. Pel, M. Bliemer, S. Hoogendoorn, A review on travel behaviour modelling in dynamic traffic simulation models for evacuations, Transportation 39
(2012) 97–123.
[21] C. Brown, W. White, C. van Slyke, J.D. Benson, Development of a strategic hurricane evacuation-dynamic traffic assignment model for the Houston,
Texas, Region, Transp. Res. Rec. 2137 (2009) 46–53.
[22] M. Perc, The Matthew effect in empirical data, J. Royal Soc. Interface 11 (2014) 1–15.
[23] K. Huang, X. Zheng, Z. Li, Y. Yang, Understanding cooperative behavior based on the coevolution of game strategy and link weight, Sci. Rep. 5 (2015)
1–7.
[24] M. Perc, A. Szolnoki, Social diversity and promotion of cooperation in the spatial prisoner’s dilemma game, Phys. Rev. E 77 (2008) 1–5.
[25] M. Perc, J. Gomez-Gardenes, A. Szolnoki, L.M. Floria, Y. Moreno, Evolutionary dynamics of group interactions on structured populations: a review, J.
Royal Soc. Interface 10 (2013) 1–17.
[26] Z. Wang, A. Szolnoki, M. Perc, Optimal interdependence between networks for the evolution of cooperation, Sci. Rep. 3 (2013) 1–7.
[27] M. Perc, Premature seizure of traffic flow due to the introduction of evolutionary games, New J. Phys. 9 (2007) 1–17.
[28] Z. Wang, L. Wang, A. Szolnoki, M. Perc, Evolutionary games on multilayer networks: a colloquium, Eur. Phys. J. B 88 (2015) 1–15.
[29] M. Perc, A. Szolnoki, Coevolutionary games-A mini review, Biosystems 99 (2010) 109–125.
[30] A. Kirchner, A. Schadschneider, Simulation of evacuation processes using a bionics-inspired cellular automaton model for pedestrian dynamics, Phys.
A: Stat. Mech. Appl. 312 (2002) 260–276.
[31] M. Enquist, O. Leimar, The evolution of cooperation in mobile organisms, Animal Behav. 45 (1993) 747–757.
[32] D. Helbing, W. Yu, The outbreak of cooperation among success-driven individuals under noisy conditions, in: Proceedings of the National Academy of
Sciences of the United States of America, 106, 2009, pp. 3680–3685.