A modified floor field model is proposed to simulate the pedestrian evacuation behavior in a room with multiple exits. The modification relies upon introduction of a so-called semidynamical floor field which additionally embodies two cognition coefficients related to exit width and pedestrian distribution around exits. The energy consumption and time requirement of evacuees are numerically investigated and the Pareto optimum of evacuation efficiency is obtained by selecting the combinations of the two cognition coefficients.
National Natural Science Foundation of China714010837130102871661024Inner Mongolia Autonomous RegionNJYT15B061. Introduction
Study on pedestrian dynamics has attracted many physicists over the past few years. Various collective behaviors and self-organization phenomena have been observed from the view point of complexity of the pedestrian flow system. For understanding these phenomena, many models have been developed, for example, the social force model [1–3], the fluid dynamic model [4–9], and the cellular automata model [10–21].
The social force model is a deterministic continuum model in which the interactions between pedestrians are governed by the social force or social field [1, 2]. Using this model, such phenomena as the arching around exit, the lane formation in counter flow, and the oscillation of flow direction at narrow door can be reproduced.
The fluid dynamic model describes how density and velocity change over time with the use of partial differential equation. This model can be used to investigate exactly the dynamics at the exits during an evacuation [7] and study some typical pedestrian behaviors, for example, the “thinking fluid” behavior and the overcompression effect [8, 9].
The rule-based cellular automata model is discrete in space and time. All pedestrians are represented by particles. Such typical phenomena as arching, jamming, kin behavior, faster-is-slower, lane formation, and sidle effect can be simulated by this kind of models [10–21]. Floor field model is one kind of discrete simulation models. At each time step, every pedestrian moves from one lattice to the neighboring lattice in terms of a transition probability. This model can be used to investigate various collective effects and the self-organization encountered in pedestrian dynamics [10, 11].
Modeling the evacuation process of pedestrians in a room with multiple exits has to consider the strategy of selecting exits. In the conventional floor field models, the exit selection strategy is mainly based on the herding behavior and the use of knowledge about the shortest path to each exit [10, 11]. In some works, the occupant density around each exit is considered as an important factor affecting the exit selection strategy [15, 16]. In this paper, we improve the floor field model through introducing two cognition coefficients associated with the exit width and the pedestrian distributions around exits, respectively.
The energy consumed in evacuation process is seldom studied by previous researches. However, if a space for evacuation is large enough to make some pedestrians expend more energy than their capabilities, he or she will fail to evacuate successfully and, on the contrary, they may become obstacles to other pedestrians. In this study, we will investigate the energy consumption of evacuees besides the time requirement and find the Pareto optimum of evacuation efficiency.
2. Model
Let a space for evacuation be represented by two-dimensional square lattices. The size of each lattice is approximately 40×40 cm2. Each lattice can be either empty or occupied by one pedestrian. Each time step is real time of evacuation based on different movement velocity of pedestrians. The length of one time step is 0.3 seconds in this study. This implies a walking speed of approximately 1.33 m/s.
In each time step, pedestrians move only one lattice site in the forward, backward, left, or right direction or remain unmoving. They probabilistically select a neighboring lattice site in these directions to move using the transition probability. The transition probability is computed using the static and dynamic floor fields of those lattices.
In the simulation of evacuation process, the static floor field S represents the degree of attractiveness of each lattice for pedestrians. The static floor field does not evolve with time and is not changed by the presence of the pedestrians. It is often given by the value depending on the distance from the exit. The dynamic floor field D represents the characteristic that pedestrians tend to follow their predecessors. This is implemented using bosons dropped by pedestrians, having their own dynamics through diffusion and decay controlled by probabilities σ and δ. The dynamic floor field D is the number of bosons remaining on the cell at each time step. These two types of floor fields have been historically used as fundamental and essential components of the floor field model.
The static floor field defined above reflects the attractiveness degree of each lattice for pedestrians, but the exit width and the pedestrian distribution around exits have not been considered. These two factors should have impacts on the attractiveness of some lattices and eventually influence the movement behavior of pedestrians. For this reason, the floor field model has to be improved so as to simulate the evacuation process more precisely.
2.1. Occupant Density
The occupant density around exit is introduced to characterize the pedestrian distribution around exits in the evacuation process. The occupant density Orm of exit m can be defined as the number of occupied cells in the effect area of exit m with radius r. The effect area is a special region around the exit. Figure 1 provides an illustration for calculating the occupant density Orm [16]. The width of the exit is 2 cells. The effect area contains sixteen intact cells which are located within a half-round with radius of 4 cells. Five black colored cells are occupied. Thus, the occupant density of this exit is 5.
Illustration for calculating occupant density Orm.
2.2. Semidynamical Floor Field
The semidynamical floor field is introduced for considering the effects of exit width and pedestrian distribution around exit. Let L be the total width of all exits, lm the width of exit m, and T the total number of exits. We have ∑m=1Tlm=L.
The semidynamical floor field SDijm of lattice site (i,j) to exit m is defined as(1)SDijm=maxmmaxk,hZkhm-Zijm,where (h,k) is a lattice. Zijm is determined by the shortest distance Mijm from lattice (i,j) to exit m, the width lm of exit m, and the occupant density Orm of exit m, that is,(2)Zijm=Mijm+αL-lm+βOrm,where Mijm=minni-inm2+j-jnm2, (inm,jnm) is the coordinate of the nth cell of exit m. The parameter α is a cognition coefficient which indicates the pedestrian’s capability of perceiving the effect of exit width. A larger α value means a bigger consideration of the effect of exit width or pedestrian being more sensible to the width effect. The parameter β is another cognition coefficient which indicates the pedestrian’s sensitivity to the people distribution around each exit. A larger β means more consideration on the occupant density. When α=0 and β=0, the effects of both exit width and pedestrian distribution are not considered and the semidynamical floor field is reduced to the static floor field.
2.3. Transition Probability
The transition probability from a lattice site to its neighboring lattice site (i,j) toward exit m, Pijm, is computed using the following:(3)Pijm=NexpkSDSDijmexpkDDijm1-μijξij,where N is a normalization factor for ensuring ∑(i,j)Pijm=1. kSD and kD are two sensitivity parameters for scaling SDijm and Dijm, respectively. The value of kSD can be regarded as a measure of the pedestrians’ perception about the attractiveness of each lattice site for each exit. The parameter kD for coupling to the dynamic field governs the tendency to follow the lead of others. A large value of kD implies a strong herding behavior. The parameter μij indicates whether the neighboring lattice (i,j) is occupied or not. It equals 1 if the lattice is occupied and 0 otherwise. The parameter ξij is related to the existence of obstacles. It is 0 if (i,j) is occupied by obstacles (e.g., walls) and 1 otherwise.
2.4. Unit Energy Consumed
We use the total movement steps of all pedestrians, rather than the total evacuation time, to represent the energy consumption. This says that when a pedestrian moves to one of neighboring lattices, one unit of energy consumption is added to the counter. The average energy consumed per person is termed as unit energy consumed. These two performance indexes are generally not equivalent to each other because, in congested situation, some pedestrians may not move if their neighboring sites are occupied by others while the time is spent. In addition, if some pedestrians change the exit choice frequently in the evacuation process, the energy consumption may increase although the evacuation time is reduced.
2.5. Pareto Optimum
The multicriteria optimization method has widely been applied in designing various mechanical structures and dynamic systems. Due to the basic nature of conflict among multiple objectives, a set of compromising solutions called Pareto optima exists. The determination of this set is essential for decision-makers to understand the trade-off relationship among alternatives [21].
In this study, we use two criteria, namely, the unit energy consumed and the evacuation time, to measure the efficiency of evacuation mechanism. We investigate how the two criteria are affected by the cognition coefficients α and β. We certainly hope to achieve such a result that these two indexes are simultaneously minimized by optimizing the two parameters α and β. However, this is basically impossible as discussed in the previous subsection, and we have to turn to finding the Pareto optimum of the evacuation mechanism.
3. Simulation
Consider a special zone in a room and suppose pedestrians are randomly distributed in the zone and attempt to escape. Without loss of generality, in Figure 2, the room is discretized into 50×50 cells and has four exits allocated at the centre of each wall. The width of exit n is n cells, n=1,2,3,4. In the specified zone (the small square marked by dotted line, with a size 22×22 cells), there are 400 pedestrians at the initial stage of the simulation.
Room structure.
The sensitivity parameters adopted here are kSD=3 and kD=0.5. The decay probability is σ=0.5 and the diffusion probability δ=0.5. The radius of effect area is r=10 cells. We carry out 10 simulations for each set of parameters and compute the averages.
To investigate the role of each exit, the pedestrian number evacuated from each exit is shown in Figures 3 and 4, when α and β take different values. It can be observed from Figures 3(a)–3(d) that, for each α value, with increase of β value, Exits 2 and 3 are first chosen by many pedestrians and then by less ones. This is because the special zone is located closer to these two exits and pedestrians do not care much about the occupant densities around exits. A larger β value means pedestrians are more sensitive to the occupant densities; as a result, some pedestrians turn to choosing exits 1 and 4. Another important finding is that the α value influences the number of pedestrians evacuated from each exit, but the influence degree is not linear and monotonic.
Pedestrian number evacuated from each exit against cognition coefficient β for different α values.
Pedestrian number evacuated from each exit against cognition coefficient α for different β values.
Figure 4 shows that, with the increase of α value, for each β value, the numbers of pedestrians evacuated from Exits 1 and 2 decrease, the one from Exit 4 increases, and the one from Exit 3 first ascends and then descends basically. This is because in the evacuation process, the occupant density in the effect area of each exit intends to be identical and the cognitive coefficient α gradually comes to play a role. Consequently, more pedestrians turn to evacuating from Exit 4 due to its width.
Figure 5 shows that the unit energy consumed monotonically increases with β for each α value. This verifies such a view point that the energy consumption is proportional to the attitude toward congestion degree.
Unit energy consumed against β for different α values.
Finally, we investigate the two criteria for evaluating the evacuation efficiency. Due to the conflict of the two criteria, it is difficult to find the optimal α and β values that simultaneously minimize the unit energy consumed and evacuation time. For solving this problem, the Pareto concept is used here. Figure 6 depicts these two indexes against combination of α and β values. The Pareto optimum points are (α∗,β∗)∈{(0,0.4),(0,0.3),(1,0.3),(3,0.3),(4,0.3),(5,0.3),(0,0.2), (1,0.2), (1,0),(0,0.1),(1,0.1)}.
Pareto optima between unit energy consumed and evacuation time.
4. Conclusions
In this paper, two criteria, namely, the unit energy consumed and the evacuation time, are exploited to evaluate the evacuation mechanism. The unit energy consumed is seldom mentioned by previous researchers. An improved floor field model is proposed to simulate the evacuation process in a room with multiple exits where pedestrians are distributed heterogeneously. The improvement relies on the modification to the conventional static floor field by incorporating such new factors as exit width and the pedestrian distribution around exits. Simulation results show that the evacuation capacity of each exit is sensitive to the pedestrians’ cognition to these two factors. The Pareto optima are numerically obtained, which verifies the conflict of the two criteria.
Conflicts of Interest
The authors declare that there are no conflicts of interest regarding the publication of this paper.
Acknowledgments
This study was supported by the National Natural Science Foundation of China (Grants nos. 71401083, 71301028, and 71661024) and the Program for Young Talents of Science and Technology in Universities of Inner Mongolia Autonomous Region (NJYT15B06).
HelbingD.MolnarP.Social force model for pedestrian dynamics19955154282428610.1103/PhysRevE.51.42822-s2.0-35949007691HelbingD.FarkasI.VicsekT.Simulating dynamical features of escape panic200040768034874902-s2.0-003472707910.1038/35035023ZengW.ChenP.NakamuraH.Iryo-AsanoM.Application of social force model to pedestrian behavior analysis at signalized crosswalk2014401431592-s2.0-8489387713710.1016/j.trc.2014.01.007HendersonL. F.The statistics of crowd fluids197122952843813832-s2.0-3524885671310.1038/229381a0JiangY.-Q.ZhangW.ZhouS.-G.Comparison study of the reactive and predictive dynamic models for pedestrian flow201644151612-s2.0-8494190482310.1016/j.physa.2015.08.04716364ZhengX.ZhongT.LiuM.Modeling crowd evacuation of a building based on seven methodological approaches20094434374452-s2.0-5634912822710.1016/j.buildenv.2008.04.002AndreianovB.DonadelloC.RosiniM. D.Crowd dynamics and conservation laws with nonlocal constraints and capacity drop201424132685272210.1142/S0218202514500341MR3260283Zbl1307.351662-s2.0-84929606725HughesR. L.A continuum theory for the flow of pedestrians20023665075352-s2.0-003664331810.1016/S0191-2615(01)00015-7ColomboR. M.RosiniM. D.Pedestrian flows and non-classical shocks200528131553156710.1002/mma.624MR2158218Zbl1108.900162-s2.0-23944512223BursteddeC.KlauckK.SchadschneiderA.ZittartzJ.Simulation of pedestrian dynamics using a two-dimensional cellular automaton20012953-450752510.1016/S0378-4371(01)00141-8Zbl0978.90018KirchnerA.SchadschneiderA.Simulation of evacuation processes using a bionics-inspired cellular automaton model for pedestrian dynamics20023121-226027610.1016/S0378-4371(02)00857-9Zbl0997.900172-s2.0-0036723401YuY. F.SongW. G.Cellular automaton simulation of pedestrian counter flow considering the surrounding environment200775410.1103/PhysRevE.75.0461120461122-s2.0-34247364626HuangH. J.GuoR. Y.Static floor field and exit choice for pedestrian evacuation in rooms with internal obstacles and multiple exits2008782610.1103/PhysRevE.78.0211310211312-s2.0-50249155889YangX.WangB.QinZ.Floor Field Model Based on Cellular Automata for Simulating Indoor Pedestrian Evacuation201520151082030610.1155/2015/8203062-s2.0-84947474524XuY.HuangH.-J.Simulation of exit choosing in pedestrian evacuation with consideration of the direction visual field2012391499110002-s2.0-8465516313010.1016/j.physa.2011.10.022ZhaoH.GaoZ.Reserve capacity and exit choosing in pedestrian evacuation dynamics201043101010500110.1088/1751-8113/43/10/105001MR2593990GuoR.-Y.HuangH.-J.Logit-based exit choice model of evacuation in rooms with internal obstacles and multiple exits20101932-s2.0-7794967894310.1088/1674-1056/19/3/030501030501JianX. X.ZhangX.Potential field cellular automata model for pedestrian evacuation in a domain with a ramp20142014710.1155/2014/7142677142672-s2.0-84896264236GuoR.-Y.HuangH.-J.WongS. C.A potential field approach to the modeling of route choice in pedestrian evacuation2013201322-s2.0-8487474178910.1088/1742-5468/2013/02/P02010P02010GoldszteinG. H.A Mathematical Model of the Formation of Lanes in Crowds of Pedestrians Moving in Opposite Directions20152015787280710.1155/2015/8728072-s2.0-84934300473ZhangW. H.GaoT.A min-max method with adaptive weightings for uniformly spaced Pareto optimum points20068428176017692-s2.0-3374870425610.1016/j.compstruc.2006.04.007