† Corresponding author. E-mail:
Project supported by the National Natural Science Foundation of China (Grant No. 71904116) and the Fund from the Shanghai Municipal Commission of Science and Technology, China (Grant Nos. 19DZ1209600 and 18DZ1201500).
Exit choice is one of the most important pedestrian behaviors during evacuation. Distance to the exit is a generally recognized factor influencing expected moving time to the exit. Visual range determines how much information a pedestrian can perceive, thus the number of pedestrians within the visual field can be used to estimate waiting time at the exit. Besides, the choice firmness that reflects the degree to which a pedestrian would persist in his/her previous choice of exit is proposed. By integrating game theory into a cellular automata simulation framework, the pedestrian exit choice mechanism is investigated and explicitly modeled in this paper. A systematic analysis of the key factors influencing pedestrian evacuation is conducted, including visual radius and choice firmness of a pedestrian, initial crowd distribution of the room, exit layout as well as exit width. It is found that low choice firmness level can lead to unnatural pedestrian behavior such as wandering, which is adverse to evacuation. The longer the pedestrian’s visual radius, the earlier the pedestrian can determine his/her final selection of the exit. Compared with the scenario where the pedestrians are randomly distributed, pedestrians clustered together in a corner of the room lead to high crowd density and imbalanced use of exits. Furthermore, the exit layout and exit width also have a certain influence on pedestrian evacuation process. The results of this paper may be of benefit to the formulation of behavioral rules in other pedestrian simulation models.
With social and economic development as well as the acceleration of urbanization, urban population expanded rapidly in recent years. The gathering of many people becomes a common phenomenon. However, due to the lack of effective management and control, crowd-related disasters such as stampede are frequently reported, which has brought serious threats to public safety. Therefore, pedestrian evacuation dynamics has attracted increasing attention of researchers in transportation, urban planning and other relevant fields. Establishing a simulation model and carrying out scenario-based simulations is one of the most effective ways to study pedestrian evacuation dynamics and evaluate performance of building layouts. The pedestrian evacuation models can usually be divided into two categories: macroscopic models and microscopic models. The macroscopic models treat the population as a whole, ignoring the interaction between individuals. They can describe pedestrian agglomeration characteristics, such as flow and speed, but may fail to demonstrate more detailed information about pedestrian flows. The pedestrian simulation model is usually microscopic, which takes each pedestrian as an individual, and focuses on the interactions between pedestrians and between the pedestrian and environment. Considering different divisions on space and time, microscopic models can further be divided into continuous and discrete ones, of which the social force model and cellular automata (CA) model are the most important representatives, respectively. The social force model was proposed by Helbing et al.,[1,2] which has been improved and applied in varied scenarios.[3,4] The CA model is greatly employed because of its simple rules and high computational efficiency.[5] Moreover, the original CA model can be combined, extended or improved according to the characteristics of each pedestrian and the surrounding environment.[6–11]
To better represent pedestrian evacuation dynamics, pedestrian behavior should be carefully considered and reasonably modeled. In terms of pedestrian behavior modeling, conflict over the moving space[12–18] and decision-making in exit selection[19–21] are the most widely studied topics. In fact, the exit selection is a complex decision-making process. Individuals would usually consider a number of factors in exit selection, and the distance to the exit is the most influencing one. Other import factors include crowd density around the exit, discomfort level, etc. In the modeling of pedestrian selection behavior, commonly used methods include random utility theory[22] and multi-logit model.[23] Game theory is a recognized tool to explain human behavior, especially when individuals face competing situation with each other. Ehtamo et al.[20] established a model of exit selection based on the game theoretic concept of best-response dynamics, where each player updates his/her strategy periodically by optimally responding to other players’ strategies. Lo et al.[21] proposed a game theory based exit choice model, in which the distance to the exit, crowd density and exit width are taken into account for calculating the payoff matrix, and the Nash equilibrium of the mixed strategy reflects the balance between evacuees and exit congestion. By introducing the floor field, the exit and the optimal path can be obtained. Xu and Huang[24] proposed a modified floor field model to simulate the multi-exit evacuation selection process, in which two cognitive coefficients of exit width and congestion around exit are explicitly considered. Yue et al.[25] combined distance-based strategy and time-based strategy with cognitive coefficients to establish a hybrid strategy of exit selection in pedestrian evacuation. Zhang et al.[11] defined a cost potential field, considering the effects of travel time and discomfort. As an optimal path-choice strategy was obtained, the model demonstrated a faster evacuation of pedestrians from a room and a shorter computation time than the classical floor field CA model. As visibility plays a vital role in pedestrian’s decision-making, and only information within a certain range could be observed and used for selecting exits, an individual-based visual field was considered in some pedestrian evacuation models.[26–29] For example, Xu et al.[29] proposed a modified floor field model, by which the pedestrian evacuation dynamics in a room with multiple exits by considering the directional visual field is simulated. However, most of the existing researches used a unique value for representing the visual range, which usually varies from each other and lacks empirical evidence. Thus a systematic analysis of the influence of visual range on pedestrian evacuation behavior can be interesting and of importance. Furthermore, unnatural movements can occur if the evacuation model does not take a certain pedestrian behavior into consideration, such as choice firmness. It may not be an influencing factor for exit selection but could lead to a hesitation phenomenon that pedestrians wander back and forth between different exits, resulting in the inaccurate simulation of pedestrian evacuation dynamics.[30]
In this paper, the game theory is introduced into a cellular automata model to simulate pedestrian exit choice during evacuation. Meanwhile, the influence of visual range and choice firmness which may lead the pedestrians to hesitate between exits on pedestrian evacuation dynamics are explicitly considered and analyzed. The rest of the paper is organized as follows. In Section
It is assumed that every pedestrian knows the locations of all the exits. The space is divided into a number of equally sized cells in the proposed model. The cell may be occupied by one pedestrian at most, by the space boundary or empty. And the Moore neighborhood is used as shown in Fig.
In this paper, the pedestrian’s visual field is a fan-shaped area with the current position of the pedestrian as the vertex, 2θ as the angle, R as the radius, and the line between the current position of the pedestrian and the center of the exit as the bisector, as shown in Fig.
In the process of evacuation, the exit selection strategy of a pedestrian will be affected by the strategies of other pedestrians. The relationship between pedestrians is competitive, and the goal is to evacuate the room as quick as possible. Assuming that all the pedestrians are rational, the game theory can be used to describe pedestrians’ exit choice behavior. First, the multi-exit room evacuation problem is transformed into a standard game expression: (i) game players; (ii) strategy space; (iii) payoff function; and (iv) the update strategy. Suppose that there are N people in the room for being evacuated, and the number of exits is K. Thus the number of game players is N, the strategy space set is S = {S1,…,SN}, the optional strategy set of player i is Si = {e1,…,eK}, where Si ∈ S. The payoff function is used to estimate the evacuation time of a pedestrian under different strategies. At each time step, the estimated evacuation time of pedestrian i choosing exit ek is Ti(ek). It is composed of
The estimated moving time of pedestrian i choosing exit ek is calculated by Eq. (
Pedestrians will update their strategies at each time step according to other pedestrians in the visual field and surrounding environment. Then, at time step t + 1, individual i updates his/her strategy according to the following equation:
In this paper, pedestrian movement is simulated based on the static field model. In particular, the movement of pedestrian i is determined by the static floor field value SFi, which is measured as the reciprocal of the distance from pedestrian i to the exit ek, and is shown below
The simulation is performed using the algorithm as follows.
(I) Set the initial parameters including room layout and pedestrian distribution.
(II) At each time step, select an exit for each pedestrian and make all the pedestrians move towards their selected exits by the following rules.
(II)-1 Calculate the number of people in the visual field of each pedestrian, get the payoff function by Eq. (
(II)-2 Calculate the floor field according to the selected exit by Eq. (
(II)-3 Record target grids of all the pedestrians. If two or more pedestrians choose the same grid, only one pedestrian is randomly selected to move to it.
(II)-4 Update the positions of all the pedestrians in parallel.
(III) Repeat steps (II)-1 to (II)-4 until all the pedestrians evacuate from the room.
A series of simulations is performed in a rectangular room (12 m × 12 m) with two exits. The cell size is set to be 0.4 m × 0.4 m, thus the average movement distance of pedestrians at each time step is 0.48 m (parallel movement is 0.4 m or diagonal movement is
Simulation cases are conducted by considering different pedestrian visual radii, choice firmness levels, crowd distributions and exit layouts. Rooms with two exits usually have three typical layouts, that is, two exits are on the same side, on the adjacent sides, and on the counter sides, as shown in Fig.
In these simulation cases, 225 pedestrians are distributed randomly or in cluster in the room where two exits are on the counter sides and the exit width is 1.2 m. The range of choice firmness level α is set to be between 0 and 0.2. Specifically, the pedestrian hesitates to choose the exit when the value is small. On the contrary, the larger the value, the more the pedestrian tends to stick to his/her initial selection. When pedestrians are randomly distributed in the room, we can observe from Fig.
When pedestrians are initially distributed in cluster, the evacuation performance is different from that when randomly distributed. It can be seen from Fig.
In simulation,we consider the case where there are 225 pedestrians with choice firmness level α = 0.1, located in the room where two exits are on the counter sides and the exit width is 1.2 m. Two different crowd distributions are considered, that is, pedestrians distributed randomly or in cluster. Figure
Figures
Figure
Three exit layouts are considered, that is, two exits are on the same side, on the adjacent sides and on the counter sides of the room. In these simulation cases, exit width is 1.2 m and 225 pedestrians are randomly distributed in the room. As can be seen from Fig.
In the following simulation cases (Fig.
In this paper, we developed a game theory-based pedestrian exit choice model for pedestrian evacuation by using the cellular automata simulation framework. On the assumption that all the pedestrians are rational, the estimated evacuation time of a pedestrian through a specified exit is composed of estimated moving time and estimated queuing time towards it. Distance to the exit is used to obtain the expected moving time, and the number of pedestrians within the visual field of a pedestrian and exit flow volume are utilized in the calculation of expected queuing time. To reflect the hesitation phenomenon in evacuation, the parameter of choice firmness is introduced into the game theory model to better represent pedestrian’s exit choice strategy. By designing a scenario of a rectangular room with two exits, a number of simulation cases are conducted. And influences of certain factors on pedestrian evacuation performance are investigated and analyzed, including visual radius and choice firmness level of the pedestrian, initial crowd distributions in the room, exit layout as well as exit width. The simulation results are indicated below. i) Choice firmness level reflects the pedestrian’s persistence in choosing his/her initial exit. Obviously, frequent change of target exit leads to unnatural pedestrian behavior such as wandering between different exits, which is adverse to evacuation. However, it could be avoided when choice firmness level is set to be a relatively large value in the model, such as α = 0.2. ii) Compared with random distribution, pedestrians clustered together in a corner of the room leads to high crowd density and imbalanced use of exits, which is not beneficial to evacuation. iii) For visual radius, the longer the radius, the earlier the pedestrian can determine his/her final exit. Specifically, when R = 0 m, the distance to the exit becomes the only factor in making exit choice, therefore the exit far from the crowd is less used during evacuation. iv) In terms of exit layout, when pedestrian wandering behavior is not considered, the simulation time of the case where exits are on the counter sides of the room is minimum, followed by the case where exits are on the same side, and pedestrians in the case where exits are on the adjacent sides experience the longest evacuation. v) Generally, the exit with a larger width leads to a quicker evacuation, and the effect of exit width on evacuation time is not linear. To sum up, the proposed model is sound in theory and credible in simulation performances. The sensitivity analyses of choice firmness level and visual radius of this paper may benefit the formulation of behavioral rules in other pedestrian simulation models. Controlled experiments on pedestrian exit choice in a room with two exits are conducted, and detailed moving trajectories are obtained and analyzed, which is conducive to further investigating the choice firmness level in future work.
[1] | |
[2] | |
[3] | |
[4] | |
[5] | |
[6] | |
[7] | |
[8] | |
[9] | |
[10] | |
[11] | |
[12] | |
[13] | |
[14] | |
[15] | |
[16] | |
[17] | |
[18] | |
[19] | |
[20] | |
[21] | |
[22] | |
[23] | |
[24] | |
[25] | |
[26] | |
[27] | |
[28] | |
[29] | |
[30] | |
[31] | |
[32] |