Multi-level personnel evacuation behavior modeling method in complex building scene

Abstract

The invention discloses a multi-level personnel evacuation behavior modeling method in a complex building scene. The method comprises the following steps: (1) constructing ''individual-group-crowd-authority'' multi-level agent attributes; (2) formulating an intelligent agent movement rule in four aspects of step length, direction, step time and position updating; (3) establishing an intelligent agent behavior of ''individual-group-crowd-authority'' multi-level interaction; and (4) forming a'perception-decision-execution 'behavior mechanism of the intelligent agent. Wherein the multi-level interaction behaviors of the intelligent agents comprise an interaction behavior between the intelligent agents and a space environment, an interaction behavior of the intelligent agents using experience knowledge, an interaction behavior between the intelligent agents in a group, an interaction behavior between unfamiliar intelligent agents in a crowd, and an interaction behavior between an evacuation intelligent agent and an authoritative intelligent agent. According to the method disclosed by the invention, the personnel evacuation behavior under the emergency condition of the complex building can be highly restored, and technical support is provided for emergency management.

Description

Technical Field

[0001] The present invention belongs to the field of personnel evacuation and relates to a multi-level personnel evacuation behavior modeling method in a complex building scene. Background Art

[0002] Large public buildings are densely populated and have complex building layouts. Once chemical leaks, fires, explosions and other emergencies occur, the safety of public life and property will face great risks. In this context, the development of key emergency evacuation technologies is an important basis for improving public safety protection capabilities.

[0003] The performance of building evacuation is highly dependent on the evacuation behavior of personnel, which is difficult to evaluate and quantitatively describe. The International Society of Fire Engineers (SFPE) pointed out that the evacuation design in traditional building codes and standards assumes that personnel evacuate in a specific way, which is insufficient to describe personnel decision-making and behavior in dynamic and uncertain environments. In recent years, industry insiders have turned to performance-based analysis based on computer modeling and simulation for the evacuation design evaluation of buildings, especially large-scale infrastructure. However, there is still a certain gap between the evacuation behavior of personnel simulated in existing emergency evacuation simulation models and the actual situation.

[0004] Therefore, how to establish a simulation model that can highly restore the decision-making mechanism and behavior patterns of personnel during the evacuation process is an urgent problem to be solved at this stage. Summary of the invention

[0005] In view of the shortcomings of the existing technology, the present invention proposes a multi-level personnel evacuation behavior modeling method in complex building scenes to provide decision support for building design and emergency evacuation management.

[0006] To achieve the above object, the technical solution of the present invention is as follows: a multi-level personnel evacuation behavior modeling method in a complex building scene is provided, comprising the following steps:

[0007] Construct the agent attribute vector A i , including sub-vectors representing the following attributes respectively: individual attributes, group attributes, crowd attributes, and authority attributes;

[0008] Based on the concept of steps, formulate the movement rules of the agent and construct the agent position vector POS i 、Direction vector DIR i , step length sl i , step time i ;

[0009] Construct agent behavior vector B i, including component vectors representing the following behaviors respectively: the interaction between individual agents and the spatial environment, the interaction using knowledge and experience of individual agents, the interaction between agents in a group, the interaction between unfamiliar agents in a crowd, and the interaction between evacuation agents and authoritative agents, wherein the evacuation agents include individual agents and group agents; wherein the interaction between evacuation agents and authoritative agents is specifically: agents evacuate through exits or stairs indicated by authoritative agents, authoritative agents guide the flow of people at fixed locations, or patrol all rooms in their assigned areas one by one, scan all areas of each room, and ensure that all agents start to evacuate;

[0010] Construct the action vector ACT of the intelligent agent i (t), the state vector S of the agent i (t) and the state vector E(t) of the environment, forming an intelligent agent behavior mechanism of “perception-decision-execution”.

[0011] Furthermore, the individual attributes include the type of the agent, body size, free movement speed, field of vision, degree of injury, and familiarity with the building layout; the types are divided into: adult male, adult female, child, and elderly; the body size includes the radius r of the agent's trunk t , the radius of the agent's shoulder r s , half of the width of the agent r w ; The injury degree is divided into: low, medium, and high; the free movement speed includes: horizontal free speed, downstairs free speed, and upstairs free speed; the field of vision includes field of vision distance, field of vision shape, and field of vision angle; the familiarity with the building layout is divided into: 0, 1, 2, and 3;

[0012] The group attributes include whether the group leader, intimacy, distance threshold and waiting time threshold;

[0013] Said crowd attributes include stress levels, shared social identities;

[0014] The authoritative attributes include evacuation tasks and guidance strategies.

[0015] Furthermore, the step-based concept is used to formulate the movement rules of the agent, specifically, to construct the agent position vector POS i (t):

[0016] POS i (t+st i )=POS i (t)+sl i (t)×DIR i (t);

[0017] Among them, POSi (t), sl i (t), DIR i (t) represent the position vector, step length and direction vector of agent t at time t, respectively. i represents the step time of agent i;

[0018] Step length sl i (t) is divided into the step length in the planar environment and step length in stair environments Step size in a planar environment for:

[0019]

[0020] in, is the maximum step size of agent i, Among them, FMS i is the free movement speed of the agent, α is the reduction of the step size;

[0021] Step length in stair environment for:

[0022]

[0023] Among them, ss depth is the depth of the stair steps;

[0024] Construct the agent direction vector DIR i (t):

[0025]

[0026] in, is the preferred direction of agent i at time t, which is determined by the behavior that the agent chooses to perform, DIR β is the adjustment amount of the direction vector;

[0027] Set the agent's obstacle avoidance rules: In a planar environment, agent i first checks whether it can move along the preferred direction at time t. Maximum moving step length If taking this step will cause a collision with an obstacle or other agents, the agent will try to reduce the step length or change the direction of movement. In the stair environment, agent i first checks whether it can move directly to the next step at time t. If taking this step will cause a collision with other agents, the agent will wait or detect other areas on the steps.

[0028] Further, the interaction behavior between the individual agent and the spatial environment is specifically that the agent uses the direction information provided by the exit sign to move to the nearest staircase or exit, or evacuates by exploring the spatial layout of the building;

[0029] The interactive behavior of the individual intelligent agent using experiential knowledge is specifically that the intelligent agent selects the nearest known exit or the most familiar exit according to its familiarity with the building layout, and uses known stairs during the evacuation process if it is not initially on the floor where the exit is located; or the intelligent agent evacuates using exits or stairs within the visual range;

[0030] The interaction behaviors between the agents in the group are specifically: group members follow the leader's decision or group members jointly decide which staircase or exit to move to, and group agents determine whether to wait for other members based on the distance threshold and the waiting time threshold;

[0031] The interactive behaviors between the unfamiliar intelligent agents in the crowd specifically include: the intelligent agents' herd behavior, queuing behavior, overtaking behavior, competitive behavior, close-range interactive behavior, and helping behavior.

[0032] Furthermore, the queuing behavior, overtaking behavior, and competitive behavior are specifically as follows: the queuing behavior prioritizes reducing the step length in a flat environment, and prioritizes waiting in place in a stair environment without adjusting the body posture; the overtaking behavior prioritizes changing the moving direction in a flat environment, and prioritizes moving to other areas of the steps in a stair environment without adjusting the body posture; the competitive behavior prioritizes changing the moving direction in a flat environment, and prioritizes moving to other areas of the steps in a stair environment, and adjusting the body posture.

[0033] Furthermore, the authoritative agent inspects all rooms in its assigned area one by one, scans all areas of each room, and ensures that all agents begin to evacuate. Specifically, the authoritative agent arrives at the entrance of a room and confirms whether the positions with integer coordinates in the room are all within the circular field of view of the authoritative agent: if so, the authoritative agent moves to another room until the authoritative agent has visually checked all rooms; if not, the authoritative agent enters the room and confirms whether the positions with integer coordinates in the room are all within the circular field of view of the authoritative agent; if not, the authoritative agent moves to the nearest position with integer coordinates that has not appeared in its circular field of view once, and repeats until the positions with integer coordinates in the room are all within the circular field of view of the authoritative agent.

[0034] Furthermore, the “perception-decision-execution” intelligent agent behavior mechanism is specifically: constructing the action vector ACT of the intelligent agent i (t),

[0035] ACT i (t) = f excecution (f deciSiOn (f perception (S i (t),E(t)));

[0036] Among them, ACT i (t) is the action vector of agent i at time t, indicating the action performed by the agent, S i (t) is the state vector of agent i itself, including attribute vector and position vector, E(t) is the state vector of the environment, f perception 、f decision 、f excecution They are perception function, decision function and execution function;

[0037] In the perception stage, construct the agent perception vector PER i (t):

[0038] PER i (t) = f perception (s i (t), E(t));

[0039] Among them, PER i (t) is the perception vector of agent i at time t, representing the perceived environmental information. The environmental state vector E(t) includes four sub-vectors, which represent the information obtained by the agent from the spatial environment, group agents, stranger agents in the crowd, and authoritative agents respectively;

[0040] In the decision-making stage, construct the agent behavior vector B i (t):

[0041] B i (t) = f decision (S i (t),PER i (t));

[0042] Among them, B i (t) is the behavior vector of agent i at time t, representing the behavior chosen after decision making, f decision is the decision function;

[0043] The behavior vector B in the execution phase i (t) converted to direction vector

[0044]

[0045] Among them, f direction is the mapping function from behavior vector to direction vector;

[0046] Action vector ACT in the execution phase i (t) can be decomposed into moving step length and moving direction:

[0047] ACT i (t)=(sli (t),DIR i (t));

[0048] get:

[0049]

[0050] Furthermore, the decision function is:

[0051]

[0052] in, They represent the decision functions of the individual, group, crowd, and authority of agent i at time t. γ, δ, θ, and μ represent the weights of each level, indicating the degree of influence of each level on the agent's decision.

[0053] Table 1 summarizes six relatively comprehensive modeling methods in the field of building evacuation (MASSEgress, FDS+Evac, ESCAPES, SAFEgress, EgressSFM and IMPACT) and the evacuation behavior factors considered by this method. Compared with the existing methods, the present invention has the following beneficial effects:

[0054] (1) The present invention takes into account the multi-level intelligent agent attributes and behaviors based on "individual-group-crowd-authority", optimizes the personnel decision-making mechanism and behavior pattern, and establishes a more comprehensive modeling method.

[0055] (2) The present invention establishes new behaviors for the authoritative agent. During the evacuation process, the authoritative agent will patrol the building and urge the agents to start evacuation. After completing the evacuation task, it will start its own evacuation.

[0056] (3) The present invention designs the behavior mechanisms of intelligent agents in plane environments and stair environments respectively. The established multi-level intelligent agent behavior is suitable for multi-story buildings with complex layouts. It can explore the impact of the interaction between intelligent agents and multi-story buildings and between intelligent agents within multi-story buildings on evacuation performance.

[0057] Table 1 Evacuation behavior factors considered by existing methods and this method

[0058]

[0059] BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a flow chart of the method of the present invention;

[0061] Figure 2 is a schematic diagram of the body dimensions of the intelligent agent of the method of the present invention;

[0062] Figure 3 Schematic diagram of the field of view of the intelligent agent in the method of the present invention;

[0063] Figure 4 It is a schematic diagram of the planar movement rule of the intelligent body in the method of the present invention;

[0064] Figure 5 It is a schematic diagram of the stair movement rule of the intelligent agent of the method of the present invention;

[0065] Figure 6 is a schematic diagram of the obstacle avoidance rules of the intelligent agent in the method of the present invention;

[0066] Figure 7 It is a schematic diagram of the process flow of the authoritative intelligent agent inspecting the building according to the method of the present invention;

[0067] Figure 8 is a schematic diagram of a four-story building in a case study of the method of the present invention;

[0068] Fig. 9 is a schematic diagram of the evacuation of an intelligent agent in a plane environment and a staircase environment in a case study of the method of the present invention;

[0069] Fig.10 is a schematic diagram of simulation results of a case study of the method of the present invention;

[0070] In the figure, 1 is the radius r of the agent’s trunk t ; 2 is the radius r of the agent's shoulder s , 3 is half of the width of the agent r w , 4 is the agent's field of vision angle; 5 is the agent's field of vision distance; 6 is the agent's preferred direction; 7 is the agent's step length sl; 8 is a schematic agent; 9 is the agent's preferred target area on the stairs; 10 is the stair step; 11 is the agent's starting area; 12 is the agent's target area; 13 is the path from the starting area to the target area; 14 is the stairs; 15 is the exit; 16 is a simulation experiment based on the "individual-crowd" level; 17 is a simulation experiment based on the "individual-group-crowd" level; 18 is a simulation experiment based on the "individual-group-crowd-authority" level. DETAILED DESCRIPTION

[0071] The following is a further detailed description of a multi-level personnel evacuation behavior modeling method in a complex building scene according to the present invention in conjunction with the accompanying drawings. Figure 1 As shown, the following steps are included:

[0072] A. Construct multi-level agent attributes (as shown in Table 2):

[0073] A i =(IA i,GA i ,CA i ,AA i ) (1)

[0074] Among them A i is the attribute vector of agent i, which contains four sub-vectors IA i , GA i , CA i AA i , represent the individual attribute vector, group attribute vector, crowd attribute vector and authority attribute vector respectively;

[0075] Table 2 Agent attributes

[0076]

[0077] B. Formulate the movement rules of the agent. The movement rules are based on the concept of steps. Each agent moves forward step by step, neither moving between grids nor moving continuously:

[0078] POS i (t+st i )=POS i (t)+sl i (t)×DIR i (t) (2)

[0079] Among them, POS i (t), sl i (t), DIR i (t) represent the position vector, step length (the length of each step) and direction vector of agent i at time t, respectively. i represents the step time of agent i (the time it takes to take one step); agent i starts taking a step at time t and at time t+st i After this step, the position changes from POS i (t) Update to POS i (t+st i );

[0080] Figure 4 For the movement of the agent in a planar environment, on the plane, the agent searches for the target location based on the moving direction and step length, and the agent prioritizes the preferred direction;

[0081] Figure 5 The agent moves in a stair environment. On the stairs, the step length is limited by the depth of the stair steps, and the agent will move to the area directly in front of it first.

[0082] Figure 6The rules for the agent to determine whether there is a collision. If the agent wants to move one step from the starting area to the target area, it needs to determine whether the target area and path are occupied by other agents or obstacles;

[0083] C. Establish multi-level agent behavior (Table 4):

[0084] B i =(IA_SE i ,IA_KE i ,CA_GA i ,CA_CA i ,EA_AA i ) (3)

[0085] Among them, B i Represents the behavior vector of agent i, which contains five sub-vectors IA_SE i IA_KE i ,GA_GA i 、CA_CA i ,EA_AA i , respectively representing: the interaction between individual agents and the spatial environment, the interaction using knowledge and experience of individual agents, the interaction between agents in a group, the interaction between unfamiliar agents in a crowd, and the interaction between evacuation agents (individual or group agents) and authoritative agents;

[0086] D. Forming the “perception-decision-execution” intelligent agent behavior mechanism:

[0087] ACT i (t) = f excecution (f decision (f perceptipn (S i (t),E(t))) (4)

[0088] Among them, ACT i (t) is the action vector of agent i at time t, indicating the action performed by the agent, S i (t) is the state vector of agent i itself, including its attributes and current position, E(t) is the state vector of the environment, and f perception 、f decision 、f excecution They are perception function, decision function and execution function;

[0089] The individual attributes in step A are:

[0090] I A i =(AT i ,BS i ,FMS i ,FVi ,IL i ,FL i ) (5)

[0091] Among them, AT i ,BS i 、FMS i 、FVi i IL i , FL i Represents the type (gender and age), body size, free movement speed, field of vision, degree of injury and familiarity with the building layout of agent i. The free movement speed has three components: horizontal free speed Free speed downstairs and free speed upstairs

[0092] The group attributes in step A are:

[0093] GA m =(GL m ,IL m ,DT m ,WTT m ) (6)

[0094] Among them, GL m IL m , DT m 、WTT m They represent the leader situation, intimacy, distance threshold and waiting time threshold of group m respectively;

[0095] The crowd attributes in step A are:

[0096] CA i =(SL i ,SSI i ) (7)

[0097] Among them, SL i 、SSI i They represent the stress level and shared social identity of agent i respectively. The shared social identity represents a sense of social identity generated by facing an emergency together.

[0098] The authoritative attributes in step A are:

[0099] AA i =(ET i ,GS i ) (8)

[0100] Among them, E.T. i GS i They represent the evacuation task and guidance strategy of agent i respectively.

[0101] The step size in the planar environment of step B for:

[0102]

[0103] in, is the maximum step length of agent i, which depends on the free movement speed FMS of the agent i According to the study [Natural discretization of pedestrian movement in continuous space], the calculation formula is:

[0104]

[0105] α is the amount by which the step size is reduced; the step size in the staircase environment is Affected by the stair step depth ss dept The impact of

[0106]

[0107] The direction vector DIR in step B i (t) is:

[0108]

[0109] in, is the preferred direction of agent i at time t, depending on the action the agent chooses to perform, DIR β is the adjustment amount of the direction vector;

[0110] The obstacle avoidance rule of the agent in step B is: In a planar environment, agent i first checks whether it can move along the preferred direction at time t. Maximum moving step length If taking this step will cause a collision with an obstacle or other agents, the agent will try to reduce the step length or change the direction of movement. In the stair environment, agent i first checks whether it can move directly to the next step at time t. If taking this step will cause a collision with other agents, the agent will wait or detect other areas on the steps.

[0111] In step C, based on the interaction behavior between the individual intelligent agent and the spatial environment (IA_SE i ), the agent can use the directional information provided by the exit signs to move to the nearest staircase or exit, or evacuate by exploring the spatial layout of the building;

[0112] The interactive behavior (IA_KE) based on the knowledge experience of individual agents in step Ci ), the agent selects the nearest known exit or the most familiar exit based on its familiarity with the building layout. If it is not initially on the floor where the exit is located, it needs to use the known stairs during the evacuation process; the agent can evacuate using the exits or stairs within the visual range;

[0113] In step C, based on the interaction behavior (CA_GA) between the agents in the group i ), group members follow the leader's decision, or group members jointly decide which staircase or exit to move to, and the group agent determines whether to wait for other members based on the distance threshold and waiting time threshold;

[0114] In step C, based on the interaction behavior (CA_CA) between stranger agents in the crowd i ), the agent may follow the majority of agents to evacuate, that is, follow the crowd behavior; the agent can perform queuing behavior, overtaking behavior or competitive behavior; in a planar environment, when the agent cannot evacuate along the preferred direction in order to avoid obstacles Maximum moving step length When , the probability of the agent changing the moving direction and reducing the step size first is: CMD and p RSL (p RSL +p CMD =1); In the stair environment, when the agent cannot move directly to the next step to avoid obstacles, the probability of the agent choosing to move to other areas on the steps and waiting in place is: CO and p w (p COR +p w =1); Table 3 shows the values ​​and behavior descriptions of the above probabilities when the agents perform queuing, overtaking and competing behaviors in a plane environment and a stair environment; compared with the agents that perform overtaking behaviors, the agents that perform competing behaviors will adjust their body postures to force their way through a smaller area; close-range interactions between agents may lead to negative consequences such as falling and trampling; agents may perform helping behaviors to help the fallen agents.

[0115] Table 3 Queuing, overtaking and competitive behavior

[0116]

[0117] In step C, based on the interaction between the evacuation agent and the authority agent (EA_AA i ), the agents evacuate through the exits or stairs indicated by the authoritative agent; the authoritative agent can guide the flow of people at a fixed location; the authoritative agent can patrol the building to urge the agents to start evacuating, that is, the authoritative agent patrols all rooms in its assigned area one by one, scans all areas of each room, and ensures that all agents start evacuating. Figure 7 As shown, the authoritative agent arrives at the entrance of a room and confirms whether the positions with integer coordinates in the room are all within the circular field of view of the authoritative agent: if so, the authoritative agent moves to another room until the authoritative agent has visually checked all rooms; if not, the authoritative agent enters the room and confirms whether the positions with integer coordinates in the room are all within the circular field of view of the authoritative agent, if still not, the authoritative agent moves to the nearest position with integer coordinates that has not appeared in its circular field of view once, and repeats until the positions with integer coordinates in the room are all within the circular field of view of the authoritative agent.

[0118] The perception function of the perception stage in step D is:

[0119] PER i (t) = f perception (S i (t),E(t)) (13)

[0120] Among them, PER i (t) is the perception vector of agent i at time t, representing the perceived environmental information. The environmental state vector E(t) is:

[0121] E(t)=(E SE (t),E GA (t),E CA (t),E AA (t)) (14)

[0122] Among them, E SE (t), E GA (t), E CA (t), E AA (t) represents the information obtained by the agent from the spatial environment, group agents, stranger agents in the crowd, and authoritative agents;

[0123] The decision-making stage in step D is that the agent selects its behavior based on the perceived information. The decision function can be expressed as:

[0124] B i (t) = f decision (S i (t),PER i (t)) (15)

[0125] Among them, B i (t) is the behavior vector of agent i at time t, representing the behavior selected after decision making (taken from the behavior in formula (3)), and the decision function f decision Can be decomposed into:

[0126]

[0127] in, They represent the decision functions of the individual, group, crowd, and authority of agent i at time t. γ, δ, θ, and μ represent the weights of each level, indicating the degree of influence of each level on the decision of the agent.

[0128] In the execution phase of step D, the behavior vector obtained in the decision phase is converted into a direction vector:

[0129]

[0130] Among them, f direction is the mapping function from behavior vector to direction vector;

[0131] Action vector ACT in the execution phase i (t) can be decomposed into moving step length and moving direction:

[0132] ACT i (t)=(sl i (t),DIR i (t)) (18)

[0133] Combining formulas (4), (9), (12), (17) and (18), we obtain:

[0134]

[0135] Formula (19) represents the mapping process from selected behavior to taken action.

[0136] Figure 8 The four-story building used for the case study, Fig. 9 Schematic diagram of the evacuation of the agent in the exit area and on the stairs. Fig.10 The simulation results record the evacuation time of the agent based on different levels of behavior. The results show that:

[0137] (1) When more people share a social identity, resulting in more people queuing, the evacuation time of the first 90% of people who successfully escape can be increased by up to 19%;

[0138] (2) When more people share a social identity and more people are lined up, falls occur more frequently, and the number of falls can increase by up to 107%. In addition, when more than 20% of the people share a social identity and help the fallen person, all the fallen people can be successfully evacuated;

[0139] (3) When the group intimacy is higher, the total evacuation time increases by 26% in 75% of cases. However, members of the low-intimacy group are more likely to get lost. The lost members choose the familiar and farther exit, resulting in longer (up to 33%) evacuation time;

[0140] (4) When an authority inspects the building, the evacuation time can be shortened by 58%, and when the authority's initial position is random, the evacuation time can be shortened by 19%.

[0141] Table 4 Agent behavior

[0142]

[0143] The contents described in the implementation cases of this specification are merely an enumeration of the implementation forms of the inventive concept. The protection scope of the present invention should not be deemed to be limited to the specific forms described in the implementation cases. The protection scope of the present invention also extends to the equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A multi-level personnel evacuation behavior modeling method in a complex building scene, characterized in that: The following steps are involved: Construct the agent attribute vector A i , including sub-vectors representing the following attributes respectively: individual attributes, group attributes, crowd attributes, and authority attributes; Based on the concept of steps, formulate the movement rules of the agent and construct the agent position vector POS i 、Direction vector DIR i , step length sl i , step time i ; Construct agent behavior vector B i , including component vectors representing the following behaviors respectively: the interaction between individual agents and the spatial environment, the interaction using knowledge and experience of individual agents, the interaction between agents in a group, the interaction between unfamiliar agents in a crowd, and the interaction between evacuation agents and authoritative agents, wherein the evacuation agents include individual agents and group agents; wherein the interaction between evacuation agents and authoritative agents is specifically: agents evacuate through exits or stairs indicated by authoritative agents, authoritative agents guide the flow of people at fixed locations, or patrol all rooms in their assigned areas one by one, scan all areas of each room, and ensure that all agents start to evacuate; Construct the action vector ACT of the intelligent agent i (t), the state vector B of the agent i (t) and the state vector E(t) of the environment, forming an intelligent agent behavior mechanism of "perception-decision-execution".

2. The method according to claim 1, characterized in that The individual attributes include the type of the intelligent agent, body size, free movement speed, field of vision, degree of injury, and familiarity with the building layout; the types are divided into: adult male, adult female, child, and elderly; the body size includes the radius r of the intelligent agent's trunk t , the radius of the agent's shoulder r s , half of the width of the agent r w ; The injury levels are classified into: low, medium, and high; The free movement speed includes: horizontal free speed, downstairs free speed and upstairs free speed; the field of view includes field of view distance, field of view shape and field of view angle; the familiarity with the building layout is divided into: 0, 1, 2, 3; The group attributes include whether the group leader, intimacy, distance threshold and waiting time threshold; Said crowd attributes include stress levels, shared social identities; The authoritative attributes include evacuation tasks and guidance strategies.

3. The method according to claim 1, characterized in that Based on the concept of step, the movement rules of the intelligent agent are formulated, specifically, the position vector POS of the intelligent agent is constructed i (t): POS i (t+st i )=POS i (t)+sl i (t)×DIR i (t); Among them, POS i (t), sl i (t), DIR i (t) represent the position vector, step length and direction vector of agent i at time t, respectively. i represents the step time of agent i; Step length sl i (t) is divided into the step length in the planar environment and step length in stair environments Step size in a planar environment for: in, is the maximum step size of agent i, Among them, FMS i is the free movement speed of the agent, α is the reduction of the step size; Step length in stair environment for: Among them, SS depth is the depth of the stair steps; Construct the agent direction vector DIR i (t): in, is the preferred direction of agent i at time t, which is determined by the behavior that the agent chooses to perform, DIR β is the adjustment amount of the direction vector; Set the agent's obstacle avoidance rules: In a planar environment, agent i first checks whether it can move along the preferred direction at time t. Maximum moving step length If taking this step will cause a collision with an obstacle or other agents, the agent will try to reduce the step length or change the direction of movement. In the stair environment, agent i first checks whether it can move directly to the next step at time t. If taking this step will cause a collision with other agents, the agent will wait or detect other areas on the steps.

4. The method according to claim 1, characterized in that The interaction between the individual agent and the spatial environment, specifically, the agent uses the directional information provided by the exit sign to move to the nearest staircase or exit, or evacuates by exploring the spatial layout of the building; The interactive behavior of the individual intelligent agent using experiential knowledge is specifically that the intelligent agent selects the nearest known exit or the most familiar exit according to its familiarity with the building layout, and uses known stairs during the evacuation process if it is not initially on the floor where the exit is located; or the intelligent agent evacuates using exits or stairs within the visual range; The interaction behaviors between the agents in the group are specifically: group members follow the leader's decision or group members jointly decide which staircase or exit to move to, and group agents determine whether to wait for other members based on the distance threshold and the waiting time threshold; The interactive behaviors between the unfamiliar intelligent agents in the crowd specifically include: the intelligent agents' herd behavior, queuing behavior, overtaking behavior, competitive behavior, close-range interactive behavior, and helping behavior.

5. The method according to claim 4, characterized in that The queuing behavior, overtaking behavior, and competitive behavior are specifically as follows: the queuing behavior prioritizes reducing the step length in a flat environment, and prioritizes waiting in place in a stair environment without adjusting the body posture; the overtaking behavior prioritizes changing the moving direction in a flat environment, and prioritizes moving to other areas of the steps in a stair environment without adjusting the body posture; the competitive behavior prioritizes changing the moving direction in a flat environment, and prioritizes moving to other areas of the steps in a stair environment, and adjusting the body posture.

6. The method according to claim 1, characterized in that The authoritative agent inspects all rooms in its assigned area one by one, scans all areas of each room, and ensures that all agents begin to evacuate. Specifically, the authoritative agent arrives at the entrance of a room and confirms whether the positions with integer coordinates in the room are all within the circular field of view of the authoritative agent: if so, the authoritative agent moves to another room until the authoritative agent has visually checked all rooms; if not, the authoritative agent enters the room and confirms whether the positions with integer coordinates in the room are all within the circular field of view of the authoritative agent; if not, the authoritative agent moves to the nearest position with integer coordinates that has not appeared in its circular field of view once, and repeats until the positions with integer coordinates in the room are all within the circular field of view of the authoritative agent.

7. The method according to claim 1, characterized in that The "perception-decision-execution" intelligent agent behavior mechanism is specifically: constructing the action vector ACT of the intelligent agent i (t), ACT i (t)=f excecution (f decision (f perception (S i (t),E(t))); Among them, ACT i (t) is the action vector of agent i at time t, indicating the action performed by the agent, S i (t) is the state vector of agent i itself, including attribute vector and position vector, E(t) is the state vector of the environment, f perception 、f decision 、f excecution They are perception function, decision function and execution function; In the perception stage, construct the agent perception vector PER i (t): PER i (t)=f perception (S i (t),E(t)); Among them, PER i (t) is the perception vector of agent i at time t, representing the perceived environmental information. The environmental state vector E(t) includes four sub-vectors, which represent the information obtained by the agent from the spatial environment, group agents, stranger agents in the crowd, and authoritative agents respectively; In the decision-making stage, construct the agent behavior vector B i (t): B i (t)=f decision (S i (t),PER i (t)); Among them, B i (t) is the behavior vector of agent i at time t, representing the behavior chosen after decision making, f decision is the decision function; The behavior vector B in the execution phase i (t) converted to direction vector Among them, f direction is the mapping function from behavior vector to direction vector; Action vector ACT in the execution phase i (t) can be decomposed into moving step length and moving direction: ACT i (t)=(sl i (t),DIR i (t)); get:

8. The method according to claim 7, characterized in that The decision function is: in, They represent the decision functions of the individual, group, crowd, and authority of agent i at time t. γ, δ, θ, and μ represent the weights of each level, indicating the degree of influence of each level on the agent's decision.

Images