A subway fire scene robot-assisted pedestrian flow evacuation simulation method

Abstract

The application is suitable for the field of crowd evacuation simulation technology, and provides a subway fire scene robot-assisted pedestrian flow evacuation simulation method, which is applied to a subway fire scene, and the method comprises the following steps: obtaining a first position of a fire in a subway station and second positions of all pedestrians in a dangerous area of the fire; determining, according to the first position and the second positions, third positions of guide robots for evacuating and shunting the pedestrians in the dangerous area, and controlling the guide robots to move to the third positions; inputting the first position, the second positions and the third positions into an improved social force model for processing to obtain accelerations of each pedestrian in the dangerous area; determining, according to the accelerations of the pedestrians, positions of the pedestrians at a next moment, and guiding the pedestrians to move to the positions; and detecting whether there is an unevacuated person in the dangerous area, if there is, continuing to evacuate the unevacuated person, otherwise ending the simulation. The application has high intelligence and automation levels, and can quickly and effectively evacuate pedestrians.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of crowd evacuation simulation, and particularly relates to a robot-assisted pedestrian flow evacuation simulation method for a subway fire scene. BACKGROUND

[0002] In a busy and complex public transportation facility such as a subway passenger layer, especially in an emergency such as a fire, rapid and orderly evacuation is crucial to protect the safety of passengers. The traditional evacuation strategy often fails to cope with the panic and chaos caused by fire, therefore, it is of great significance to use modern information technology for simulation and use the simulation results to develop more reasonable, scientific and efficient emergency evacuation plans to improve the response capability of the subway station to emergencies such as fire.

[0003] Although the existing simulation methods can provide directions for pedestrian evacuation to some extent, the intelligence level and automation level of these methods are not high enough, resulting in difficulty in quickly and effectively evacuating pedestrians in a subway fire scene. SUMMARY

[0004] The application provides a robot-assisted pedestrian flow evacuation simulation method for a subway fire scene, which can solve the problem of difficulty in quickly and effectively evacuating pedestrians in a subway fire scene.

[0005] The application provides a robot-assisted pedestrian flow evacuation simulation method for a subway fire scene, which can solve the problem of difficulty in quickly and effectively evacuating pedestrians in a subway fire scene.

[0006] Obtain a first position of a fire in a subway station, and a second position of all pedestrians in a dangerous area of the fire;

[0007] According to the first position and the second position, determine a third position of a guide robot for evacuating and diverting the pedestrians in the dangerous area, and control the guide robot to move to the third position;

[0008] Input the first position, the second position and the third position into an improved social force model for processing to obtain an acceleration of each pedestrian in the dangerous area from the dangerous area to a safe area; the improved social force model is a social force model integrated with fire scene characteristic parameters, the fire scene characteristic parameters including a driving force, an interaction force between pedestrians, an interaction force between pedestrians and obstacles, and an interaction force between pedestrians and fire; the driving force represents a difference between a current speed of the pedestrian and an expected speed thereof, the interaction force between pedestrians represents a mutual interaction force generated when the pedestrians physically contact each other, the interaction force between pedestrians and obstacles represents an interaction force between the pedestrians and the obstacles, and the interaction force between pedestrians and fire represents an interaction force of the pedestrians away from the fire;

[0009] For each pedestrian in the dangerous area, the position of the pedestrian at the next moment is determined according to the acceleration of the pedestrian from the dangerous area to the safe area, and the pedestrian is guided to move to the position;

[0010] When the next moment arrives, it is detected whether there is an unevacuated person in the dangerous area of the fire, if there is, the step of obtaining the first position of the fire in the subway station, the second position of all pedestrians in the dangerous area of the fire is returned, otherwise the simulation is ended.

[0011] Optionally, according to the first position and the second position, a third position of a guide robot for evacuation and shunting of pedestrians in the dangerous area is determined, comprising:

[0012] The first position and the second position are input into a learned deep neural network for processing to obtain the third position of the guide robot for evacuation and shunting of pedestrians in the dangerous area.

[0013] Optionally, the improved social force model is:

[0014]

[0015] Wherein, m i represents the mass of the ith pedestrian, represents the acceleration a t of the ith pedestrian from the dangerous area to the safe area, i represents the driving force of the ith pedestrian, f ij represents the force between the ith pedestrian and the jth pedestrian, f iw represents the force between the ith pedestrian and the obstacle, f if represents the force between the ith pedestrian and the fire.

[0016] Optionally, Wherein, represents the expected speed of the ith pedestrian, represents the expected direction of motion of the ith pedestrian at time t, v i (t) represents the true speed of the ith pedestrian at time t, τ i represents a time interval;

[0017]

[0018] Wherein, v min represents a preset minimum speed, v max represents a preset maximum speed, α represents a preset parameter, d fire represents the distance between the ith pedestrian and the fire, rfire Indicates the radius of the fire.

[0019] Optional, f ij =A i exp[(r ij -d ij ) / B i ]n ij +kg(r ij -d ij )n ij , where A i =2000N, r ij Let d represent the sum of the radii of the i-th pedestrian and the j-th pedestrian. ij B represents the distance between the i-th pedestrian and the j-th pedestrian. i =0.08m, n ij Let k represent a unit vector, k = 1.2 × 10⁻⁶. 5 kg·s -2 If the i-th pedestrian and the j-th pedestrian do not touch each other, then the function g(x) takes the value of zero, where x represents the variable of the function g(x); otherwise, g(x) equals x.

[0020] Optional, f iw =A i exp[[(r i -d iw ) / B i ]]n iw +kg(r i -d iw )n iw , where r i Let d represent the radius of the i-th pedestrian. iw Let n represent the distance between the i-th pedestrian and the obstacle w. iw Let g(x) represent a unit vector. If the i-th pedestrian does not touch the obstacle w, then the function g(x) takes the value of zero, where x represents the variable of the function g(x). Otherwise, g(x) equals x.

[0021] Optional, f if =λexp(-d fire (1+I)n if Where λ = 480N, n if Represents a unit vector.

[0022] Optionally, the pedestrian's position at the next moment can be determined based on the pedestrian's acceleration from the danger zone to the safe zone, including:

[0023] The position of the i-th pedestrian at time t+1 is obtained using the position update formula; the position update formula is:

[0024]

[0025] wherein, x t+1 represents the position of the i-th pedestrian at t+1 moment, x t represents the position of the i-th pedestrian at t moment, t1 is 0.005 seconds, t moment is the current moment, and t+1 moment is the next moment of the current moment.

[0026] The above-mentioned scheme of the present application has the following beneficial effects:

[0027] In some embodiments of the present application, the position of the guide robot for diverting the pedestrians for evacuation is determined based on the position of the fire in the subway station and the position of the pedestrians in the fire danger area, the guide robot is controlled to move to the position, the acceleration of each pedestrian from the danger area to the safe area is determined based on the improved social force model, the position of the pedestrian at the next moment is determined based on the acceleration, the pedestrian is guided to move to the position, and when it is judged that there are still people in the danger area at the next moment, the above-mentioned process is repeated to evacuate the pedestrians, so as to evacuate all the pedestrians in the danger area to the safe area. Wherein, the guide robot has high intelligentization and automation level, it can analyze the specific position of the fire, the spreading situation, the flow state of the pedestrians and the availability of the evacuation channel in real time, and quickly move to the specific area near the fire, reduce the crowd of the pedestrians, guide the pedestrians to evacuate along the safest evacuation path, so as to evacuate the pedestrians quickly and effectively.

[0028] Other beneficial effects of the present application will be described in detail in the subsequent specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0030] Figure 1 The flow chart of the subway fire scene robot-assisted pedestrian flow evacuation simulation method provided by an embodiment of the present application;

[0031] Figure 2 The stair area in the subway boarding layer in an example of the present application;

[0032] Figure 3 The pedestrian motion trajectory graph in the no-machine scene in an example of the present application;

[0033] Figure 4 The time heat map in the static robot scene in an example of the present application;

[0034] Figure 5 A plot of the robot motion trajectory for the DRL control robot scenario in one example of the present application;

[0035] Figure 6 A plot of the robot motion trajectory fitting for the robot in one example of the present application;

[0036] Figure 7 A plot of the instantaneous density for Case 1 in one example of the present application;

[0037] Figure 8 A plot of the instantaneous density for Case 2 in one example of the present application;

[0038] Figure 9 A plot of the average distance of pedestrians to fire in one example of the present application;

[0039] Figure 10 A plot of the average distance of the last 10 people to fire in one example of the present application. DETAILED DESCRIPTION

[0040] In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular architectures, technologies, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

[0041] It is to be understood that the terminology “includes,” “has,” “holds,” “contains,” “comprises,” “comprised of,” “comprising,” or variants thereof does not exclude the presence of additional

[0042] It is to be understood that the terminology “includes,” “has,” “holds,” “contains,” “comprises,” “comprised of,” “comprising,” or variants thereof does not exclude the presence of additional

[0043] As used in the present specification and claims, the term “if’ can be interpreted as meaning “when” or “once” or “in response to a determination” or “in response to a detection” depending on the context. Similarly, the phrase “if it is determined” or “if [a described condition or event] is detected” can be interpreted as meaning “once it is determined” or “in response to the determination” or “once [a described condition or event] is detected” or “in response to the detection” depending on the context.

[0044] In addition, in the description of the present application and the appended claims, the terms "first", "second", "third", etc. are used only to distinguish descriptions and cannot be understood as indicating or implying relative importance.

[0045] Reference to "one embodiment" or "some embodiments" or "one implementation" or "some implementations" or "one example" or "some examples" described in the present application means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present application. Therefore, the phrases "in one embodiment", "in some embodiments", "in other some embodiments", "in yet some embodiments", etc. appearing in different places in the specification are not necessarily all referring to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized. The terms "include", "contain", "have" and their variants mean "including but not limited to", unless otherwise specifically emphasized.

[0046] In view of the problem that it is difficult to quickly and effectively evacuate pedestrians in the current subway fire scene, the present application provides a subway fire scene robot-assisted pedestrian flow evacuation simulation method, which is applied to the subway fire scene. The method determines the position of the guide robot for over-evacuation and diversion of pedestrians based on the position of the fire in the subway station and the position of the pedestrians in the fire danger area, and controls the guide robot to move to the position. At the same time, the acceleration of each pedestrian from the danger area to the safe area is determined based on the improved social force model, the position of the pedestrian at the next time is determined based on the acceleration, the pedestrian is guided to move to the position, and when it is judged that there are still personnel in the danger area at the next time, the above process is repeated to evacuate the pedestrians. The method evacuates all pedestrians in the danger area to the safe area. The guide robot has a high level of intelligence and automation, which can analyze the specific position of the fire, the spread of the fire, the flow state of the pedestrians and the availability of the evacuation channel in real time, and quickly move to a specific area near the fire. The guide robot reduces the crowding of pedestrians and guides the pedestrians to evacuate along the safest evacuation path, thereby quickly and effectively evacuating the pedestrians.

[0047] The subway fire scene robot-assisted pedestrian flow evacuation simulation method provided by the present application will be exemplarily described in combination with specific embodiments.

[0048] As shown in Figure 1 The subway fire scene robot-assisted pedestrian flow evacuation simulation method provided by the present application includes the following steps:

[0049] Step 11, obtaining a first position of a fire in a subway station and a second position of all pedestrians in a danger area of the fire.

[0050] The first position of the fire refers to the regional position of the fire, for example, a certain area in the passenger layer of the subway station. The dangerous area can be understood as an area with the first position as the center and a preset distance as the radius. The second position refers to the position of the pedestrian in the dangerous area.

[0051] Step 12, determining the third position of the guide robot for evacuation and shunting of the pedestrian in the dangerous area according to the first position and the second position, and controlling the guide robot to move to the third position.

[0052] The guide robot can autonomously perceive the environment, analyze the specific position of the fire, the spread of the fire (i.e., the identification of the dangerous area), the flow state of the pedestrian, and the availability of the evacuation passage in real time. It can be understood that these functions can be realized based on the sensors of the guide robot itself (such as smoke sensors, temperature sensors, acceleration sensors, etc.), internal integrated positioning algorithms (which can be common positioning algorithms), etc.

[0053] That is, in some embodiments of the present application, the first position of the fire and the second position of all pedestrians can be obtained by the guide robot.

[0054] After determining the first position of the fire and the second position of the pedestrian, the processor of the guide robot can determine the third position of the guide robot using a deep neural network, and control the guide robot to move to the third position (which is usually a key position for evacuating pedestrians, such as a stairwell).

[0055] For example, the guide robot can be in the form of an obstacle, thereby shunting the pedestrian, reducing the crowd of pedestrians, and guiding the pedestrian to evacuate.

[0056] In some embodiments of the present application, the first position and the second position can be input into a learned deep neural network for processing to obtain the third position of the guide robot for evacuation and shunting of the pedestrian in the dangerous area.

[0057] The deep neural network can be a common neural network based on deep reinforcement learning (DRL) technology. The learning refers to training the deep neural network using training samples so that it can output the position of the guide robot based on the position of the fire and the position of the pedestrian. It can be understood that since the structure and training method of the deep neural network are not improved in the present application, the structure and training process of the deep neural network will not be described in detail here. For example, the deep neural network can be a policy network.

[0058] Step 13, inputting the first position, the second position, and the third position into an improved social force model for processing to obtain the acceleration of each pedestrian in the dangerous area from the dangerous area to the safe area.

[0059] The improved social force model is a social force model integrated with fire scene characteristic parameters, the fire scene characteristic parameters including a driving force, an interaction force between pedestrians, an interaction force between a pedestrian and an obstacle, and an interaction force between a pedestrian and a fire; the driving force represents a difference between a current speed of the pedestrian and a desired speed of the pedestrian, the interaction force between pedestrians represents an interaction force generated when physical contact occurs between the pedestrians, the interaction force between the pedestrian and the obstacle represents an interaction force between the pedestrian and the obstacle, and the interaction force between the pedestrian and the fire represents an interaction force of the pedestrian away from the fire.

[0060] The improved social force model is as follows:

[0061]

[0062] wherein, m i represents a mass of the ith pedestrian, which can be pre-set according to a situation, represents an acceleration a t of the ith pedestrian from a dangerous area to a safe area, i represents the driving force of the ith pedestrian, f ij represents the interaction force between the ith pedestrian and the jth pedestrian, f iw represents the interaction force between the ith pedestrian and the obstacle, f if represents the interaction force between the ith pedestrian and the fire, i = 1, …, I, and I represents a total number of pedestrians.

[0063] wherein, represents a desired speed of the ith pedestrian, represents an expected moving direction of the ith pedestrian at a time t, v i represents a real speed of the ith pedestrian at the time t (the real speed is a quantity that needs to be solved by the social force model, i.e., v i in the social force model), τ i represents a time interval (taking a value of 0.5 seconds).

[0064]

[0065] wherein, v min represents a preset minimum speed, which can be set according to an actual situation, v max represents a preset maximum speed, which can be set according to the actual situation, α represents a preset parameter (taking a value of 2), d fire represents a distance between the ith pedestrian and the fire, r fire represents a radius of the fire.

[0066] wherein, fij = A i exp[(r ij -d ij ) / B i ]n ij + kg(r ij -d ij )n ij , A i = 2000 N, r ij denotes the sum of the radii of the ith pedestrian and the jth pedestrian, d ij denotes the distance between the ith pedestrian and the jth pedestrian, B i = 0.08 m, n ij denotes a unit vector, k = 1.2 x 10 5 kg s -2 , specifically, A i exp[(r ij -d ij ) / B i ]n ij denotes the psychological repulsion between the ith pedestrian and the jth pedestrian, kg(r ij -d ij )n ij denotes the physical repulsion between the ith pedestrian and the jth pedestrian, if the ith pedestrian and the jth pedestrian are not in contact with each other, the function g(x) takes a value equal to zero, x denotes a variable of the function g(x), otherwise, g(x) is equal to x (i.e. r ij -d ij ).

[0067] wherein f iw = A i exp[[(r i -d iw ) / B i ]]n iw + kg(r i -d iw )n iw , r i denotes the radius of the ith pedestrian, d iw denotes the distance between the ith pedestrian and the obstacle w, n iw denotes a unit vector, specifically A i exp[[(r i -d iw ) / B i ]]n iw denotes the psychological repulsion between the ith pedestrian and the obstacle w, kg(r i -d iw )n iwrepresents the physical repulsion between the ith pedestrian and the obstacle w, if the ith pedestrian and the obstacle w are not in contact with each other, the function g(x) takes a value equal to zero, x represents the variable of the function g(x), otherwise, g(x) is equal to x (i.e. r i -d iw ). For example, the above obstacle is an obstacle in the subway station, such as a wall.

[0068] wherein f if = λexp(-d fire )(1+I)n if , wherein λ = 480N, n if represents a unit vector.

[0069] Step 14, for each pedestrian in the dangerous area, respectively, according to the acceleration of the pedestrian from the dangerous area to the safe area, the position of the pedestrian at the next moment is determined, and the pedestrian is guided to move to the position.

[0070] In some embodiments of the present application, the position of the ith pedestrian at t+1 moment can be obtained by a position updating formula, which is:

[0071]

[0072] wherein x t+1 represents the position of the ith pedestrian at t+1 moment, x t represents the position of the ith pedestrian at t moment, t1 is 0.005 seconds, t moment is the current moment, and t+1 moment is the next moment of the current moment.

[0073] Step 15, when the next moment arrives, it is detected whether there is an unevacuated person in the dangerous area of the fire, if there is, the step of obtaining the first position of the fire in the subway station and the second position of all pedestrians in the dangerous area of the fire is returned to be executed, otherwise the simulation is ended.

[0074] That is, when the next moment arrives, it is judged whether there is an unevacuated person in the dangerous area, if there is, it is returned to the above step 11 to obtain the first position of the fire in the subway station and the second position of all pedestrians in the dangerous area of the fire, and sequentially execute steps 12 to 15. It is worth mentioning that in this process, the specific position of the fire, the spread, the flow state of the pedestrians and the availability of the evacuation channel are analyzed in real time by guiding the robot, and quickly move to a specific area near the fire, reduce the crowd of pedestrians, guide the pedestrians to evacuate along the safest evacuation path, so as to evacuate the pedestrians quickly and effectively.

[0075] Of course, if there is no unevacuated person in the dangerous area, the simulation is ended. At this time, the parameters of the deep neural network can be updated using the entire evacuation process. Taking the deep neural network as an example, the parameter update of the deep neural network is exemplarily illustrated.

[0076] The parameter update process includes the following steps:

[0077] Exemplarily, the deep neural network can be a policy network and a value function network, and the learning process (i.e., the training process) of the deep neural network includes the following steps:

[0078] (1) Environment interaction and data collection: under the current policy, the action probability distribution output by the policy network is used as a guide to interact with the environment to generate a series of trajectory data. Each trajectory data contains a series of states, actions (i.e., the position of the guided robot), rewards, and next states. The generated trajectory data is stored in an experience replay buffer for subsequent training process;

[0079] (2) Calculate the advantage function: use the value function network to estimate the value function of each state in the experience replay buffer, i.e., predict the expected cumulative reward that can be obtained by following the current policy from the state. According to the value function estimate and the reward in the trajectory data, the advantage function of each state-action pair is calculated;

[0080] (3) Calculate the probability ratio: for each state-action pair in the experience replay buffer, calculate the probability ratio of the new policy and the old policy to produce the action;

[0081] (4) Maximize the clipping objective to update the policy: the clipping objective function is a function of the probability ratio of the new and old policies and the advantage function. Use an optimization algorithm to maximize the clipping objective function to update the parameters of the policy network. This step aims to improve the performance of the policy while maintaining a certain stability;

[0082] (5) Mean squared error regression to update the value function: calculate the mean squared error according to the current value function estimate and the target value function calculated based on the trajectory data. Use gradient descent to optimize the mean squared error to update the parameters of the value function network. This step aims to improve the accuracy of the value function estimate to provide more accurate guidance for policy optimization.

[0083] In summary, the subway fire scene robot-assisted pedestrian flow evacuation simulation method of the present application introduces deep reinforcement learning (DRL) technology, which brings revolutionary changes to the emergency evacuation of subway stations in the event of fire. DRL technology combines the perception ability of deep learning and the decision-making ability of reinforcement learning, giving the guide robots strong intelligence. In emergency situations such as fires, these guide robots can independently perceive the environment, learn and adjust the best evacuation strategy in real time. They can analyze the specific location of the fire, the spread of the fire, the flow state of the passengers, and the availability of evacuation passages in real time, and quickly move to a specific area near the fire. Through DRL technology, guide robots can play a key role at critical moments. They can provide necessary support and guidance at key locations such as stairways to ensure that pedestrians can reach the stairway area more safely and quickly and evacuate to a safe area (e.g., if the fire is in the boarding layer, the safe area can be the upper layer of the boarding layer). Therefore, the subway station fire emergency evacuation system based on DRL technology exhibits a high level of intelligence and automation. It can provide timely and effective safety protection for pedestrians in emergency situations such as fires, greatly improving the safety performance of subway stations.

[0084] That is, the subway fire scene robot-assisted pedestrian flow evacuation simulation method of the present application has the following advantages:

[0085] First, the present application integrates fire characteristics into the social force model, successfully constructing a pedestrian motion model for subway fire scenarios. Not only can it accurately simulate the movement of pedestrians under subway fire conditions, but it can also provide strong data support for subsequent evacuation strategy development.

[0086] Second, the present application uses deep reinforcement learning to control guide robots, enabling them to dynamically adjust evacuation strategies based on the location and intensity of fire danger zones. This intelligent decision support capability enables guide robots to respond more flexibly to complex fire scene situations, ensuring the smooth progress of the evacuation process.

[0087] Third, the present application uses dynamic guide robots for pedestrian evacuation guidance, significantly improving the evacuation efficiency of subway fire scenes. In fire danger zones, robots can quickly navigate to the best location, effectively guiding pedestrians to avoid danger, reducing congestion, and ensuring that personnel can quickly and orderly evacuate to a safe area.

[0088] Fourth, the present application provides a practical guidance scheme for pedestrian evacuation management in subway fire accidents and lays a solid theoretical foundation for improving evacuation strategies and safety measures in subway fire scenes.

[0089] The subway fire scene robot-assisted pedestrian flow evacuation simulation method will be described below with reference to specific examples.

[0090] In this example, the fire occurs in the boarding layer of the subway station, focusing on the area in front of the stairs of the subway boarding layer, as shown in Figure 2 The area is subdivided into three parts: the pedestrian waiting area, the robot movement area, and the fire area. In the waiting area, pedestrians continuously enter. In order to ensure the orderly progress of the evacuation process and avoid interference with the flow of pedestrians in front of the stairs, it is particularly stipulated that the robot can only move within the designated area. The fire occurrence area is located above the subway boarding layer, and the main escape direction of the pedestrians is towards the stairs to quickly move away from the fire source. In addition, in order to simulate the authenticity, each pedestrian is abstracted as a cylinder with a radius of 0.3 meters, and its mass is set to 80 kilograms; the robot (i.e. the guiding robot mentioned above) is abstracted as a cylinder with a radius of 0.25 meters. Such setting makes the simulation more close to the actual situation, improving the practicality and reliability of the present application. Figure 2 The left lower diagonal area represents the pedestrian waiting area, and the right lower diagonal area represents the robot movement area. In this example, x and y represent the horizontal and vertical coordinates of the position, respectively.

[0091] (1) Pedestrian movement rules

[0092] The present application focuses on improving the evacuation efficiency of the subway boarding layer in the fire emergency situation, and particularly concerns how pedestrians quickly transfer from the boarding layer to the escalator area, thereby safely evacuating the subway station. In simulating the real situation, the initial state of a pedestrian in the waiting area is set. Once the fire occurs, all pedestrians will immediately begin to evacuate and move collectively towards the stair exit. In each time step, the pedestrian will dynamically update its position according to the improved social force model. This model not only considers the avoidance behavior of pedestrians towards the fire area, but also fully simulates the urgency of pedestrians moving towards the stair exit and the characteristics of collective behavior. During the evacuation process, the pedestrians will follow a certain speed range, with the maximum speed set to 3 m / s and the minimum speed set to 1.2 m / s. In order to more specifically simulate the real situation, the total number of pedestrians is set to 50. This setting aims to reflect the possible passenger density of the subway boarding layer during peak hours, while ensuring the reliability and practicality of the simulation results.

[0093] (2) Robot movement rules

[0094] In the daily operation of the subway station, the primary consideration is to ensure that the robot does not interfere with or affect the normal operation of the subway station. However, in emergency evacuation scenarios such as fire, although the presence and action of the robot is crucial for assisting evacuation, its action must also be strictly controlled within a safe range to avoid exacerbating the chaos and danger during the evacuation process. Therefore, the robot is explicitly limited to Figure 2The robot is shown moving within a restricted area. This restricted area ensures that the robot does not interfere with other normal operations in the subway station while performing its evacuation assistance tasks. At the same time, in emergency situations such as fires, the robot will update its position in real time according to the optimal action recommendations provided by the deep neural network analysis, so that it can respond quickly and accurately when necessary. It is particularly worth noting that considering the possibility of pedestrian congestion in front of the escalator exit, if the robot moves extensively in such areas, it may cause panic among the pedestrians, leading to dangerous incidents such as stampedes. Therefore, it is particularly stipulated that within the first 5 seconds after evacuation begins, the robot can move freely as needed to assist in evacuation; but after that, to ensure safety, the robot will be limited to a range of 1 meter from its own radius for fine-tuning movements to avoid unnecessary interference or panic among pedestrians.

[0095] To further explore the specific impact of different fire locations on the evacuation process of the subway platform, two typical fire scenarios were carefully selected: in case 1, the fire is located in the upper middle of the scene, and in case 2, the fire is located in the upper left of the scene. To comprehensively evaluate the impact of these different fire danger zones on pedestrian evacuation behavior, detailed trajectory data of pedestrians was extracted and analyzed.

[0096] As Figure 3 shown, when pedestrians approach the fire danger zone, they will instinctively choose to avoid the fire source, which fully validates the rationality and effectiveness of the improved social force model. This model not only accurately reflects the avoidance behavior of pedestrians when facing fire danger, but also embodies their urgent need to move towards the safe exit. It is worth noting that according to the different locations of the fire, pedestrians also show significant differences in choosing the route and timing of bypassing the fire. This finding provides important data support and theoretical basis for further optimizing the fire emergency evacuation strategy of the subway station.

[0097] To further analyze the impact of the robot on the evacuation process, a study was carefully planned focusing on the static robot scenario. The robot was placed in different positions to evaluate how these position changes affect evacuation efficiency. Specifically, within the robot movement area, robots were placed statically at intervals of 0.5 meters to explore their potential impact on the evacuation process. The experiment was repeated ten times, and the evacuation time was recorded in detail each time, and the average evacuation time of these experiments was calculated.

[0098] As Figure 4As shown, the evacuation time data for the robot placed statically in different positions under various conditions are detailed. Positions with evacuation time greater than 17s are marked with left-top to right-bottom diagonal lines, indicating that the robot's presence in these positions has a negative effect on evacuation. Positions with evacuation time less than 16s are marked with right-bottom to left-top diagonal lines, indicating that the robot's presence in these positions has a positive effect on evacuation. Through comparative analysis, it is found that when the robot is located in the region y∈(3.5, 6), the evacuation time shows a clear growth trend. This is because this region is the only way for most pedestrians to reach the stairs, and the static presence of the robot forces pedestrians to change their route and detour, significantly increasing the overall evacuation time. However, it is worth noting that when the robot is placed in the specific region x∈(8, 10) and y∈(7, 8), the evacuation time is reduced. This finding shows that the position of the robot has a significant impact on evacuation efficiency, and this impact is not a single trend, but a complex diversity with the change of position. Figure 4 In the figure, left-bottom diagonal lines represent time greater than 17s, indicating a greater negative effect on evacuation; right-bottom diagonal lines represent time less than 16s, indicating a greater positive effect on evacuation.

[0099] Therefore, it is concluded that the position selection of the robot has a crucial impact on the evacuation effect when dealing with different fire situations. In order to optimize the evacuation process, the position of the robot needs to be adjusted flexibly according to the specific fire scene and personnel distribution.

[0100] Next, consider the scenario of human evacuation under the control of deep reinforcement learning robots. First, the evacuation time of three different scenarios (no robot scenario, robot random movement scenario and DRL controlled robot scenario) will be compared and analyzed in detail. In the robot random movement scenario, these robots move freely in the specified area without a clear goal or strategy.

[0101] The evacuation time of different scenarios shown in Table 1 intuitively shows the average evacuation time comparison of the two cases under three different conditions. From the data, it can be seen that the randomly moving robots result in relatively long evacuation time, highlighting the importance of orderly and strategic movement of robots in improving evacuation efficiency. In the DRL controlled robot scenario, the evacuation time is significantly reduced. This fully proves that DRL can effectively guide the robot to move to the area most beneficial to evacuation, thereby greatly improving the overall evacuation efficiency. It is worth noting that although the robot scenario under DRL control has achieved remarkable results, its average evacuation time has not completely reached the optimal evacuation time under the static robot scenario. This difference is mainly due to the time required for the robot to navigate from the initial position to the best evacuation area.

[0102]

[0103] Table 1

[0104] The analysis of the present application will deeply dissect the motion patterns of the robot under different scenarios and initial positions, especially the changes in its trajectory. The initial positions and convergence intervals of the robot shown in Table 2 detail two different initial positions set for the robot in the two case studies: Position A and Position B. Position A is selected based on the position with the longest evacuation time in the static robot scenario, while Position B is randomly determined. In addition, special attention is paid to the motion convergence area of the robot, and key data are extracted from it.

[0105]

[0106] Table 2

[0107] From Table 2, an interesting phenomenon can be observed: although Position A and Position B differ in space, under the control of deep reinforcement learning, both initial positions eventually enable the robot to transition to similar motion convergence areas. This phenomenon is further confirmed in Figure 5 , which shows the motion trajectories of pedestrians in the robot scenario under DRL control. Obviously, regardless of the changes in the fire scenario or the initial position of the robot, DRL can effectively guide the robot to quickly move into a specific area, thereby significantly optimizing the evacuation process of pedestrians.

[0108] Further, in order to comprehensively verify the universality and robustness of the proposed method, the motion results of the robot at 10 different initial positions are presented in Figure 6 Figure 6 The two thin solid lines inrepresent the maximum and minimum values of the 10 sets of motion trajectories of the robot, and the thick solid line represents the average value of the 10 sets of motion trajectories. From the figure, it can be clearly seen that regardless of the changes in the location and intensity of the fire danger zone, the robot can successfully and consistently move into the corresponding evacuation optimal interval within 5 seconds. This result shows that the robot control method based on deep reinforcement learning adopted has high adaptability and flexibility, which can quickly adapt to various complex and dynamic environmental conditions, thereby effectively guiding the evacuation of pedestrians. This provides new solutions and ideas for future emergency management and public safety fields. Figure 6 The horizontal axis in represents time.

[0109] This paper investigates how DRL can always lead the robot to the optimal position in various fire hazard areas regardless of the initial position of the robot, thereby significantly improving evacuation efficiency. Next, we will discuss how to further optimize the evacuation process through strategic robot placement from the unique perspective of instantaneous density. In calculating the instantaneous density, an accurate and efficient method is adopted: a circular area with an area of 0.1 m2 is drawn around each monitoring point, and then the number of pedestrians in this area is counted, and this count is divided by the area of the drawn circle to obtain the instantaneous density of the area. In Figure 7 and Figure 8 , the density of pedestrians is represented by the density of the oblique lines, and the more dense the oblique lines, the greater the density. As shown in Figure 7 , when the robot is placed near the lower part of the exit, its movement trajectory significantly reduces the pedestrian density in the surrounding area, thereby effectively shortening the overall evacuation time. As shown in Figure 8 , if the robot is placed near the upper part of the exit, its movement effectively disperses the surrounding pedestrians, thereby significantly reducing congestion and further improving evacuation efficiency.

[0110] As can be seen, placing the robot in the right place can indeed effectively alleviate congestion and improve evacuation efficiency. However, in the special situation of fire, it is also crucial to analyze whether the placement of the robot may increase the risk of injury to pedestrians. For this purpose, the average distance of all pedestrians to the fire hazard area and the average distance of the nearest 10 pedestrians to the fire hazard area are considered as indicators to evaluate the risk index of pedestrians.

[0111] As shown in Figure 9 , in case 2, the robot in the specific area effectively increases the average distance between pedestrians and the fire hazard area in the later stage of evacuation. This significant observation shows that in these cases, the robot under DRL control does not negatively affect the safety of pedestrians. In particular, in the later stage of evacuation, the robot in the specific area significantly increases the average distance between pedestrians and the fire hazard area, not only improving evacuation efficiency, but also significantly enhancing the safety of evacuation. However, case 1 shows another situation where the robot in the specific area actually reduces the average distance between pedestrians and the fire hazard area, which implies that in this specific case, the robot under DRL control may sacrifice safety to some extent to improve evacuation efficiency. As shown in Figure 10As shown, in case 2, the average distance per person between the 10 most dangerous pedestrians in the specific area and the fire danger zone increased to varying degrees, which further proves that the robot under DRL control does not pose a threat to the safety of most pedestrians. However, in case 1, the average distance during evacuation decreased, indicating that in some cases, although the DRL-controlled robot can improve evacuation efficiency, it may also sacrifice some safety to a certain extent. The fundamental reason for this trade-off between safety and efficiency is presumably that safety considerations have not been fully incorporated into the reward function settings. Therefore, in future research, the scheme of the present application can be further optimized. Figure 9 and Figure 10 The hollow circles in the above figures represent the scheme without a robot, and the solid circles represent the scheme of the present application.

[0112] The above describes the preferred embodiments of the present application. It should be pointed out that for ordinary skilled persons in the technical field, several improvements and refinements can be made without departing from the principles described in the present application, and these improvements and refinements should also be considered within the protection scope of the present application.

Claims

1. A method for subway fire scenario robot-assisted pedestrian flow evacuation simulation, characterized in that, The pedestrian flow evacuation simulation method comprises: obtaining a first position of a fire in a subway station and second positions of all pedestrians in a dangerous area of the fire; determining a third position of a guide robot for evacuation shunting of the pedestrians in the dangerous area according to the first position and the second positions, and controlling the guide robot to move to the third position; inputting the first position, the second position and the third position into an improved social force model to obtain an acceleration of each pedestrian in the dangerous area from the dangerous area to a safe area; the improved social force model is a social force model into which fire scene characteristic parameters are fused, the fire scene characteristic parameters including a driving force, an interaction force between pedestrians, an interaction force between pedestrians and obstacles, and an interaction force between pedestrians and fire; the driving force represents a difference between a current speed of the pedestrian and an expected speed thereof, the interaction force between pedestrians represents an interaction force generated when physical contact occurs between the pedestrians, the interaction force between the pedestrians and the obstacles represents an interaction force between the pedestrians and the obstacles, and the interaction force between the pedestrians and the fire represents an interaction force of the pedestrians away from the fire; for each pedestrian in the dangerous area, determining a position of the pedestrian at a next time according to the acceleration of the pedestrian from the dangerous area to the safe area, and guiding the pedestrian to move to the position; when the next time arrives, detecting whether there is an unevacuated person in the dangerous area of the fire, and if there is, returning to perform the steps of obtaining the first position of the fire in the subway station and the second positions of all the pedestrians in the dangerous area of the fire, otherwise ending the simulation.

2. The pedestrian flow evacuation simulation method according to claim 1, wherein, The determining of the third position of the guide robot for evacuation shunting of the pedestrians in the dangerous area according to the first position and the second positions comprises: inputting the first position and the second position into a learned deep neural network to obtain the third position of the guide robot for evacuation shunting of the pedestrians in the dangerous area.

3. The pedestrian flow evacuation simulation method of claim 1, wherein, The improved social force model is: wherein m i denotes the mass of the i-th pedestrian, denotes the acceleration a t , f i denotes the driving force of the i-th pedestrian, f ij denotes the force between the i-th pedestrian and the j-th pedestrian, f iw denotes the force between the i-th pedestrian and the obstacle, f if denotes the force between the i-th pedestrian and the fire.

4. The pedestrian flow evacuation simulation method according to claim 3, characterized in that, wherein, vi(t) denotes the expected speed of the i-th pedestrian, vi(t) denotes the expected speed of the i-th pedestrian, i vi(t) denotes the expected speed of the i-th pedestrian, i denotes the time interval; wherein v min represents a preset minimum speed, v max represents a preset maximum speed, α represents a preset parameter, d fire represents a distance between the ith pedestrian and the fire, r fire represents a radius of the fire.

5. The pedestrian flow evacuation simulation method according to claim 4, characterized in that, f ij = A i exp[(r ij -d ij ) / B i ]n ij + kg(r ij -d ij )n ij where A i = 2000 N, r ij denotes the sum of the radii of the ith pedestrian and the jth pedestrian, d ij denotes the distance between the ith pedestrian and the jth pedestrian, B i = 0.08 m, n ij denotes a unit vector, k = 1.2 x 10 5 kg s -2 , the function g(x) takes a value equal to zero if the ith pedestrian and the jth pedestrian are not in contact, x denotes a variable of the function g(x), otherwise g(x) is equal to x.

6. The pedestrian flow evacuation simulation method according to claim 5, characterized in that, f iw = A i exp[[(r i -d iw ) / B i ]]n iw + kg(r i -d iw )n iw where r i represents the radius of the ith pedestrian, d iw represents the distance between the ith pedestrian and the obstacle w, n iw represents a unit vector, g(x) takes a value equal to zero if the ith pedestrian and the obstacle w are not in contact with each other, x represents a variable of the function g(x), otherwise, g(x) is equal to x.

7. The pedestrian flow evacuation simulation method according to claim 6, characterized in that, f if = λ exp(-d fire )(1 + I) n if where λ = 480N, n if denotes a unit vector.

8. The pedestrian flow evacuation simulation method of claim 3, wherein, The determining of the position of the pedestrian at the next time according to the acceleration of the pedestrian from the dangerous area to the safe area comprises: obtaining the position of the ith pedestrian at t+1 by a position update formula; the position update formula is: wherein x t+1 represents the position of the i-th pedestrian at time t+1, x t represents the position of the i-th pedestrian at time t, t1 is 0.005 seconds, time t is the current time, and time t+1 is the next time of the current time.

Images