Dynamic crowd evacuation simulation method based on generative artificial intelligence and application thereof

Abstract

The invention discloses a dynamic crowd evacuation simulation method based on generative artificial intelligence and an application thereof. The method comprises the following steps: constructing an evacuation scene model; initializing a behavior state model; dynamically generating individual behavior parameters based on the state information of the pedestrian and a pre-trained generative artificial intelligence model; performing evacuation simulation in the evacuation scene model, planning an evacuation path based on the individual behavior parameters and the state information in each time step of simulation, simulating movement behaviors of pedestrians based on the state information, the individual behavior parameters and the evacuation path of the pedestrians, and updating the behavior state model, and the simulation is finished. According to the method, the individual behavior parameters of environment perception are generated based on the generative artificial intelligence model, and path generation, selection and re-planning are performed based on conflict perception, so that the stability, authenticity and application range of the crowd evacuation model in a complex scene are effectively improved.

Description

Technical Field

[0001] This application belongs to the field of crowd evacuation modeling and simulation technology, specifically involving a dynamic crowd evacuation simulation method based on generative artificial intelligence and its application. Background Technology

[0002] With the continuous advancement of urbanization, the population density in large public buildings, transportation hubs, underground spaces, and complexes is constantly increasing. Under conditions of emergencies such as fires, earthquakes, and explosions, efficient and reliable crowd evacuation capabilities have become one of the key technical issues in the fields of public safety engineering and urban emergency management. In order to assess evacuation safety during the planning and design phase and to assist in decision analysis during emergency management, the construction of evacuation modeling and simulation methods that can realistically depict the characteristics of crowd movement and behavioral evolution has significant engineering application value.

[0003] Existing crowd evacuation modeling methods can be broadly categorized into three types: macroscopic models, microscopic models, and macro-micro hybrid models. Macroscopic models typically treat crowds as a continuous medium, employing fluid dynamics or continuous medium theory to describe the overall density and flow evolution. While computationally efficient, they struggle to characterize individual differences and local interactions. Microscopic models, on the other hand, focus on individuals, using rule-driven or dynamic equations to describe pedestrian movement and interaction in space. They can realistically reflect local congestion, conflicts, and path selection, and are therefore widely used in evacuation simulation and safety assessment.

[0004] Among numerous micro-level crowd modeling methods, the Social Force Model (SFM) has become one of the most representative fundamental models in current crowd evacuation research and engineering applications due to its advantages such as clear physical meaning, simple model structure, and ease of integration with numerical simulation platforms. This model constructs continuous-time dynamic equations by introducing terms such as expected force, pedestrian-pedestrian interaction force, and pedestrian-obstacle interaction force, thereby describing the evolution of individual movement behavior during evacuation. Crowd simulation methods based on the social force model have been integrated into various evacuation analysis software and research platforms, demonstrating a high level of engineering maturity.

[0005] However, with the increasing complexity of evacuation scenarios, traditional social force models have gradually revealed a series of unavoidable technical shortcomings in practical applications. First, the behavioral parameters in the model (such as expected speed, social force strength, and psychological distance) typically rely on manual experience or static calibration, making it difficult to dynamically and adaptively adjust according to changes in crowd density, environmental constraints, or sudden disturbances. This limits the model's applicability in multi-density, multi-exit, and complex obstacle environments. Second, traditional social force models have limited ability to characterize individual heterogeneity; behavioral differences between individuals are often approximated by a few parameter perturbations, failing to truly reflect the diverse decision-making and response behaviors of people in emergency situations. Furthermore, in high-density evacuation or bottleneck scenarios, traditional social force models are prone to non-physical phenomena such as local congestion, amplified oscillations, and accumulated conflicts, leading to biased evacuation efficiency assessments. Existing improvement methods mainly alleviate these problems by introducing additional constraint rules, local heuristic strategies, or adjusting parameters for specific scenarios, but these remain passive corrections, lacking a unified model of the pedestrian behavior strategy generation mechanism, making it difficult to achieve transfer and generalization across different scenarios. Summary of the Invention

[0006] The purpose of this application is to provide a dynamic crowd evacuation simulation method based on generative artificial intelligence and its application, so as to solve the problems of behavioral parameters being difficult to adaptively adjust and individual behavioral differences being difficult to truly reflect in existing crowd evacuation modeling using traditional social force models.

[0007] To achieve the above objectives, the first aspect of this application provides a dynamic crowd evacuation simulation method based on generative artificial intelligence, comprising:

[0008] Construct an evacuation scenario model, which describes the spatial extent of the evacuation area, the location of the exit area, and the location of obstacles;

[0009] Initialize the behavior state model, which is used to describe the state information of each pedestrian in the evacuation crowd at any time.

[0010] Based on the pedestrian's state information and a pre-trained generative artificial intelligence model, individual behavioral parameters are dynamically generated.

[0011] Evacuation simulation is performed in the evacuation scenario model. At each time step of the simulation, an evacuation path is planned based on the individual behavior parameters and the state information. Based on the pedestrian's state information, the individual behavior parameters, and the evacuation path, the pedestrian's movement behavior is simulated, and the behavior state model is updated until the simulation ends.

[0012] In one or more embodiments, the individual behavior parameters include expected speed parameters, reaction time parameters, pedestrian avoidance weights, obstacle avoidance weights, and conflict avoidance intensity parameters.

[0013] In one or more embodiments, the status information includes position, velocity, and target exit;

[0014] The step of dynamically generating individual behavior parameters based on pedestrian state information and a pre-trained generative artificial intelligence model includes:

[0015] Based on the state information, the environmental semantic information and evacuation task constraints of the target pedestrian are obtained. The environmental semantic information includes the crowd density in the local neighborhood where the target pedestrian is located, the distance from the target pedestrian to the target exit, and the distance between the target pedestrian and the nearest obstacle.

[0016] The state information, the environmental semantic information, the evacuation task constraints, and the random latent variables are input into the generative artificial intelligence model to obtain the individual behavior parameters of the target pedestrian.

[0017] In one or more embodiments, the step of planning evacuation routes based on the individual behavioral parameters and the state information includes:

[0018] Based on the individual behavior parameters, a local conflict neighborhood of the target pedestrian is constructed;

[0019] The state information, the local conflict neighborhood, the environmental semantic information, the individual behavior parameters, and the random latent variables are input into the generative path strategy model to obtain multiple candidate paths;

[0020] Calculate the speed conflict risk between the target pedestrian and other pedestrians in the local conflict neighborhood in each of the candidate paths, and obtain the conflict cost;

[0021] Calculate the spatial conflict risk between the target pedestrian and obstacles in each of the candidate paths to obtain the obstacle cost;

[0022] Calculate the path length cost for each of the candidate paths;

[0023] Based on the conflict cost, obstacle cost, and path length cost, the comprehensive cost of each candidate path is obtained, and the candidate path with the optimal comprehensive cost is selected as the evacuation path.

[0024] In one or more embodiments, the formula for the local conflict neighborhood is as follows:

[0025] ;

[0026] In the formula, the conflict sensing radius ,in, For weight parameters, For the desired speed parameter, The reaction time parameter is given.

[0027] In one or more embodiments, the conflict cost is calculated using the following formula:

[0028] ,

[0029] In the formula, Here, s represents a candidate path, and s represents a path point within that candidate path. Let i represent the target pedestrian, and j represent other pedestrians located in the local conflict neighborhood. This represents the risk of velocity conflict at path point s. , The position at time step t. Let t be the velocity at time step t.

[0030] In one or more embodiments, the formula for calculating the obstacle cost is as follows:

[0031] ,

[0032] In the formula, Here, s represents a candidate path, and s represents a path point within that candidate path. This represents the spatial conflict risk at path point s. , The conflict avoidance strength parameter is... Let t be the distance between the target pedestrian and the nearest obstacle at time step t.

[0033] In one or more embodiments, the path length cost is calculated using the following formula:

[0034] .

[0035] In one or more embodiments, the state information includes position, velocity, and target exit; the step of simulating pedestrian movement behavior based on the pedestrian's state information, individual behavior parameters, and evacuation path includes:

[0036] Based on the target exit and the evacuation route, determine the expected direction of movement of the target pedestrian;

[0037] Based on the target pedestrian's desired direction of movement, desired velocity parameter, reaction time parameter, and velocity, the target pedestrian's desired driving force is obtained;

[0038] Based on the continuous social force model, determine the pedestrian repulsion force and obstacle repulsion force of the target pedestrian;

[0039] Based on the pedestrian repulsion force and the pedestrian avoidance weight, the pedestrian driving force of the target pedestrian is obtained;

[0040] Based on the obstacle repulsion force and the obstacle avoidance weight, the obstacle driving force of the target pedestrian is obtained;

[0041] Based on the expected driving force, pedestrian driving force, and obstacle driving force, the instantaneous acceleration of the target pedestrian is determined, and the state information of the target pedestrian is updated based on the instantaneous acceleration.

[0042] In one or more embodiments, the step of determining the desired movement direction of the target pedestrian based on the target exit and the evacuation route is specifically as follows:

[0043] ;

[0044] In the formula, This indicates the geometric direction in which the target pedestrian points towards their target exit. This indicates the path direction determined based on the evacuation route. The response coefficient of the target pedestrian to the policy correction.

[0045] In one or more embodiments, the instantaneous acceleration is calculated using the following formula:

[0046] ;

[0047] In the formula, i represents the target pedestrian, j represents other pedestrians, and k represents obstacles. Indicates the desired direction of motion. This represents the desired speed parameter. This refers to the reaction time parameter. This indicates the pedestrian avoidance weight. This represents the pedestrian repulsion force. Indicates obstacle avoidance weight. This indicates the repulsive force of the obstacle.

[0048] In one or more embodiments, it further includes:

[0049] After the simulation, a global feedback index is constructed based on the average speed, average conflict rate, and evacuation success rate during the evacuation process.

[0050] Based on the global feedback metric, a performance evaluation function is constructed, which is specifically expressed as follows:

[0051] ;

[0052] In the formula, These are the weighting coefficients. This represents the average speed during the evacuation process. This represents the average conflict occurrence rate per unit of time. This indicates the percentage of individuals who successfully reach the evacuation exit within the specified time.

[0053] In one or more embodiments, the training method for the generative artificial intelligence model includes:

[0054] Based on a generative artificial intelligence model, multiple evacuation simulations are performed in the target evacuation scenario. After each simulation, the performance evaluation function is constructed.

[0055] Based on the performance evaluation function, a loss function is constructed, and the loss function is specifically as follows: In the formula, Let be the performance evaluation function. Indicates a smoothing regularization term. These are the individual behavioral parameters in the k-th simulation;

[0056] Based on the loss function, the parameters of the generative artificial intelligence model are updated along the direction of gradient descent until the loss function converges.

[0057] In one or more embodiments, it further includes:

[0058] Obtain the performance evaluation function of the generative artificial intelligence model trained under different evacuation scenarios and the individual behavior parameters generated in the simulation;

[0059] Based on the performance evaluation function and individual behavior parameters under different evacuation scenarios, an objective function is constructed, which is specifically as follows: In the formula, Let represent the expected value of the performance evaluation function in the m-th evacuation scenario. This represents the consistency weight coefficient across different scenarios. Let represent the individual behavior parameters in the m-th evacuation scenario. The Euclidean distance used to describe the individual behavioral parameters in the m-th and n-th evacuation scenarios;

[0060] The parameters of the generative artificial intelligence model are updated in the direction of minimizing the objective function.

[0061] To achieve the above objectives, a second aspect of this application provides a dynamic crowd evacuation simulation device based on generative artificial intelligence, comprising:

[0062] The scene construction module is used to build an evacuation scene model, which describes the spatial range of the evacuation area, the location of the exit area, and the location of obstacles.

[0063] The state model construction module is used to initialize the behavioral state model, which describes the state information of each person in the evacuation crowd at any time.

[0064] The individual parameter generation module is used to dynamically generate individual behavioral parameters based on the pedestrian's state information and a pre-trained generative artificial intelligence model.

[0065] The simulation module is used to perform evacuation simulation in the evacuation scenario model. At each time step of the simulation, it plans the evacuation path based on the individual behavior parameters and the state information, simulates the pedestrian's movement behavior based on the pedestrian's state information, the individual behavior parameters and the evacuation path, and updates the behavior state model until the simulation ends.

[0066] To achieve the above objectives, a third aspect of this application provides an electronic device, comprising:

[0067] At least one processor; and

[0068] A memory that stores instructions that, when executed by the at least one processor, cause the at least one processor to perform the dynamic crowd evacuation simulation method based on generative artificial intelligence as described in any of the above embodiments.

[0069] To achieve the above objectives, a fourth aspect of this application provides a machine-readable storage medium storing executable instructions that, when executed, cause the machine to perform the dynamic crowd evacuation simulation method based on generative artificial intelligence as described in any of the above embodiments.

[0070] The advantages of this application, which differ from existing technologies, are:

[0071] This application generates individual behavioral parameters for environmental perception based on a generative artificial intelligence model, and performs path generation, selection and replanning based on conflict perception, which effectively improves the stability, realism and applicability of crowd evacuation models in complex scenarios. Attached Figure Description

[0072] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0073] Figure 1 This is a flowchart illustrating one implementation method of the dynamic crowd evacuation simulation method based on generative artificial intelligence in this application.

[0074] Figure 2 yes Figure 1 A flowchart illustrating one embodiment corresponding to S300;

[0075] Figure 3 This is a flowchart illustrating one implementation method of the evacuation route planning method of this application;

[0076] Figure 4 This is a flowchart illustrating one implementation method of the evacuation simulation method in this application;

[0077] Figure 5 This is a flowchart illustrating one implementation method of the generative artificial intelligence model of this application;

[0078] Figure 6 This is a schematic diagram of one embodiment of the dynamic crowd evacuation simulation device based on generative artificial intelligence in this application;

[0079] Figure 7 This is a structural block diagram of one embodiment of the electronic device of this application. Detailed Implementation

[0080] To enable those skilled in the art to better understand the technical solutions in this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this disclosure.

[0081] When traditional social force models are used for evacuation simulations, the behavioral parameters of different pedestrians in the model are set by human experience or statically calibrated. This makes it difficult to dynamically and adaptively adjust to changes in crowd density, environmental constraints, or sudden disturbances, thus limiting the model's applicability in multi-density, multi-exit, and complex obstacle environments. Furthermore, traditional social force models have limited ability to characterize individual heterogeneity; behavioral differences between individuals are often approximated by a few parameter perturbations, failing to accurately reflect the diverse decision-making and response behaviors of crowds in emergency situations. In addition, in high-density evacuation or bottleneck scenarios, traditional social force models are prone to non-physical phenomena such as localized congestion, amplified oscillations, and accumulated conflicts, leading to biased evacuation efficiency assessments.

[0082] Some existing improvement methods alleviate the above problems by introducing additional constraint rules, local heuristic strategies, or adjusting parameters for specific scenarios, but they are still passive corrections overall. They lack a unified model of the pedestrian behavior strategy generation mechanism and are difficult to transfer and generalize across different scenarios.

[0083] To address the aforementioned issues, the applicant has developed a novel dynamic crowd evacuation simulation method based on generative artificial intelligence. This method utilizes generative artificial intelligence to dynamically generate individual behavioral parameters for different pedestrians. It can dynamically generate and optimize individual behavioral characteristics of pedestrians based on environmental perception and group status, thereby improving the stability and realism of the crowd evacuation simulation model in complex scenarios and expanding the applicability of the crowd evacuation simulation model.

[0084] The technical solution of this application is described in detail below with reference to the accompanying drawings. Please refer to the accompanying drawings. Figure 1 , Figure 1 This is a flowchart illustrating one implementation method of the dynamic crowd evacuation simulation method based on generative artificial intelligence in this application.

[0085] like Figure 1 As shown, the method includes:

[0086] S100, Construct an evacuation scenario model.

[0087] The evacuation scenario describes the spatial extent of the evacuation area, the location of the exit area, and the location of obstacles.

[0088] In one implementation, a two-dimensional evacuation zone can be constructed. This represents the actual space available for pedestrian movement. This area can be defined by wall boundaries and includes multiple pre-designed exit areas. .

[0089] Each exit Its location, width, and direction of passage can be determined together to describe the final destination of pedestrians during evacuation.

[0090] Obstacles can be set up within the evacuation area. It is used to simulate pillars, partitions, or impassable areas that exist in a real environment.

[0091] Based on the above steps, a unified representation of the evacuation scenario is constructed.

[0092] S200, Initialize the behavior state model.

[0093] Among them, the behavioral state model is used to describe the state information of each person in the evacuation crowd at any time.

[0094] In one implementation, for the first A pedestrian, at any time Define its state as

[0095] ;

[0096] in: pedestrian Current location; pedestrian The instantaneous velocity vector; pedestrian Selected target export.

[0097] S300 dynamically generates individual behavioral parameters based on pedestrian state information and a pre-trained generative artificial intelligence model.

[0098] Unlike traditional social force models, this application uses a generative artificial intelligence model to dynamically generate individual behavioral parameters. These individual behavioral parameters fully consider the characteristics of the actual environment and are used to characterize the behavioral differences of different pedestrians in different evacuation environments.

[0099] The training methods for generative artificial intelligence models will be discussed in detail later.

[0100] In one implementation, the individual behavioral parameters include expected speed parameters, reaction time parameters, pedestrian avoidance weights, obstacle avoidance weights, and conflict avoidance intensity parameters, which can be expressed as follows:

[0101] ;

[0102] in, The desired velocity parameter reflects pedestrian i's willingness to move autonomously under undisturbed conditions; is the reaction time parameter, describing how quickly pedestrian i responds to changes in the external environment; The avoidance weight between pedestrians characterizes the intensity of pedestrian i's avoidance of others approaching; Obstacle avoidance weights characterize the degree to which pedestrian i avoids walls or obstacles; The conflict avoidance strength parameter is used to adjust the intensity of pedestrian i's response to local density and conflict.

[0103] Please see Figure 2 , Figure 2 yes Figure 1 A flowchart of one embodiment corresponding to S300.

[0104] like Figure 2 As shown, methods for generating individual behavioral parameters may include:

[0105] S301. Based on state information, obtain the environmental semantic information of the target pedestrian and the evacuation task constraints.

[0106] Among them, environmental semantic information is used to describe the environmental characteristics of the target pedestrian. It can include the crowd density in the local neighborhood of the target pedestrian, the distance from the target pedestrian to the target exit, and the distance between the target pedestrian and the nearest obstacle. It can be obtained based on the position of each pedestrian in the state information.

[0107] Specifically, environmental semantic information can be expressed as follows:

[0108] ;

[0109] In the formula, pedestrian Population density in the local neighborhood; pedestrian The Euclidean distance to its target exit; pedestrian Distance to the nearest obstacle.

[0110] The evacuation task constraint is used to represent the evacuation target of the target pedestrian, which may include information such as the number or priority of the target exit.

[0111] S302. Input the state information, environmental semantic information, evacuation task constraints, and random latent variables into the generative artificial intelligence model to obtain the individual behavior parameters of the target pedestrian.

[0112] Specifically, conditional generative models can be constructed based on generative artificial intelligence models:

[0113] .

[0114] in, Represents the individual behavioral parameters in the k-th simulation. Indicates parameters Generative artificial intelligence models; Status information; For environmental semantic information; These are random latent variables that follow a predefined probability distribution (such as a standard Gaussian distribution) and are used to introduce individual behavioral differences to avoid generating completely consistent behavioral patterns for all individuals. pedestrian Corresponding evacuation task constraints.

[0115] Based on the above-mentioned conditional generation model, individual behavioral parameters that are aware of the environment can be generated, which can characterize the differences between different individuals in different evacuation scenarios. This realizes the transformation of social force model behavioral parameters from static preset constants to dynamically generated variables driven by environmental and task conditions, thus avoiding damage to the original dynamic structure and interpretability of the social force model.

[0116] S400. Perform evacuation simulation in the evacuation scenario model. At each time step of the simulation, plan the evacuation path based on individual behavior parameters and state information. Simulate the pedestrian's movement behavior based on the pedestrian's state information, individual behavior parameters, and evacuation path, and update the behavior state model until the simulation ends.

[0117] Simulations are performed based on individual behavioral parameters. In the simulation, the evacuation dynamics evolution based on the continuous social force model is used to update the position of each pedestrian at each time step until the simulation ends.

[0118] Unlike conventional simulations, this implementation dynamically plans evacuation routes based on the pedestrian's environment at each time step of the simulation. Then, it uses a continuous social force model to perform evacuation simulation based on the evacuation routes, so as to achieve a safe and efficient crowd evacuation strategy.

[0119] Specifically, please refer to Figure 3 , Figure 3 This is a flowchart illustrating one implementation method of the evacuation route planning method of this application.

[0120] like Figure 3 As shown, methods for planning evacuation routes may include:

[0121] S401a. Construct a local conflict neighborhood for the target pedestrian based on individual behavior parameters.

[0122] The local conflict neighborhood is used to represent the conflict perception radius that the target pedestrian can perceive, and the pedestrians and obstacles within this radius have a conflict risk with the target pedestrian.

[0123] The local conflict neighborhood is determined by the location of the target pedestrian and individual behavioral parameters, and its formula can be as follows:

[0124] ;

[0125] In the formula, the conflict sensing radius ,in, For the desired speed parameter, This refers to the reaction time parameter.

[0126] S402a: Input state information, local conflict neighborhood, environmental semantic information, individual behavior parameters, and random latent variables into the generative path strategy model to obtain multiple candidate paths.

[0127] After obtaining the local conflict neighborhood, multiple candidate paths are further generated based on the generative path strategy model.

[0128] Specifically, a conditional generation model for the current time step can be constructed based on the generative path strategy model:

[0129] ;

[0130] In the formula, Represents candidate paths, This is a generative path strategy model. These are random latent variables used to enhance path diversity. This refers to the state information at time step t; For the environmental semantic information at time step t, These are parameters representing individual behavior.

[0131] S403a. Calculate the speed conflict risk between the target pedestrian and other pedestrians in the local conflict neighborhood in each candidate path, and obtain the conflict cost.

[0132] Specifically, the formula for calculating the cost of conflict is as follows:

[0133] ,

[0134] In the formula, Here, s represents a candidate path, and s represents a path point within that candidate path. Let i represent the target pedestrian, and j represent other pedestrians located in the local conflict neighborhood. This represents the risk of velocity conflict at path point s.

[0135] ;

[0136] In the formula, The position at time step t. Let t be the velocity at time step t.

[0137] S404a. Calculate the spatial conflict risk between the target pedestrian and obstacles in each candidate path to obtain the obstacle cost.

[0138] Specifically, the formula for calculating the cost of the obstacle can be as follows:

[0139] ,

[0140] In the formula, Here, s represents a candidate path, and s represents a path point within that candidate path. This represents the spatial conflict risk at path point s;

[0141] ;

[0142] For conflict avoidance strength parameters, Let t be the distance between the target pedestrian and the nearest obstacle at time step t.

[0143] S405a. Calculate the path length cost for each candidate path.

[0144] Specifically, the formula for calculating path length cost is as follows:

[0145] .

[0146] S406a. Based on the conflict cost, obstacle cost, and path length cost, obtain the comprehensive cost of each candidate path, and select the candidate path with the optimal comprehensive cost as the evacuation path.

[0147] Based on the calculated costs of conflict with pedestrians, obstacles, and path length, a comprehensive cost for each evacuation path can be constructed, and the candidate path with the lowest comprehensive cost can be selected as the evacuation path.

[0148] Specifically, the comprehensive cost function can be expressed as follows:

[0149] .

[0150] Based on the above comprehensive path cost function, the optimal path for pedestrian i at time step t is selected:

[0151] .

[0152] In one implementation, when faced with sudden congestion or conflict, the path can be dynamically updated in the direction of minimizing the above comprehensive path cost function, as shown in the following equation: to achieve The adaptive adjustment of evacuation routes in response to sudden congestion or conflict further improves safety and traffic efficiency in high-density evacuation scenarios.

[0153] Based on the individual behavior parameters generated above and the selected evacuation path, simulation can be further performed in the evacuation model. At each time step of the simulation, the pedestrian's expected driving force and repulsive force are comprehensively considered to calculate the pedestrian's acceleration vector and thus update the pedestrian's position.

[0154] Specifically, please refer to Figure 4 , Figure 4 This is a flowchart illustrating one implementation method of the evacuation simulation method in this application.

[0155] like Figure 4 As shown, methods for simulating pedestrian movement behavior may include:

[0156] S401b: Determine the expected direction of movement of the target pedestrian based on the target exit and evacuation route.

[0157] First, the desired movement direction of the target pedestrian can be determined by combining the location of the target exit and the evacuation direction of the target pedestrian's path point in the evacuation route.

[0158] Specifically, it can be expressed as follows: ;

[0159] In the formula, This indicates the geometric direction in which the target pedestrian points towards their target exit. Indicates the path direction determined based on the evacuation route. The response coefficient of the target pedestrian to the policy correction.

[0160] S402b: Based on the target pedestrian's desired motion direction, desired velocity parameters, reaction time parameters, and velocity, the desired driving force of the target pedestrian is obtained.

[0161] First, the expected driving force of the target pedestrian can be calculated, which is determined by the target pedestrian's expected speed and reaction time parameters.

[0162] S403b. Based on the continuous social force model, determine the pedestrian repulsion force and obstacle repulsion force of the target pedestrian.

[0163] In the continuous social force model, the basic pedestrian repulsion force and obstacle repulsion force can be determined based on the distance between the target pedestrian and other pedestrians, as well as the distance between the target pedestrian and the obstacle.

[0164] S404b: Based on pedestrian repulsion force and pedestrian avoidance weights, the pedestrian driving force of the target pedestrian is obtained.

[0165] S405b: Based on obstacle repulsion force and obstacle avoidance weight, the obstacle driving force of the target pedestrian is obtained.

[0166] Furthermore, based on the pedestrian avoidance weight and obstacle avoidance weight, combined with the pedestrian repulsion force and obstacle repulsion force, the pedestrian driving force and obstacle driving force of the target pedestrian can be obtained.

[0167] S406b: Based on the expected driving force, pedestrian driving force, and obstacle driving force, determine the instantaneous acceleration of the target pedestrian, and update the state information of the target pedestrian based on the instantaneous acceleration.

[0168] Specifically, the formula for calculating the instantaneous acceleration of the target pedestrian can be as follows:

[0169] ;

[0170] In the formula, i represents the target pedestrian, j represents other pedestrians, and k represents obstacles. Indicates the desired direction of motion. Indicates the desired speed parameter. Indicates the reaction time parameter. Indicates the weight of pedestrians giving way. Indicates the repulsive force of pedestrians. Indicates obstacle avoidance weight. It represents the repulsive force of an obstacle.

[0171] After obtaining the instantaneous acceleration of the target pedestrian, the position of the target pedestrian in the next time step can be obtained by combining the current velocity of the target pedestrian, thereby realizing the update of the target pedestrian's state information.

[0172] Based on the methods described above, individual behavioral parameters for environmental perception are generated using a generative artificial intelligence model. Simultaneously, path generation, selection, and replanning are performed based on conflict perception, effectively improving the stability, realism, and applicability of crowd evacuation models in complex scenarios.

[0173] Furthermore, in order to provide feedback on evacuation performance, the method also includes:

[0174] After the S500 simulation is completed, a global feedback index is constructed based on the average speed, average conflict rate, and evacuation success rate during the evacuation process.

[0175] In one implementation, the global feedback metric may also include the average population density during the evacuation process.

[0176] S600: Based on global feedback metrics, construct a performance evaluation function.

[0177] Specifically, the performance evaluation function can be as follows:

[0178] ;

[0179] In the formula, The weighting coefficients are used to balance evacuation efficiency and safety requirements. This represents the average speed during the evacuation process. This represents the average conflict occurrence rate per unit of time. This represents the percentage of individuals who successfully reached the evacuation exit within the specified time; the normalized [number]th [unit / item]. Evacuation performance indicators Used to eliminate dimensional differences.

[0180] Based on the above performance evaluation function, the efficiency and stability of evacuation simulation can be effectively reflected. At the same time, generative artificial intelligence models can be trained based on this performance evaluation function.

[0181] The training methods for generative artificial intelligence models are detailed below; please refer to section 5. Figure 5This is a flowchart illustrating one implementation method of the generative artificial intelligence model training method of this application.

[0182] like Figure 5 As shown, the method includes:

[0183] S10. Based on the generative artificial intelligence model, perform multiple evacuation simulations in the target evacuation scenario, and construct a performance evaluation function after each simulation.

[0184] First, multiple simulations were performed in the target evacuation scenario based on a generative artificial intelligence model, and a performance evaluation function for each simulation was constructed.

[0185] S20. Construct a loss function based on the performance evaluation function.

[0186] In one implementation, the loss function is specifically as follows: In the formula, For performance evaluation functions, The smoothing regularization term describes the Euclidean distance between the individual behavior parameters in the k-th simulation and the (k-1)-th simulation. This constrains the variation of individual behavior parameters in adjacent simulation rounds, preventing drastic fluctuations in model parameters from causing unstable evacuation behavior. It also causes the generated model to gradually favor a distribution of behavior parameters that produce higher evacuation efficiency and lower conflict risk during multiple evacuation execution rounds.

[0187] S30. Based on the loss function, update the parameters of the generative artificial intelligence model along the direction of gradient descent until the loss function converges.

[0188] The parameters are updated based on the loss function described above until the loss function converges.

[0189] To improve the model's adaptability and transferability under different evacuation scenarios, further methods include:

[0190] S40. Obtain the performance evaluation function of the generative artificial intelligence model trained under different evacuation scenarios and the individual behavior parameters generated in the simulation.

[0191] S50. Based on the performance evaluation functions and individual behavior parameters under different evacuation scenarios, construct the objective function.

[0192] Specifically, the set of evacuation scenarios can be defined as follows: In different evacuation scenarios, there are differences in the number of exits, the layout of obstacles, the spatial scale, and the initial distribution of crowd density.

[0193] For specific evacuation scenarios The empirical distribution of individual behavioral parameters is obtained through generative artificial intelligence models. Meanwhile, the empirical distribution of individual behavioral parameters in different scenarios is measured using the mean squared error distribution similarity function. and The differences between them ultimately lead to the construction of the following cross-scenario joint optimization objective function:

[0194] ;

[0195] In the formula, This represents the expected value of the performance evaluation function in the m-th evacuation scenario, with the first term used to ensure the parameters of the generated model. In the scene It has good evacuation performance;

[0196] This represents the consistency weight coefficient across different scenarios. Let represent the individual behavior parameters in the m-th evacuation scenario. The Euclidean distance is used to describe the individual behavioral parameters of the m-th and n-th evacuation scenarios.

[0197] S60. Update the parameters of the generative artificial intelligence model in the direction of minimizing the objective function.

[0198] Through the aforementioned cross-scenario joint optimization mechanism, the generative model gradually forms stable and transferable behavior generation rules in various evacuation environments, enabling the continuous self-evolution of crowd evacuation behavior modeling and decision-making models.

[0199] This application also provides a dynamic crowd evacuation simulation device based on generative artificial intelligence. Please refer to [link / reference]. Figure 6 , Figure 6 This is a schematic diagram of one embodiment of the dynamic crowd evacuation simulation device based on generative artificial intelligence in this application.

[0200] like Figure 6 As shown, the device includes a scene construction module 21, a state model construction module 22, an individual parameter generation module 23, and a simulation module 24.

[0201] Among them, the scene construction module 21 is used to construct the evacuation scene model, which describes the spatial range of the evacuation area, the location of the exit area, and the location of obstacles.

[0202] State model building module 22 is used to initialize the behavioral state model, which describes the state information of each person in the evacuation crowd at any time.

[0203] Individual parameter generation module 23 is used to dynamically generate individual behavior parameters based on pedestrian state information and a pre-trained generative artificial intelligence model;

[0204] The simulation module 24 is used to perform evacuation simulation in the evacuation scenario model. At each time step of the simulation, it plans the evacuation path based on individual behavior parameters and state information, simulates the movement behavior of pedestrians based on pedestrian state information, individual behavior parameters and evacuation path, and updates the behavior state model until the simulation ends.

[0205] As referred above Figures 1 to 5 This specification describes a dynamic crowd evacuation simulation method based on generative artificial intelligence, according to embodiments thereof. The details mentioned in the above description of the method embodiments also apply to the dynamic crowd evacuation simulation device based on generative artificial intelligence, as described in the embodiments of this specification. The above-described dynamic crowd evacuation simulation device based on generative artificial intelligence can be implemented in hardware, software, or a combination of hardware and software.

[0206] This application also provides an electronic device, please refer to... Figure 7 , Figure 7 This is a structural block diagram of one embodiment of the electronic device of this application. Figure 7 As shown, the electronic device 30 may include at least one processor 31, a memory 32 (e.g., non-volatile memory), a RAM 33, and a communication interface 34, and the at least one processor 31, memory 32, RAM 33, and communication interface 34 are connected together via a bus 35. The at least one processor 31 executes at least one computer-readable instruction stored or encoded in the memory 32.

[0207] It should be understood that the computer-executable instructions stored in memory 32, when executed, cause at least one processor 31 to perform the above-described combinations in the various embodiments of this specification. Figures 1-5 The description includes various operations and functions.

[0208] In the embodiments of this specification, electronic device 30 may include, but is not limited to: personal computer, server computer, workstation, desktop computer, laptop computer, notebook computer, mobile electronic device, smartphone, tablet computer, cellular phone, personal digital assistant (PDA), handheld device, messaging device, wearable electronic device, consumer electronic device, etc.

[0209] According to one embodiment, a program product, such as a machine-readable medium, is provided. The machine-readable medium may have instructions (i.e., the elements implemented in software as described above), which, when executed by a machine, cause the machine to perform the above-described combinations of the various embodiments of this specification. Figures 1-5The various operations and functions described. Specifically, a system or apparatus equipped with a readable storage medium storing software program code that implements the functions of any of the embodiments described above, and enabling the computer or processor of the system or apparatus to read and execute the instructions stored in the readable storage medium.

[0210] In this case, the program code read from the readable medium itself can perform the functions of any of the above embodiments, and therefore the machine-readable code and the readable storage medium storing the machine-readable code constitute a part of this specification.

[0211] Examples of readable storage media include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD-RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer or the cloud via a communication network.

[0212] Those skilled in the art will understand that the various embodiments disclosed above can be modified and varied without departing from the spirit of the invention. Therefore, the scope of protection of this specification should be defined by the appended claims.

[0213] It should be noted that not all steps and units in the above process and system structure diagrams are mandatory; some steps or units can be omitted according to actual needs. The execution order of each step is not fixed and can be determined as needed. The device structure described in the above embodiments can be a physical structure or a logical structure. That is, some units may be implemented by the same physical client, or some units may be implemented by multiple physical clients, or they may be jointly implemented by certain components in multiple independent devices.

[0214] In the above embodiments, the hardware units or modules can be implemented mechanically or electrically. For example, a hardware unit, module, or processor may include permanent dedicated circuitry or logic (such as a dedicated processor, FPGA, or ASIC) to perform the corresponding operation. The hardware unit or processor may also include programmable logic or circuitry (such as a general-purpose processor or other programmable processor), which can be temporarily configured by software to perform the corresponding operation. The specific implementation method (mechanical, dedicated permanent circuitry, or temporarily configured circuitry) can be determined based on cost and time considerations.

[0215] The specific embodiments described above with reference to the accompanying drawings are exemplary embodiments, but do not represent all embodiments that can be implemented or fall within the scope of the claims. The term "exemplary" as used throughout this specification means "serving as an example, instance, or illustration" and does not imply that it is "preferred" or "advantageous" compared to other embodiments. Specific details are included to provide an understanding of the described techniques. However, these techniques can be practiced without these specific details. In some instances, well-known structures and apparatuses are shown in block diagram form to avoid obscuring the concepts of the described embodiments.

[0216] The foregoing description of this disclosure is provided to enable any person skilled in the art to implement or use this disclosure. Various modifications to this disclosure will be apparent to those skilled in the art, and the general principles applicable herein can be applied to other variations without departing from the scope of this disclosure. Therefore, this disclosure is not limited to the examples and designs described herein, but is consistent with the widest scope of the principles and novel features disclosed herein.

Claims

1. A dynamic crowd evacuation simulation method based on generative artificial intelligence, characterized in that, include: Construct an evacuation scenario model, which describes the spatial extent of the evacuation area, the location of the exit area, and the location of obstacles; Initialize the behavior state model, which is used to describe the state information of each pedestrian in the evacuation crowd at any time. Based on the pedestrian's state information and a pre-trained generative artificial intelligence model, individual behavioral parameters are dynamically generated. Evacuation simulation is performed in the evacuation scenario model. At each time step of the simulation, an evacuation path is planned based on the individual behavior parameters and the state information. Based on the pedestrian's state information, the individual behavior parameters, and the evacuation path, the pedestrian's movement behavior is simulated, and the behavior state model is updated until the simulation ends.

2. The dynamic crowd evacuation simulation method according to claim 1, characterized in that, The individual behavioral parameters include expected speed parameters, reaction time parameters, pedestrian avoidance weights, obstacle avoidance weights, and conflict avoidance intensity parameters.

3. The dynamic crowd evacuation simulation method according to claim 2, characterized in that, The status information includes position, speed, and target exit; The step of dynamically generating individual behavior parameters based on pedestrian state information and a pre-trained generative artificial intelligence model includes: Based on the state information, the environmental semantic information and evacuation task constraints of the target pedestrian are obtained. The environmental semantic information includes the crowd density in the local neighborhood where the target pedestrian is located, the distance from the target pedestrian to the target exit, and the distance between the target pedestrian and the nearest obstacle. The state information, the environmental semantic information, the evacuation task constraints, and the random latent variables are input into the generative artificial intelligence model to obtain the individual behavior parameters of the target pedestrian.

4. The dynamic crowd evacuation simulation method according to claim 3, characterized in that, The step of planning evacuation routes based on the individual behavior parameters and the state information includes: Based on the individual behavior parameters, a local conflict neighborhood of the target pedestrian is constructed; The state information, the local conflict neighborhood, the environmental semantic information, the individual behavior parameters, and the random latent variables are input into the generative path strategy model to obtain multiple candidate paths; Calculate the speed conflict risk between the target pedestrian and other pedestrians in the local conflict neighborhood in each of the candidate paths, and obtain the conflict cost; Calculate the spatial conflict risk between the target pedestrian and obstacles in each of the candidate paths to obtain the obstacle cost; Calculate the path length cost for each of the candidate paths; Based on the conflict cost, obstacle cost, and path length cost, the comprehensive cost of each candidate path is obtained, and the candidate path with the optimal comprehensive cost is selected as the evacuation path.

5. The dynamic crowd evacuation simulation method according to claim 4, characterized in that, The formula for the local conflict neighborhood is as follows: ； In the formula, the conflict sensing radius ,in, For weight parameters, For the desired speed parameter, The reaction time parameter; and / or, The formula for calculating the conflict cost is as follows: ， In the formula, Here, s represents a candidate path, and s represents a path point within that candidate path. Let i represent the target pedestrian, and j represent other pedestrians located in the local conflict neighborhood. This represents the risk of velocity conflict at path point s. , The position at time step t. The velocity at time step t; and / or, The formula for calculating the obstacle cost is as follows: ， In the formula, Here, s represents a candidate path, and s represents a path point within that candidate path. This represents the spatial conflict risk at path point s. , The conflict avoidance strength parameter is... Let t be the distance between the target pedestrian and the nearest obstacle at time step t; and / or, The formula for calculating the path length cost is as follows: 。 6. The dynamic crowd evacuation simulation method according to claim 2, characterized in that, The status information includes position, speed, and target exit; the steps of simulating pedestrian movement behavior based on the pedestrian's status information, individual behavior parameters, and evacuation path include: Based on the target exit and the evacuation route, determine the expected direction of movement of the target pedestrian; Based on the target pedestrian's desired direction of movement, desired velocity parameter, reaction time parameter, and velocity, the target pedestrian's desired driving force is obtained; Based on the continuous social force model, determine the pedestrian repulsion force and obstacle repulsion force of the target pedestrian; Based on the pedestrian repulsion force and the pedestrian avoidance weight, the pedestrian driving force of the target pedestrian is obtained; Based on the obstacle repulsion force and the obstacle avoidance weight, the obstacle driving force of the target pedestrian is obtained; Based on the expected driving force, pedestrian driving force, and obstacle driving force, the instantaneous acceleration of the target pedestrian is determined, and the state information of the target pedestrian is updated based on the instantaneous acceleration.

7. The dynamic crowd evacuation simulation method according to claim 6, characterized in that, The step of determining the expected movement direction of the target pedestrian based on the target exit and the evacuation route is specifically as follows: ； In the formula, This indicates the geometric direction in which the target pedestrian points towards their target exit. This indicates the path direction determined based on the evacuation route. For the target pedestrian's response coefficient to the policy modification; and / or, The formula for calculating the instantaneous acceleration is as follows: ； In the formula, i represents the target pedestrian, j represents other pedestrians, and k represents obstacles. Indicates the desired direction of motion. This represents the desired speed parameter. This refers to the reaction time parameter. This indicates the pedestrian avoidance weight. This represents the pedestrian repulsion force. Indicates obstacle avoidance weight. This indicates the repulsive force of the obstacle.

8. The dynamic crowd evacuation simulation method according to claim 1, characterized in that, Also includes: After the simulation, a global feedback index is constructed based on the average speed, average conflict rate, and evacuation success rate during the evacuation process. Based on the global feedback metric, a performance evaluation function is constructed, which is specifically expressed as follows: ； In the formula, These are the weighting coefficients. This represents the average speed during the evacuation process. This represents the average conflict occurrence rate per unit of time. This indicates the percentage of individuals who successfully reach the evacuation exit within the specified time.

9. The dynamic crowd evacuation simulation method according to claim 8, characterized in that, The training methods for the generative artificial intelligence model include: Based on a generative artificial intelligence model, multiple evacuation simulations are performed in the target evacuation scenario. After each simulation, the performance evaluation function is constructed. Based on the performance evaluation function, a loss function is constructed, and the loss function is specifically as follows: In the formula, Let be the performance evaluation function. Indicates a smoothing regularization term. These are the individual behavioral parameters in the k-th simulation; Based on the loss function, the parameters of the generative artificial intelligence model are updated along the direction of gradient descent until the loss function converges.

10. The dynamic crowd evacuation simulation method according to claim 9, characterized in that, Also includes: Obtain the performance evaluation function of the generative artificial intelligence model trained under different evacuation scenarios and the individual behavior parameters generated in the simulation; Based on the performance evaluation function and individual behavior parameters under different evacuation scenarios, an objective function is constructed, which is specifically as follows: In the formula, Let represent the expected value of the performance evaluation function in the m-th evacuation scenario. This represents the consistency weight coefficient across different scenarios. Let represent the individual behavior parameters in the m-th evacuation scenario. The Euclidean distance used to describe the individual behavioral parameters in the m-th and n-th evacuation scenarios; The parameters of the generative artificial intelligence model are updated in the direction of minimizing the objective function.

11. A dynamic crowd evacuation simulation device based on generative artificial intelligence, characterized in that, include: The scene construction module is used to build an evacuation scene model, which describes the spatial range of the evacuation area, the location of the exit area, and the location of obstacles. The state model construction module is used to initialize the behavioral state model, which describes the state information of each person in the evacuation crowd at any time. The individual parameter generation module is used to dynamically generate individual behavioral parameters based on the pedestrian's state information and a pre-trained generative artificial intelligence model. The simulation module is used to perform evacuation simulation in the evacuation scenario model. At each time step of the simulation, it plans the evacuation path based on the individual behavior parameters and the state information, simulates the pedestrian's movement behavior based on the pedestrian's state information, the individual behavior parameters and the evacuation path, and updates the behavior state model until the simulation ends.

12. An electronic device, comprising: At least one processor; as well as A memory that stores instructions that, when executed by the at least one processor, cause the at least one processor to perform the dynamic crowd evacuation simulation method based on generative artificial intelligence as described in any one of claims 1 to 11.

13. A machine-readable storage medium storing executable instructions that, when executed, cause the machine to perform the dynamic crowd evacuation simulation method based on generative artificial intelligence as described in any one of claims 1 to 11.

Images