Crowd passing management system and method and computer equipment

Abstract

The invention discloses a crowd passing management system and method and computer equipment. The crowd passing management system comprises a crowd simulator and a large language model (LLM) planning agent, the crowd simulator is configured to respond to the first decision scheme output by the LLM planning intelligent agent, simulate pedestrian motion behaviors and crowd dynamics in a scene marked with a passing area, obstacles, vertical traffic and function nodes, and output real-time data streams; the LLM planning intelligent agent is configured to analyze the real-time data flow through a semantic understanding technology and calculate a multi-dimensional congestion index according to the analyzed real-time data flow; and optimizing the first decision scheme according to the multi-dimensional congestion index, outputting a second decision scheme, and sending the second decision scheme to the crowd simulator. In the embodiment of the invention, a dual-core coupling architecture of microscopic simulation and a macroscopic semantic agent forms a data-driven and knowledge-driven fused collaborative mechanism, so that the congestion is relieved and the traffic efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of urban traffic management, and in particular to a crowd passing management system and method and a computer device. BACKGROUND

[0002] Under the background of accelerating evolution of urban traffic system, efficient management of dense crowds in large traffic hubs (such as train stations and subway stations) has become a core problem that needs to be broken through in intelligent traffic systems. In particular, during high-pressure periods such as morning and evening peak hours, holidays or emergencies, a large number of passengers flow in a space-limited environment, which can easily cause serious congestion, stagnation and even stampede accidents, and seriously restrict the operation efficiency and service quality of the traffic system.

[0003] In related technologies, crowds are guided by static identification systems, broadcast notifications, or information display based on digital maps. However, in a high-density crowd environment, due to the lack of micro and macro linkage path planning mechanisms, it is difficult to coordinate individual behavior and overall flow in a high-density passenger environment in real time, which can easily lead to increased congestion and reduced passing efficiency. SUMMARY

[0004] Based on the above problems, the present application provides a crowd passing management system, method and computer device to alleviate congestion and improve passing efficiency.

[0005] The embodiments of the present application disclose the following technical solutions:

[0006] In a first aspect, the embodiments of the present application provide a crowd passing management system, comprising: a crowd simulator and a large language model (LLM) planning agent.

[0007] The crowd simulator is configured to simulate pedestrian motion behavior and group dynamics in a scene annotated with passing areas, obstacles, vertical traffic and function nodes, and output real-time data streams in response to a first decision scheme output by the LLM planning agent.

[0008] The LLM planning agent is configured to analyze real-time data streams through semantic understanding technology, and calculate multi-dimensional congestion indicators according to the analyzed real-time data streams; optimize the first decision scheme according to the multi-dimensional congestion indicators, output a second decision scheme, and send the second decision scheme to the crowd simulator.

[0009] In one possible implementation, the crowd simulator includes a potential-driven hybrid motion model.

[0010] The hybrid motion model is configured to simulate pedestrian motion behavior and group dynamics in a scene annotated with passing areas, obstacles, vertical traffic and function nodes, and output real-time data streams according to global navigation, local obstacle avoidance, vertical traffic processing and random disturbance behavior.

[0011] In a possible implementation, the hybrid motion model is configured to obtain a current position of the pedestrian; calculate a crowd density and a crowd velocity vector sum in a neighborhood within a local observation window centered on the current position; adjust a travel direction and a travel speed of the pedestrian according to the crowd density and the crowd velocity vector sum, and output a real-time data stream.

[0012] In a possible implementation, the multi-dimensional congestion indicator includes a congestion score, and the LLM planning agent includes a tool calling module;

[0013] The tool calling module is configured to call a congestion evaluation tool to obtain, for any origin-destination OD pair, average density indicator data, a non-zero density proportion indicator data, and a longest continuous congestion proportion indicator data about each path in the OD pair according to the real-time data stream and the historical trajectory data; and obtain, for any path, a congestion score of the path according to the average density indicator data, the non-zero density proportion indicator data, and the longest continuous congestion proportion indicator data corresponding to the path.

[0014] In a possible implementation, the multi-dimensional congestion indicator includes a congestion score, and the LLM planning agent includes a tool calling module;

[0015] The tool calling module is further configured to call the congestion evaluation tool to obtain, for any origin-destination OD pair, an average path length of the OD pair according to the historical trajectory data, and take the average path length as a congestion baseline; and obtain a congestion score of each path of the OD pair according to the congestion baseline.

[0016] In a possible implementation, the multi-dimensional congestion indicator includes real-time population density and unit time in-out flow, and the LLM planning agent includes a tool calling module;

[0017] The tool calling module is configured to call a space-time flow analysis tool to obtain, for a target region, real-time population density and unit time in-out flow about the target region.

[0018] In a possible implementation, the LLM planning agent includes a tool calling module;

[0019] The tool calling module is configured to call a recommendation module to calculate, for a corresponding origin-destination OD pair, a multi-dimensional congestion indicator according to departure location information of a target pedestrian, and generate a candidate path list according to the multi-dimensional congestion indicator data, and output a path planning and guidance suggestion.

[0020] In a possible implementation, the mixed motion module is configured to acquire a topological structure of the scene, and calculate a shortest path potential field of each main exit by using a Bellman-Ford algorithm; and guide the pedestrian to move along a gradient direction of the shortest path potential field, so as to realize global navigation.

[0021] In a second aspect, the embodiments of the present application provide a crowd passing management method, which comprises the following steps:

[0022] The crowd simulator simulates the motion behavior of the pedestrian and the group dynamics in the scene marked with the passing area, the obstacle, the vertical traffic and the functional node, and outputs a real-time data stream in response to the first decision scheme output by the LLM planning agent.

[0023] The LLM planning agent analyzes the real-time data stream by using a semantic understanding technology, calculates a multi-dimensional congestion index according to the analyzed real-time data stream, optimizes the first decision scheme according to the multi-dimensional congestion index, outputs a second decision scheme, and sends the second decision scheme to the crowd simulator.

[0024] In a third aspect, the embodiments of the present application provide a computer device, which comprises a memory and a processor.

[0025] The memory has a computer program stored thereon.

[0026] The processor is configured to execute the computer program in the memory, so as to implement the steps of the crowd passing management method in the second aspect.

[0027] In order to alleviate congestion and improve passing efficiency, the embodiments of the present application provide a crowd passing management system, which comprises a crowd simulator and an LLM planning agent of a large language model. The crowd simulator is configured to simulate the motion behavior of the pedestrian and the group dynamics in the scene marked with the passing area, the obstacle, the vertical traffic and the functional node, and output a real-time data stream in response to a first decision scheme output by the LLM planning agent. The LLM planning agent is configured to analyze the real-time data stream by using a semantic understanding technology, calculate a multi-dimensional congestion index according to the analyzed real-time data stream, optimize the first decision scheme according to the multi-dimensional congestion index, output a second decision scheme, and send the second decision scheme to the crowd simulator. In the embodiments of the present application, a dual-core coupling architecture of micro-simulation and macro-semantic agent is provided. By coupling the micro-granularity crowd behavior simulation and the macro-level semantic reasoning system, a collaborative mechanism of data-driven and knowledge-driven fusion is formed, a closed-loop management of passenger path optimization and transfer strategy generation is realized, the response capability and the real-time scheduling level of the system to the high-density crowd scene are greatly improved, and then the congestion is alleviated and the passing efficiency is improved. BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0029] Figure 1 A schematic diagram of a crowd passage management system provided by an embodiment of the present application;

[0030] Figure 2 A flowchart of a crowd passage management method provided by an embodiment of the present application;

[0031] Figure 3 A schematic diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0032] In order to make the technical personnel in the art better understand the present application, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0033] The terms "first" and "second" and the like in the specification and claims of the present application are used to distinguish different objects, not to describe a specific order of the objects. For example, the first switch and the second switch are used to distinguish different switches, not to describe a specific order of the switches.

[0034] In the embodiments of the present application, the words "exemplary" or "for example" are used to mean serving as an example, instance, or illustration. Any embodiment or design presented as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or advantageous than other embodiments or design solutions. Rather, the use of "exemplary" or "for example" is intended to present concepts in a concrete manner.

[0035] In the description of the embodiments of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more, for example, a plurality of processing units means two or more processing units, and the like; a plurality of elements means two or more elements, and the like.

[0036] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the following will introduce the crowd passage management system in conjunction with the drawings.

[0037] Referring toFigure 1 FIG. 1 is a schematic diagram of a crowd passage management system according to an embodiment of the present application.

[0038] As shown in Figure 1 The crowd passage management system includes a crowd simulator 100 and a large language model (LLM) planning agent 200.

[0039] The crowd simulator 100 is configured to simulate the motion behavior of pedestrians and the group dynamics in a scene annotated with passage areas, obstacles, vertical traffic, and functional nodes, and output real-time data streams in response to a first decision scheme output by the LLM planning agent 200.

[0040] In an embodiment of the present application, the crowd simulator can be a highly specialized and integrated simulation engine. It receives high-level decision instructions and drives a cognitive pedestrian model in a structured virtual scene, thereby generating realistic group behavior dynamics and providing continuous and reliable data feedback for upper-level intelligent decision-making.

[0041] The first decision scheme includes path optimization suggestions, transfer guidance strategies, and scheduling instructions, etc. The real-time data streams include pedestrian positions, densities, and speeds, etc.

[0042] In one possible implementation, the crowd simulator includes a potential-driven hybrid motion model. The hybrid motion model is configured to simulate the motion behavior of pedestrians and the group dynamics in a scene annotated with passage areas, obstacles, vertical traffic, and functional nodes, and output real-time data streams according to global navigation, local obstacle avoidance, vertical traffic processing, and random disturbance behavior.

[0043] The vertical traffic includes stairs, elevators, escalators, etc. The functional nodes include subway entrances, subway exits, stores, and online car pickup points, etc.

[0044] The global navigation, local obstacle avoidance, vertical traffic processing, and random disturbance behavior will be introduced respectively as follows.

[0045] Global navigation refers to the global path of pedestrians from the starting point to the target point. In an embodiment of the present application, the algorithm used for global navigation is not specifically limited, for example, the Bellman-Ford algorithm can be used to realize global navigation.

[0046] In one possible implementation, the hybrid motion module is configured to obtain the topological structure of the scene, calculate the shortest path potential field of each main exit through the Bellman-Ford algorithm, and guide the pedestrians to move along the gradient direction of the shortest path potential field to realize global navigation.

[0047] Exemplarily, nodes and edges are extracted from the scene annotation; the Bellman-Ford algorithm is used to calculate the shortest path from each node to each exit; the potential energy value of each node (the shortest distance from each node to the exit as the potential energy value of the node) is calculated, and thus the potential field of the entire scene is obtained; when a pedestrian is at a certain node, the pedestrian moves in the direction of the fastest potential energy decrease (i.e., the gradient direction). In the figure, the gradient direction can be understood as moving from the current node to the node with the smallest potential energy value among the adjacent nodes.

[0048] Local obstacle avoidance, that is, when a pedestrian moves along the global path, the pedestrian needs to avoid collision with obstacles and other pedestrians. In the embodiments, the algorithm or model used for local obstacle avoidance is not specifically limited, for example, a social force model (Social Force Model) or a method based on velocity obstacles (Velocity Obstacles), such as the RVO2 library, can be used to realize local obstacle avoidance.

[0049] Vertical traffic processing, that is, when a pedestrian needs to change floors, the pedestrian must use vertical traffic facilities (such as stairs or elevators). In the embodiments, these facilities can be modeled, in which, for the stairs, the pedestrian can move up and down, and needs to deal with the change of walking speed on the stairs (usually slower) and avoidance; for the elevator, the pedestrian needs to wait for the elevator, enter the elevator, be transported by the elevator, and then leave the elevator, which involves queuing and time delay.

[0050] Random disturbance behavior, that is, a pedestrian may randomly pause, slightly change direction, or temporarily change the target during walking (for example, be attracted by a store). In the embodiments, by introducing random disturbance behavior, the behavior dynamics of the crowd can be more accurately simulated.

[0051] Exemplarily, in combination with global navigation, local obstacle avoidance, vertical traffic processing, and random disturbance behavior, in a local observation window centered on the current position of the pedestrian, the crowd simulator dynamically calculates the crowd density, velocity vector, and sum of the neighborhood, and then adjusts the overall direction of travel and the step speed, so that the pedestrian actively avoids high-density areas while complying with the maximum speed limit (the default upper limit is 3 m / s) and physical constraints. In the floor conversion scenario, the crowd simulator simulates the traffic delay caused by the vertical traffic bottleneck, sets the stair waiting state (the default delay time is 5 seconds), and truly reflects the flow line conflict in the multi-layer traffic hub.

[0052] In addition, to improve the robustness of the model, the system introduces a random walk mechanism with a certain probability (default 10%) to help individuals escape from local potential field minimum value traps and avoid the simulation process from falling into a convergence dead zone. The simulation environment supports spatial modeling with a precision of meters. For example, in the application of Station A, the system constructs a high-precision indoor map with a resolution of 781x2748, accurately labeling passable areas, obstacles, elevators / escalators, and various functional nodes (such as subway exits, online car connection points, etc.), providing a solid foundation for precise navigation and behavior simulation.

[0053] The LLM planning agent 200 is configured to analyze real-time data streams through semantic understanding technology and calculate multi-dimensional congestion indicators based on the analyzed real-time data streams; optimize the first decision scheme based on the multi-dimensional congestion indicators, output the second decision scheme, and send the second decision scheme to the crowd simulator 100.

[0054] For example, to improve the accuracy and interpretability of the model in specific scenarios, the system designs and constructs a structured station knowledge graph (StationKG) in the LLM planning agent, which covers core entity types such as "origin-destination (OD) pairs", "travel trajectories", "areas of interest (AOI)", "density records", "speed tracking", and key semantic relationships such as "have", "pass", and "share". This structured station knowledge graph is constructed from simulation-enhanced historical trajectory data, forming a semantic foundation with retrievability and spatial logical constraints, and then accurately identifying information and data in real-time data streams so that the LLM can optimize decision schemes based on the identified data.

[0055] In addition, the structured station knowledge graph StationKG is integrated into the knowledge retrieval augmented generation (KG-RAG) framework, which can effectively alleviate the "hallucination" problem of large language models in spatial decision tasks and ensure that the recommended results conform to the spatial topological relationship and logical operation rules.

[0056] In one possible implementation, the multi-dimensional congestion indicators include congestion scores, and the LLM planning agent 200 includes a tool calling module; the tool calling module is configured to call a congestion evaluation tool to obtain, for any origin-destination (OD) pair, average density indicator data, non-zero density proportion indicator data, and longest continuous congestion proportion indicator data for each path in the OD pair based on real-time data streams and historical trajectory data; and for any path, obtain the congestion score of the path based on the average density indicator data, the non-zero density proportion indicator data, and the longest continuous congestion proportion indicator data corresponding to the path.

[0057] The average density index data reflects the average number of vehicles per unit road length of each path in the OD pair, directly reflecting the "crowded base" of path traffic (the higher the value, the more crowded the path as a whole during the statistical period, which is the core basic index for judging congestion or not); the non-zero density proportion index data reflects the proportion of time when each path in the OD pair has vehicle traffic, excluding the time period when there is no vehicle empty running (the higher the proportion, the more frequently the path is used, and the higher the probability of congestion; a low proportion means that the path is relatively empty and has a small risk of congestion); the longest continuous congestion proportion index data reflects the proportion of the longest continuous duration of each path in the OD pair in the congestion state during the statistical period (the higher the proportion, the longer the duration of congestion, and the longer the impact on traffic efficiency, which is a key indicator of the "stubbornness" of congestion).

[0058] For example, for an OD pair with a departure location A and a destination B, the OD pair includes paths a, b, and c. For path a, the average density index data, the non-zero density proportion index data, and the longest continuous congestion proportion index data of path a are calculated, and the congestion score of path a is obtained according to the average density index data, the non-zero density proportion index data, and the longest continuous congestion proportion index data of path a; for path b, the average density index data, the non-zero density proportion index data, and the longest continuous congestion proportion index data of path b are calculated, and the congestion score of path b is obtained according to the average density index data, the non-zero density proportion index data, and the longest continuous congestion proportion index data of path b; for path c, the average density index data, the non-zero density proportion index data, and the longest continuous congestion proportion index data of path c are calculated, and the congestion score of path c is obtained according to the average density index data, the non-zero density proportion index data, and the longest continuous congestion proportion index data of path c.

[0059] In one possible embodiment, the multi-dimensional congestion index includes a congestion score, and the LLM planning agent includes a tool calling module; the tool calling module is further configured to call a congestion evaluation tool, so that the congestion evaluation tool obtains the average path length of the OD pair according to the historical trajectory data for any origin-destination OD pair, and takes the average path length as a congestion baseline; and obtains the congestion score of each path of the OD pair according to the congestion baseline.

[0060] For example, all valid pass records of the target OD pair in the historical trajectory data are extracted; the average pass duration of all records is calculated, which is the congestion baseline of the target OD pair; the actual pass duration of each path to be evaluated (real-time data) or the predicted pass duration of the target period is extracted, and the congestion deviation is calculated according to the following formula: congestion deviation = (current path duration-congestion baseline) / congestion baseline x 100% (positive number indicates congestion, negative number indicates smoothness); congestion deviation ≤-20% (more than 20% faster than the baseline): 0-20 points (smoothness), -20% < congestion deviation ≤10% (close to the baseline or slightly slower): 21-40 points (basically smooth), 10% < congestion deviation ≤30% (10%-30% slower than the baseline): 41-70 points (mild to moderate congestion), and congestion deviation > 30% (more than 30% slower than the baseline): 71-100 points (severe congestion).

[0061] In a possible implementation, the multi-dimensional congestion index includes real-time population density and unit time access flow, and the LLM planning agent includes a tool calling module; the tool calling module is configured to call a space-time flow analysis tool, so that the space-time flow analysis tool analyzes the target area to obtain the real-time population density and the unit time access flow of the target area.

[0062] For example, the real-time population density of the target area is obtained according to the real-time population number of the target area and the area of the target area; the unit time access flow of the target area is obtained according to the number of people entering the target area, the number of people leaving the target area and the statistical duration, and is stored in a structured form of “timestamp-area identifier-real-time density-access flow” (supporting time series databases such as MySQL / InfluxDB).

[0063] The following several visualization forms are also provided in the embodiments of the application.

[0064] Trend chart: time axis + density / flow double-Y-axis line chart (supporting historical comparison, such as today vs. yesterday same period).

[0065] Heat map: target area space heat distribution (the deeper the color, the higher the density).

[0066] Data table: exportable time series data details (including time, density, flow, and year-on-year / compared-year change).

[0067] In addition, the space-time flow analysis tool can also be configured to automatically generate a “safety-efficiency” two-dimensional evaluation report, so as to identify potential high-risk or low-efficiency areas and provide data support for facility optimization and operation scheduling.

[0068] In a possible implementation, based on the multi-dimensional congestion index in the foregoing embodiments, the tool calling module in the LLM planning agent is configured to call the recommendation module, so that the recommendation module calculates the multi-dimensional congestion index for the corresponding OD pair according to the departure location information of the target pedestrian, generates a candidate path list according to the multi-dimensional congestion index data, and outputs a decision scheme.

[0069] It should be understood that the tool calling module in the embodiments of the present application can call the recommendation module in the initial stage to output an initial decision scheme, and call the recommendation module in the loop stage to output a first decision scheme, a second decision scheme, and the like.

[0070] For example, a new passenger entering the scene and the departure location of the passenger are identified, a multi-dimensional congestion index is calculated for the corresponding OD pair (preferably using D-Mode data), a candidate path list is generated according to a hybrid sorting strategy (combining congestion degree, average flow rate, path length, and the like), a decision scheme (optimal destination and path recommendation) is output, and structured support parameters and natural language explanations (for example, "recommended to go to the B exit, because the current density is 1.2 people / m², which is lower than the threshold value 2.5, and the average flow rate is 1.1 m / s") are generated. This mechanism significantly improves the transparency and user trust of the guidance strategy.

[0071] In the embodiments of the present application, a dual-core coupling architecture of micro-simulation and macro-semantic agent is adopted, the micro-granularity crowd behavior simulation and the macro-level semantic reasoning system are coupled, a collaborative mechanism of data-driven and knowledge-driven fusion is formed, the closed-loop management of passenger path optimization and transfer strategy generation is realized, the response capability and real-time scheduling level of the system to high-density crowd scenes are greatly improved, and then the congestion is relieved and the traffic efficiency is improved.

[0072] In addition, the crowd simulator in the embodiments of the present application is equivalent to a simulation verification module, which aims to build a strategy verification closed loop with "counterfactual evaluation" capability. The path recommendation generated by the LLM planning agent is fed back to the micro-simulation environment for behavior simulation to verify its feasibility and effect under the constraints of real space and group dynamics.

[0073] For example, after the LLM planning agent outputs the path recommendation and guidance recommendation, the system injects the path recommendation and the guidance recommendation as input parameters into the simulation environment, runs the micro-simulator to replay the path, observes key behavior indicators such as local density change, individual delay, and conflict frequency, and compares them with the current system default path to evaluate the improvement of the new strategy in terms of traffic efficiency, space load balancing, and safety. If the strategy performs poorly in the simulation, the agent will adjust the strategy parameters and generate them again until the performance constraints are met or the scheduling target is reached.

[0074] In the embodiments of the present application, the mechanism makes the system have verifiability and behavior level explainability, ensures that each recommended strategy has sufficient effectiveness and robustness in real scenarios, and provides low illusion, strong explainability and evidence-based auxiliary decision support for managers.

[0075] In combination with the foregoing embodiments, the present application adopts a multi-level path planning mechanism and a strategy verification closed loop design; the system combines a potential field driven hybrid motion model, a density perception navigation mechanism and a path recommendation method based on a knowledge graph to form a multi-level path planning system across granularities and spacetimes; meanwhile, a simulation verification module is introduced to realize counterfactual evaluation of the recommended strategy, effectively guaranteeing the real feasibility and behavior stability of the strategy, avoiding path deadlock, local congestion and model illusion problems, and improving the credibility and practicality of system decision-making.

[0076] Based on the crowd passage management system provided in the foregoing embodiments, the embodiments of the present application further provide a crowd passage management method.

[0077] Referring to Figure 2 , the figure is a flowchart of a crowd passage management method provided in the embodiments of the present application.

[0078] As Figure 2 shown, the method comprises the following steps:

[0079] S100: The crowd simulator outputs real-time data streams by simulating pedestrian motion behavior and group dynamics in a scene marked with passage areas, obstacles, vertical traffic and function nodes in response to the first decision scheme output by the LLM planning agent.

[0080] S200: The LLM planning agent analyzes the real-time data streams by semantic understanding technology, and calculates multi-dimensional congestion indicators according to the analyzed real-time data streams; the first decision scheme is optimized according to the multi-dimensional congestion indicators, a second decision scheme is output, and the second decision scheme is sent to the crowd simulator.

[0081] In the embodiments of the present application, a dual-core coupling architecture of micro-simulation and macro-semantic agent is adopted, the micro-granularity crowd behavior simulation and the macro-level semantic reasoning system are coupled, a data-driven and knowledge-driven fusion collaborative mechanism is formed, the closed loop management of passenger path optimization and transfer strategy generation is realized, the response capability and real-time scheduling level of the system to high-density crowd scenes are greatly improved, and then the congestion is relieved and the passage efficiency is improved.

[0082] In a possible implementation, in the scenario of marking the passing area, the obstacle, the vertical traffic and the functional node, the motion behavior and the group dynamics of the pedestrian are simulated according to the global navigation, the local obstacle avoidance, the vertical traffic processing and the random disturbance behavior, and the real-time data stream is output.

[0083] In a possible implementation, in the scenario of marking the passing area, the obstacle, the vertical traffic and the functional node, the motion behavior and the group dynamics of the pedestrian are simulated according to the global navigation, the local obstacle avoidance, the vertical traffic processing and the random disturbance behavior, and the real-time data stream is output, including: obtaining the current position of the pedestrian; calculating the crowd density, the crowd speed vector sum and the like in the neighborhood within the local observation window centered on the current position; adjusting the moving direction and the moving speed of the pedestrian according to the crowd density and the crowd speed vector sum, and outputting the real-time data stream.

[0084] In a possible implementation, the LLM planning agent parses the real-time data stream through the semantic understanding technology, and calculates the multi-dimensional congestion index according to the parsed real-time data stream, including: calling a congestion evaluation tool, so that the congestion evaluation tool obtains, for any origin-destination OD pair, average density index data, non-zero density proportion index data and longest continuous congestion proportion index data about each path in the OD pair according to the real-time data stream and historical trajectory data; for any path, the congestion score of the path is obtained according to the average density index data, the non-zero density proportion index data and the longest continuous congestion proportion index data corresponding to the path.

[0085] In a possible implementation, the LLM planning agent parses the real-time data stream through the semantic understanding technology, and calculates the multi-dimensional congestion index according to the parsed real-time data stream, including: calling a congestion evaluation tool, so that the congestion evaluation tool obtains, for any origin-destination OD pair, average density index data, non-zero density proportion index data and longest continuous congestion proportion index data about each path in the OD pair according to the real-time data stream and historical trajectory data; for any path, the congestion score of the path is obtained according to the average density index data, the non-zero density proportion index data and the longest continuous congestion proportion index data corresponding to the path.

[0086] In a possible implementation, the LLM planning agent parses the real-time data stream through the semantic understanding technology, and calculates the multi-dimensional congestion index according to the parsed real-time data stream, including: calling a congestion evaluation tool, so that the congestion evaluation tool obtains, for any origin-destination OD pair, average density index data, non-zero density proportion index data and longest continuous congestion proportion index data about each path in the OD pair according to the real-time data stream and historical trajectory data; for any path, the congestion score of the path is obtained according to the average density index data, the non-zero density proportion index data and the longest continuous congestion proportion index data corresponding to the path.

[0087] In a possible implementation, the method further comprises: calling the recommendation module, so that the recommendation module calculates a multi-dimensional congestion index for a corresponding origin-destination OD pair according to the departure location information of the target pedestrian, generates a candidate path list according to the multi-dimensional congestion index data, and outputs a path planning and guidance suggestion.

[0088] In a possible implementation, the global navigation is achieved in the following manner: the topological structure of the scene is acquired, and the shortest path potential field of each main exit is calculated through the Bellman-Ford algorithm; and the pedestrian is guided to move in the gradient direction of the shortest path potential field to achieve the global navigation.

[0089] In addition, the embodiment of the present application further provides a computer device, comprising: a memory and a processor;

[0090] The memory has a computer program stored thereon;

[0091] The processor is configured to execute the computer program in the memory to implement the steps of the crowd passage management method in any of the foregoing embodiments.

[0092] As shown in Figure 3 The computer device 12 is merely an example and should not impose any limitation on the functions and use range of the embodiment of the present application.

[0093] As shown in Figure 3 The computer device 12 is in the form of a general-purpose computing device. The components of the computer device 12 can include but are not limited to one or more processors or processing units 16, a system memory 28, and a bus 18 connecting different system components, including the system memory 28 and the processing unit 16.

[0094] The bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor or local bus using any of a variety of bus architectures. For example, these architectures include but are not limited to an industry standard architecture (ISA) bus, a microchannel architecture (MAC) bus, an enhanced ISA bus, a video electronics standards association (VESA) local bus, and a peripheral component interconnect (PCI) bus.

[0095] The computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0096] System memory 28 can include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. Computer device 12 can further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 can be provided for reading from and writing to a non-removable, non-volatile magnetic media (not shown and typically called a "hard drive"). Figure 3 Although not shown, computer device 12 can employ other, different or Figure 3 additional hardware and / or software modules that can be used in conjunction with the computer device 12, including but not limited to a microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc.

[0097] Program / utility 40, having a set (at least one) of program modules 42, can be stored in, for example, memory 28 by way of example, and not limiting, as suitable program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, each or some combination thereof,

[0098] Computer device 12 can also communicate with one or more external devices 14 such as a keyboard, a pointing device, a display 24, etc.; one or more devices that enable a user to interact with computer device 12; and / or one or more devices that enable computer device 12 to communicate with one or more other computing devices. Such communication can be via input / output (I / O) interfaces 22. Still yet, computer device 12 can communicate with one or more networks, such as a local area network (LAN), a wide area network (WAN), and / or the Internet, through network adapter 20. As Figure 3 illustrated, network adapter 20 can communicate with the other components of computer device 12 via bus 18. It should be understood that although not shown, other hardware and / or software components could be used in conjunction with computer device 12. Examples, include, but are not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc. Figure 3

[0099] ​The processor unit 16 performs various function applications and data processing by running programs stored in the system memory 28, such as implementing the crowd passage management provided by the embodiments of the present application.

[0100] The above merely provides a specific implementation of the present application, but the protection scope of the present application is not limited thereto, any changes or replacements within the technical scope disclosed by the present application can be easily thought of by those skilled in the art, and should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A crowd access management system, characterized in that, include: Crowd simulator and large language model LLM planning agent; The crowd simulator is configured to respond to the first decision scheme output by the LLM planning agent, and simulate pedestrian movement behavior and crowd dynamics in a scenario with marked passage areas, obstacles, vertical transportation and functional nodes, and output real-time data streams. The LLM planning agent is configured to parse the real-time data stream using semantic understanding technology and calculate multi-dimensional congestion indicators based on the parsed real-time data stream. The first decision scheme is optimized based on the multidimensional congestion index, a second decision scheme is output, and the second decision scheme is sent to the crowd simulator.

2. The crowd access management system according to claim 1, characterized in that, The crowd simulator includes a potential energy-driven hybrid motion model; The hybrid motion model is configured to simulate pedestrian movement behavior and group dynamics in a scenario with marked passage areas, obstacles, vertical traffic, and functional nodes, based on global navigation, local obstacle avoidance, vertical traffic processing, and random disturbance behavior, and output the real-time data stream.

3. The crowd access management system according to claim 2, characterized in that, A hybrid motion model is configured to acquire the current position of a pedestrian; calculate the crowd density and the sum of crowd velocity vectors within a local observation window centered on the current position; adjust the pedestrian's direction of travel and speed based on the crowd density and the sum of crowd velocity vectors, and output the real-time data stream.

4. The crowd access management system according to claim 1, characterized in that, The multidimensional congestion index includes a congestion score, and the LLM planning agent includes a tool invocation module; The tool invocation module is configured to invoke a congestion assessment tool to obtain, for any origin-destination (OD) pair, average density index data, non-zero density ratio index data, and longest continuous congestion ratio index data for each path in the OD pair based on the real-time data stream and historical trajectory data; and for any given path, to obtain a congestion score for that path based on the average density index data, non-zero density ratio index data, and longest continuous congestion ratio index data corresponding to that path.

5. The crowd access management system according to claim 1, characterized in that, The multidimensional congestion index includes a congestion score, and the LLM planning agent includes a tool invocation module; The tool invocation module is also configured to invoke a congestion assessment tool, so that the congestion assessment tool, for any origin-destination (OD) pair, obtains the average path length of the OD pair based on historical trajectory data, and uses the average path length as a congestion baseline; based on the congestion baseline, it obtains a congestion score for each path of the OD pair.

6. The crowd access management system according to claim 1, characterized in that, The multidimensional congestion indicators include real-time population density and inbound / outbound flow per unit time, and the LLM planning agent includes a tool invocation module; The tool invocation module is configured to invoke a spatiotemporal flow analysis tool so that the spatiotemporal flow analysis tool can analyze the target area and obtain the real-time population density and the inflow and outflow per unit time for the target area.

7. The crowd access management system according to claim 1, characterized in that, The LLM planning agent includes a tool invocation module; The tool invocation module is configured to invoke the recommendation module, so that the recommendation module calculates a multi-dimensional congestion index for the corresponding origin-destination OD pair based on the starting location information of the target pedestrian, generates a candidate route list based on the multi-dimensional congestion index data, and outputs a decision scheme.

8. The crowd access management system according to claim 2 or 3, characterized in that, The hybrid motion module is configured to acquire the topology of the scene, calculate the shortest path potential field of each main exit using the Bellman-Ford algorithm, and guide pedestrians to move along the gradient direction of the shortest path potential field to achieve global navigation.

9. A method for managing crowd access, characterized in that, The method includes: The crowd simulator responds to the first decision scheme output by the LLM planning agent, and simulates pedestrian movement behavior and crowd dynamics in a scenario with marked passage areas, obstacles, vertical transportation and functional nodes, and outputs real-time data streams. The LLM planning agent parses the real-time data stream using semantic understanding technology and calculates multi-dimensional congestion indicators based on the parsed real-time data stream; it optimizes the first decision scheme based on the multi-dimensional congestion indicators, outputs a second decision scheme, and sends the second decision scheme to the crowd simulator.

10. A computer device, characterized in that, include: Memory and processor; The memory stores computer programs thereon; The processor is configured to execute the computer program in the memory to implement the steps of the crowd access management method of claim 9.

Images