Method for identifying and early warning stability of crowd in urban rail transit station

Abstract

The invention belongs to the technical field of urban rail transit safety management, and particularly discloses an urban rail transit station crowd stability identification and early warning method, which comprises the following steps: acquiring personnel action data under various crowd flow movement scenes, quantifying conflict events among personnel based on a Worronoi diagram and a speed obstacle model, and obtaining quantitative characteristics; abstracting personnel as nodes, abstracting conflict events as directed edges, and constructing a time-varying conflict network; setting a percolation threshold value, and continuously analyzing a communication state and a conflict index set of the conflict network in a threshold value increasing process; generating a stability criterion according to the change trend of the connected subgraph along with the threshold value, judging the stability of the target scene in combination with the conflict index set, and outputting an early warning signal; the method has the following advantages that individual microscopic behaviors and system macroscopic steady-state characteristics can be described at the same time, multi-scene stability criteria are unified, steady-state real-time recognition and instability early warning of people in the rail transit station are achieved, and the in-station passenger flow safety management level can be improved.

Description

Technical Field

[0001] This invention relates to the field of urban rail transit safety management technology, and more specifically, to a method for identifying and warning of crowd stability within urban rail transit stations. Background Technology

[0002] As a crucial component of large-scale urban public transportation systems, urban rail transit stations typically feature high-density, multi-directional, and continuously flowing passenger traffic. With the continuous growth in passenger volume, accurately identifying and providing risk warnings for the stability of dense crowds within stations has become a critical issue in rail transit safety management. Disruptions to the stable state of the crowd within stations can easily trigger localized congestion, pushing, stampedes, and other mass safety incidents.

[0003] Existing methods generally rely on macroscopic average indicators or static characteristics, making it difficult to take into account the coupling relationship between microscopic individual behavior and macroscopic overall stability. This results in insufficient steady-state identification accuracy in complex station scenarios with multiple entrances and exits and multiple channels. Furthermore, existing simulation calculations based on dynamic models are complex and have low timeliness, making it difficult to meet the needs of real-time monitoring and rapid early warning. In addition, although complex network theory can characterize the structural features of population systems, there is still a lack of a quantitative evaluation framework for population stability that combines network permeation processes, making it impossible to form a unified stability criterion and risk classification standard in different scenarios.

[0004] To address this issue, a method for identifying and providing early warning of crowd stability within urban rail transit stations is proposed. Summary of the Invention

[0005] This invention aims to provide a method for identifying and warning about the stability of crowds in urban rail transit stations, in order to solve or improve the problems mentioned above, such as the difficulty of existing methods for analyzing the steady-state of crowds in urban rail transit stations in simultaneously taking into account micro-level individual interactions and macro-level overall structure, the lack of a unified framework for quantitative evaluation of stability, and insufficient real-time performance.

[0006] In view of this, the first aspect of the present invention is to provide a method for identifying and warning of crowd stability in urban rail transit stations.

[0007] The first aspect of the present invention provides a method for identifying and warning about crowd stability in urban rail transit stations, comprising the following steps: acquiring human movement data under various human movement scenarios; quantifying conflict events between each human and its neighbors and obtaining quantified features based on the human movement scenarios and the movement data; treating human beings as nodes and generating directed edges between nodes based on the conflict events; assigning edge weights to each directed edge using the quantified features to construct a conflict network; setting an adjustable seepage threshold and filtering directed edges in the conflict network using the seepage threshold and the edge weights; continuously acquiring the connectivity state and conflict index set of the conflict network as the seepage threshold increases; acquiring the changing states of various connected subgraphs in the conflict network based on the changes in connectivity state with the seepage threshold, and generating a stability criterion for the conflict network under all human movement scenarios based on the changing states; judging the stability of a target human movement scenario using the stability criterion and the conflict index set, and selecting whether to generate an alarm based on the judgment result.

[0008] The beneficial effects of this invention compared to the prior art are as follows: By collecting data on human movement in various scenarios and constructing quantitative features based on spatial and speed conflicts between neighbors, this method can simultaneously reflect local interactions at the individual level and the macroscopic evolution of the overall system. This overcomes the shortcomings of existing methods that rely solely on macroscopic average indicators and cannot accurately characterize microscopic interactions, making the results of crowd steady-state identification more physically meaningful and behaviorally interpretable.

[0009] By abstracting individuals as nodes and conflict events as directed edges, and using conflict intensity as edge weights, a time-varying conflict network is constructed, achieving a dynamic networked representation of the population state. Compared to traditional models based on fixed structures or static indicators, this method can reflect changes in conflict relationships within the population in real time, making stability analysis both temporally continuous and structurally sensitive, providing more accurate basic data for subsequent risk identification.

[0010] By introducing an adjustable infiltration threshold mechanism and continuously acquiring the connectivity state and conflict index set of the conflict network during the threshold increment process, it is possible to simulate the gradual diffusion and weakening process of conflict in the network, revealing the evolution law of the connectivity structure of the population system under different perturbation intensities from a macroscopic level. This process can not only identify the critical point of network collapse, but also capture potential instability signs before changes in connectivity state, achieving early identification of instability precursors.

[0011] By monitoring the changing trend of connected subgraph size with seepage threshold, and utilizing the abrupt change behavior of the largest and second-largest connected subgraphs to generate stability criteria, a unified quantification of stability across scenarios is achieved. Compared to existing schemes that can only evaluate stability in a single scenario, the criteria of this invention have universality and comparability, maintaining consistent judgment standards under different station, layout, and passenger flow density conditions, providing a unified benchmark framework for multi-scenario steady-state assessment within urban rail transit stations. Additional aspects and advantages of embodiments of the invention will become apparent in the following description or may be learned by practice of embodiments of the invention. Attached Figure Description

[0012] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart of the method steps of the present invention; Figure 2 These are schematic diagrams illustrating six typical scenarios of the present invention; Figure 3 Analysis of the basic topological characteristics of dynamic pedestrian conflict networks in six typical scenarios of this invention; Figure 4 Radar charts showing the direction and intensity of crowd conflict in six typical scenarios according to this invention; Figure 5 This invention provides an analysis of the seepage process in a unidirectional flow scenario from the perspective of strong connectivity. Figure 6 This invention presents the key seepage threshold changes in six different scenarios from the perspective of strong and weak connectivity. Detailed Implementation

[0013] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0014] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0015] Please see Figures 1-6 The following describes a method for identifying and warning about crowd stability in urban rail transit stations according to some embodiments of the present invention.

[0016] An embodiment of the first aspect of the present invention proposes a method for identifying and warning of crowd stability within urban rail transit stations. In some embodiments of the present invention, such as... Figures 1-6 As shown, the method includes the following steps: S101: Acquire action data of people in various pedestrian movement scenarios, and quantify the conflict events between each person and their neighbors and obtain quantitative features based on the pedestrian movement scenarios and action data.

[0017] First, it is necessary to identify various pedestrian movement scenarios based on different spatial forms and passenger flow organization methods within urban rail transit stations. Specifically, these can include one-way movement scenarios after passengers enter the station hall through turnstiles, opposing convergence scenarios between the station hall and platform, intersection scenarios where multiple directions converge simultaneously, congestion scenarios where passengers pass through narrow bottleneck sections, and circular flow scenarios in the platform return area. Each scenario has a different spatial layout and direction of travel, which affects the interaction characteristics and conflict patterns between individuals. By collecting pedestrian trajectory data in different scenarios, it is possible to comprehensively reflect the differences in spatial distribution, flow convergence, density changes, and speed coupling of the crowd.

[0018] Furthermore, using a passenger flow monitoring system or computer vision algorithms, the spatial coordinates and velocity vectors of each pedestrian at each time step are extracted from continuous video images. The position coordinates reflect the spatial distribution of pedestrians, while the velocity vectors characterize the direction and amplitude of their movement. Based on this data, at each time step, a set of neighbors directly interacting with any target pedestrian is determined. This neighbor set can be implemented using a Volonoi graph partition, where Volonoi polygons are constructed centered on each pedestrian. When two pedestrian polygons share a common boundary, they are considered spatially adjacent, thus defining a spatial neighbor relationship. The geometric properties of the Volonoi partition ensure that at any given time, the neighbor relationships of each pedestrian are unique and time-varying, dynamically reflecting changes in their relative positions within a local space.

[0019] After determining the neighbor set, the conflict between any target pedestrian and its neighbor pair is quantified. This is done by characterizing spatial conflict and velocity conflict separately. For spatial conflict, a shortest distance function from the pedestrian to the polygon boundary of the neighbor is defined, reflecting the minimum safe distance between individuals. When the distance decreases to near zero, it indicates that the activity spaces of the two overlap, posing a potential collision risk; the conflict intensity increases as the distance decreases. For velocity conflict, velocity barrier theory is used for modeling. A velocity barrier cone region is constructed based on the relative speed and position of the pedestrian and its neighbors, with the pedestrian's speed as the reference. When the pedestrian's velocity vector falls inside this cone region, it indicates that maintaining the current speed may result in a collision in the future, defined as a velocity conflict; conversely, when the velocity vector is outside the cone, the current motion state is considered safe, with no risk of velocity conflict.

[0020] As described above, the degree of conflict between a pedestrian and their neighbors is comprehensively calculated by combining the two conflict types mentioned. If both spatial overlap and velocity convergence exist simultaneously, the conflict quantification value for that neighbor pair will be negative, and its absolute value represents the minimum velocity change the pedestrian needs to adjust to avoid potential collisions. When there are no spatial or velocity conflicts, the conflict quantification value is positive, and its absolute value reflects the maximum acceleration space achievable by the pedestrian in the current environment. In this way, each pedestrian corresponds to a set of neighbor conflict quantification features at each time step, forming a dynamic conflict dataset for the entire scenario. This dataset reflects both individual-level micro-interaction behaviors and provides a quantitative basis for subsequent conflict network construction and crowd stability assessment.

[0021] Furthermore, the categories of pedestrian movement scenarios are set by variables such as the number of pedestrian flow directions, the number of open intersections, and the flow width in the scenario.

[0022] As described above, within the monitoring range of the same scenario, the main traffic directions that occur simultaneously and are distinguishable are counted based on the actual speed and direction of personnel passage. If the angle between directions is less than a preset angle threshold (e.g., 15°), they are considered to be counted together as the same direction. Only directions that account for a percentage of the preset threshold within a continuous time window are counted to eliminate occasional minor flow interference. Under the boundary conditions of the same scenario, openings that passengers can actually enter and exit are identified, and only openings that are open and accessible within the monitoring period are counted. Openings that are temporarily closed or restricted in one direction are counted according to the actual accessible direction. The effective width of passage within the monitoring range is quantified by sampling the width along the direction perpendicular to the main traffic direction. For areas with fixed obstacles, variable obstacles, and boundary contraction, the minimum effective width principle is adopted, and representative values ​​within the time window are recorded and solidified into a flow width variable of the same unit scale.

[0023] Specifically, the action data includes the location and speed of individuals at each moment, as well as the steps to quantify conflict events between each individual and their neighbors and obtain quantified characteristics, including: Based on location, the Volonoi map is used to determine the spatial conflict of each person within the Volonoi neighborhood.

[0024] Based on speed, a speed obstacle is used to determine the speed conflict of each person within the speed obstacle cone.

[0025] The quantitative values ​​of conflict events between each person and their neighbors within the Volonoi neighborhood are calculated using spatial conflict and velocity conflict methods.

[0026] Based on the specific description above, at each time point, the position of each person is represented by a unified coordinate system, recording the instantaneous position of all persons using the same coordinate system. A Volonoi map is constructed using all position points at that time as input. In this Volonoi map, each person corresponds to a unique Volonoi polygon, and the boundary shared with adjacent polygons is defined as the spatial adjacency relationship of the Volonoi neighborhood. During implementation, for any person, all neighbors within their Volonoi neighborhood are enumerated, and a spatial conflict metric is calculated based on geometric quantities such as the shortest distance from the person to the shared boundary and the proportion of that boundary's length at the current time. Smaller distances and tighter boundaries indicate lower local spatial availability and stronger potential obstacles. To ensure comparability and verifiability, all spatial conflict metrics use the same calculation rules and dimensions at the same time point, and are identified and archived using a ternary field of person, neighbor, and time, along with the Volonoi neighborhood identifier. Through this process, the spatial conflict of each person within their Volonoi neighborhood is determined using the Volonoi map based on their position, achieving datafication, standardization, and solidification.

[0027] At any given moment, the velocity of a person is vectorized and used in conjunction with the velocity and relative position of each neighbor in the Volonoi neighborhood to construct a velocity barrier cone with the person as a reference. During implementation, the person's current velocity vector is compared to the boundary of the velocity barrier cone: if the velocity vector falls inside the cone, a velocity conflict is considered; if it lies outside, no conflict is considered. To quantify the strength of the conflict, the minimum velocity change to the cone edge or an equivalent safety margin can be calculated under the same discrimination rule, representing the person's adjustable velocity space without triggering a collision. All velocity conflict determinations and their quantification results are recorded along with the velocity barrier cone's geometric parameters using a ternary field of person, neighbor, and time, ensuring consistent thresholds and repeatability in the process of determining velocity conflicts for each person within the cone based on velocity.

[0028] After calculating spatial and velocity conflicts at the same time, the two types of conflict information are merged within the same recording unit for personnel, neighbors, and time, using the Volonoi neighborhood as the adjacency range. During implementation, spatial and velocity conflicts are mapped to a single quantified value according to unified rules. This quantified value directly characterizes the strength and directionality of the conflict event between each person and their neighbor within the Volonoi neighborhood: when the quantified value is negative, its absolute value represents the minimum velocity adjustment required to avoid conflict; when the quantified value is positive, its absolute value represents the maximum allowable acceleration without triggering conflict. To ensure comparability under different personnel densities and different pedestrian flow conditions, the quantified values ​​use the same scaling specifications and notation conventions at the same time and are archived as a continuous time series, thus forming a dynamic trajectory of quantified characteristics on the time axis. Through this process, the quantified values ​​of the conflict events between each person and their neighbor within the Volonoi neighborhood, calculated through spatial and velocity conflicts, are solidified into input data that can be directly used for subsequent steps, maintaining a one-to-one correspondence with the Volonoi neighborhood and strictly aligning with the velocity obstacle determination results.

[0029] S102, personnel are treated as nodes, and directed edges between nodes are generated based on conflict events; the edge weight of each directed edge is assigned by quantization features to construct a conflict network.

[0030] Here, in this step, the quantitative characteristics of pedestrian conflicts obtained in step S101 are transformed from individual behavioral data into structured network data, thereby establishing a conflict network model that can dynamically reflect the overall stability of the crowd. This model uses the interactive conflict relationships between pedestrians as its core logic, abstracting the crowd system into a time-varying directed weighted network, providing a foundation for subsequent stability assessment and seepage analysis.

[0031] First, at each moment, all pedestrians within the rail transit station are considered nodes in the network. Each node represents an independent individual, and its state is defined by its position, velocity, direction, and conflict quantification characteristics at the current moment. The interconnections between nodes are determined based on neighbor conflict events. When two pedestrians are in spatially adjacent Volonoi neighborhoods and there is a spatial or velocity conflict between them, the system establishes an edge between the two nodes to indicate that there is a potential mutual interference relationship between them at that moment.

[0032] Secondly, to reflect the directionality of conflict between individuals, the edges established in the network are defined as directed edges, with their direction pointing from the intruding party to the intruded party. For example, when pedestrian A's movement direction obstructs pedestrian B's movement path, the system will establish a directed edge between node A and node B, pointing from A to B. This directional setting can effectively characterize the propagation path and mutual influence patterns of conflict within the group, thereby revealing potential local bottlenecks and behavioral chains in crowd movement.

[0033] After determining the network structure, each directed edge needs to be weighted according to the conflict quantification characteristics. The size of the edge weight represents the conflict intensity under that connection, and its value is directly derived from the neighbor conflict quantification value calculated in step S101. The smaller the edge weight value, the more serious the conflict between the pedestrian and its neighbor, that is, the more significant the mutual interference between them; conversely, the larger the edge weight value, the more stable the connection relationship is, the weaker the interaction between individuals, and the ability to maintain one's own speed and direction without affecting the movement of others.

[0034] To facilitate calculation and comparison, this invention performs translation and normalization on all edge weights when constructing the conflict network, ensuring their values ​​are distributed within a non-negative range, thus avoiding computational ambiguity caused by negative values. Specifically, the conflict quantification values ​​are shifted as a whole to the range above zero while maintaining their relative magnitude. This preserves the relative differences in conflict intensity and makes the network structure more consistent and stable in mathematical processing.

[0035] As described above, after the above processing, a complete directed weighted conflict network can be formed at any given time. The total number of nodes in this network corresponds to the number of pedestrians detected in the current scene, while the number and weight of edges depend on the scene density, the degree of directional convergence, and the intensity of interaction between pedestrians. Over time, a new conflict network is generated at each time step, and the network sequence at all times constitutes the temporal evolution process of the dynamic conflict network. By continuously tracking the changes in this network, the evolutionary characteristics of the overall crowd structure can be intuitively reflected: in a stable state, the edge weights in the network are generally large, the structural connectivity is high, and there are few local conflicts; while in a state where the crowd gradually becomes unstable, the edge weights in some areas begin to decrease, and local dense connections form high-conflict clusters, ultimately leading to the differentiation of the overall connectivity structure.

[0036] Furthermore, the representational meaning of quantified features is defined by the following rules: When the quantization feature is negative, the absolute value represents the minimum speed adjustment required to avoid conflict.

[0037] When the quantization feature is positive, the absolute value represents the maximum acceleration allowed without causing a conflict.

[0038] As can be seen from the above, when the quantization feature is negative, the absolute value represents the minimum speed adjustment required to avoid conflict.

[0039] In this scenario, the negative sign of the quantization feature indicates that if the current motion state remains unchanged, a conflict event will be triggered, requiring proactive correction at the velocity level. During implementation, the relevant personnel read the absolute value of this negative value as the nominal size of the minimum velocity adjustment. This size is used to constrain one-time or step-by-step velocity corrections, ensuring the correction magnitude is not less than this absolute value, thereby eliminating the impending conflict at minimal cost. To ensure the correction is directional, the minimum velocity adjustment is decomposed into the normal or tangential directions relative to the neighboring motion in actual calculations, achieving the most direct and effective resolution of the conflict source. After execution, the quantization feature should be recalculated immediately at the next time step. If it is still negative, iterative correction continues based on the new absolute value until the quantization feature is no longer negative. This process ensures conflict resolution with the minimum necessary adjustment, avoiding secondary disturbances caused by over-correction.

[0040] When the quantization feature is positive, the absolute value represents the maximum acceleration allowed without causing a conflict.

[0041] In this scenario, a positive sign for the quantization characteristic indicates that the current motion state is within a conflict-free zone, allowing for acceleration optimization without triggering conflict events. During implementation, the relevant personnel read the absolute value of this positive sign as the upper limit of the maximum acceleration. This upper limit constrains the velocity increase at this moment and even within a short time series, ensuring that the accelerated motion state remains within a safe zone free from conflict. To ensure a safety margin, the maximum acceleration can be applied directly to the current velocity direction, or deflected at a small angle before acceleration, to adapt to continuous changes in local traffic conditions. After execution, the quantization characteristic is recalculated at the next time step: if it remains positive, it indicates that the accelerated state is still within a safe zone, and acceleration can continue as long as it does not exceed the new absolute value upper limit; if it turns negative, the minimum velocity adjustment is immediately performed according to the aforementioned rules, prioritizing the elimination of potential conflicts.

[0042] S103, set an adjustable percolation threshold, filter directed edges in the conflict network by the percolation threshold and edge weight; continuously obtain the connectivity status of the conflict network and the conflict index set as the percolation threshold increases.

[0043] Here, this step involves introducing a seepage threshold control parameter to perform structural screening and hierarchical analysis on the directed weighted conflict network constructed in the previous step S102. This aims to characterize the connectivity changes of the network under different conflict intensity constraints, and thereby extract a set of key indicators reflecting overall stability. Essentially, this process, from a network science perspective, simulates a dynamic evolution of the population system, similar to a gradual weakening of connection strength, to reveal the potential critical state of the population structure from stability to instability.

[0044] First, in the initial state, the distribution range of all edge weights is determined based on the directed weighted conflict network generated in the previous stage. Since the edge weights represent the degree of conflict between pedestrians, there will be both high-weight weak conflict edges and low-weight strong conflict edges in the overall network. To perform hierarchical analysis of conflict relationships of different intensities, an adjustable percolation threshold is set. This percolation threshold serves as a boundary condition for filtering network connections, and its physical meaning can be understood as the maximum conflict intensity that the network can tolerate. When the edge weight of an edge is less than or equal to the percolation threshold, it indicates that the degree of conflict in the connection relationship exceeds the system's allowable range, and the edge is considered invalid in the overall structure and is deleted; conversely, when the edge weight is greater than the percolation threshold, it indicates that the connection relationship can still maintain a relatively stable interaction and should be retained in the network.

[0045] Next, the percolation threshold is incrementally adjusted within a preset numerical range. Typically, it starts with a low initial value and gradually increases in fixed steps. Each increase in the percolation threshold signifies a reduction in the system's tolerance for conflict, i.e., a reassessment of the network's connectivity under stricter stability conditions. As the percolation threshold continuously increases, low-weight edges in the network are gradually eliminated, the connections between nodes decrease, and the originally tightly connected overall structure gradually decomposes into multiple independent connected subgraphs.

[0046] Under each seepage threshold and corresponding screening state, the connectivity parameters of the conflict network are calculated in real time. The connectivity state reflects whether the overall network structure still possesses global connectivity, that is, whether there are still reachable paths between nodes under the current conflict intensity constraint. If a large-scale connected group remains in the network, it indicates that the overall coordination of the group structure is relatively high and it is still in a stable stage; conversely, if the network breaks significantly with a slight increase in the threshold, it indicates that the system structure is prone to collapse under external disturbances and is on the verge of potential instability.

[0047] Furthermore, to more accurately describe the internal characteristics of the network during the seepage process, a conflict index set is calculated at each threshold. The conflict index set consists of multiple statistics reflecting the state of the population, including but not limited to the number of connected subgraphs, the distribution of node sizes within each connected subgraph, the average degree, the clustering coefficient, and the rate of change in edge weight distribution. The conflict index can characterize the complexity of internal conflicts within the system from multiple dimensions: the number of connected subgraphs reflects the degree of group differentiation, the node size distribution describes the degree of local congestion, the clustering coefficient measures the tightness of connections between neighborhoods, and the rate of change in edge weight distribution reveals the evolution of conflict propagation speed and intensity.

[0048] By continuously monitoring the changing trends of these indicators as the seepage threshold increases, a set of time-series data describing the evolution of the network structure can be generated. If, within a certain threshold range, the network's connectivity or clustering characteristics exhibit significant abrupt changes—for example, a sharp decrease in the size of the largest connected subgraph or a sudden drop in the average clustering coefficient—it can be considered a key indicator of the system transitioning from a stable to an unstable state. Conversely, when the indicators change gradually and the connectivity structure maintains strong integrity, it indicates that the population is still in a coordinated and stable state of motion.

[0049] As described above, by introducing a seepage threshold into the conflict network and continuously increasing and adjusting it, a step-by-step peeling analysis of the structural robustness of the population system was achieved. This process can not only identify the connectivity evolution characteristics of the network at different conflict levels, but also extract core indicators reflecting the changing trends of system stability, providing a quantitative basis for subsequent extraction of key seepage thresholds and generation of population steady-state criteria.

[0050] Specifically, the steps for continuously acquiring the connectivity status of the conflict network and the conflict index set during the increasing seepage threshold include: The seepage threshold is continuously increased in fixed steps within a preset range.

[0051] During the scaling process, the size of the connected subgraph is obtained when directed edges are deleted from the conflict network, and the severity of conflict events in the conflict network is calculated.

[0052] At each step of the seepage threshold, a mapping relationship between the connected subgraph and the conflict severity is generated, and a conflict index set is generated based on the mapping relationship and the conflict severity.

[0053] Regarding the specific description above, before starting the calculation, a preset range and fixed step size for the seepage threshold are first set. The preset range is used to limit the value interval of the seepage threshold from the initial value to the final value, and the fixed step size is used to limit the increment of each threshold update. Subsequently, at the same time in the same conflict network, the seepage threshold is incrementally updated with a fixed step size. Each time it is incremented, the directed edges in the conflict network are immediately filtered based on the new seepage threshold to ensure that the seepage threshold, the filtering result, and the subsequently recorded data correspond one-to-one, forming a strict sequence of seepage thresholds and filtering states. The connectivity state of the conflict network under different seepage thresholds is expanded into a comparable sequence on a uniform scale, providing a basic serialized input for the subsequent generation of mapping relationships and the formation of conflict index sets.

[0054] After each increase in the seepage threshold and completion of the directed edge filtering, the directed edges deleted due to the seepage threshold change are immediately detected. Whenever a directed edge is deleted from the conflict network, the connectivity state at that moment is updated in real time, and the connected subgraphs in the current conflict network are identified and statistically analyzed to obtain the size of the connected subgraph. The size of the connected subgraph is measured by the number of nodes contained within it to ensure consistency between the size definition and subsequent trend analysis. Furthermore, within the same calculation cycle synchronized with the acquisition of the connected subgraph size, the conflict severity of conflict events in the conflict network is calculated. The conflict severity is measured by the edge weight distribution of the currently filtered directed edges and its changes caused by the seepage threshold variation. That is, under the same seepage threshold, the strength level of the conflict event at that moment is calculated based on the edge weights of the retained directed edges to reflect the intensity of the population conflict state under the current threshold condition. The function of this action is to simultaneously obtain synchronous observations of the size of the connected subgraph and the conflict severity at each real-time node where a directed edge is deleted, ensuring that structural information and conflict information are updated synchronously and correspond to each other.

[0055] After synchronously acquiring the size and conflict severity of the connected subgraphs, a mapping relationship between connected subgraphs and conflict severity is generated within the same recording unit corresponding to the same seepage threshold. This mapping relationship uses the current seepage threshold as an index to record the size of each connected subgraph and its corresponding conflict severity at that threshold, thus establishing a one-to-one correspondence between structural features and conflict intensity in the same threshold coordinate system. Subsequently, using this mapping relationship as input, combined with the conflict severity calculated at the same threshold, a conflict index set is generated. The conflict index set forms a set of entries at each step of the seepage threshold, used to continuously characterize the connectivity changes and conflict intensity changes of the conflict network as the threshold increases. The function of this action is to fix the structural information and intensity information in a mapping relationship and generate a matching conflict index set at each step, thereby obtaining serialized index data that can be directly used for subsequent judgment and comparison.

[0056] As described above, by executing the three actions sequentially, continuous observation and recording of the same conflict network during the increasing percolation threshold process can be achieved without changing the meaning of the given feature words. Each percolation threshold step corresponds to a specific filtering state, a specific set of connected subgraph sizes, a specific conflict severity value, and a set of conflict index entries generated from the mapping relationship between the two. The multidimensional sequence of percolation threshold, connected state, and conflict index set ensures both the consistency of data source and operation order, as well as the comparability between different step lengths. This provides a direct, sufficient, and verifiable quantitative basis for subsequently obtaining the changing states of various connected subgraphs based on the changes in connected state with the percolation threshold, and for generating stability criteria for the conflict network in all traffic movement scenarios based on the changing states.

[0057] Furthermore, the scale includes the types of connected subgraphs and the number of nodes contained within each connected subgraph.

[0058] The types are classified by the connection state between nodes within a connected subgraph via directed edges.

[0059] As described above, under each seepage threshold and corresponding filtering state, all connected subgraphs in the current conflict network are first identified. Then, two pieces of information are recorded for each connected subgraph: the type of the connected subgraph and the number of nodes it contains. The number of nodes within a connected subgraph is counted using the total number of nodes in that subgraph, ensuring direct comparability of the sizes of different connected subgraphs under the same seepage threshold. Furthermore, to guarantee comparability between different seepage threshold steps, this count is acquired and stored using the same measurement method at each step. In implementation, a record can be created for each connected subgraph while generating the list of connected subgraphs. This record contains at least two fields: type and number of nodes. The number of nodes is a non-negative integer representing the size of the connected subgraph. These two fields together constitute a complete definition of the size, used for subsequent sorting, trend tracking, and mapping relationship generation.

[0060] To distinguish connected subgraphs with different structural features, the types are classified and labeled based on the connection status between nodes within the connected subgraph via directed edges. During implementation, the reachability of any two nodes within the same connected subgraph is determined: If there is a mutually reachable path between any two nodes within the connected subgraph, that is, the two nodes can reach each other through a sequence of directed edges, then the connected subgraph is classified into a category according to this connection state. If a connected subgraph remains connected after ignoring the direction of the directed edges, but there is no mutual reachability between any two nodes when the direction of the directed edges is retained, then the connected subgraph is classified into another type according to this connection state.

[0061] The above classification is based entirely on the connection state between nodes within a connected subgraph via directed edges, without introducing other criteria, ensuring that the definition of the categories is strictly consistent with the feature terms of this invention. During implementation, to ensure consistency in judgment, the following operations can be performed sequentially at each step's percolation threshold: On the currently selected conflict network, identify connected subgraphs and generate a set of connected subgraph nodes; Within the node set of each connected subgraph, reachability is checked according to the connection state through directed edges, and the type of the connected subgraph is determined accordingly. Count the number of nodes contained within the connected subgraph and output the count as the node count field. The size record of the connected subgraph is fixed in the form of a binary tuple of type and number of nodes.

[0062] Furthermore, to facilitate subsequent sorting and trend analysis, under the same seepage threshold, nodes can first be grouped by type, and then within each type, they can be arranged from largest to smallest according to the number of nodes contained in the connected subgraph, forming an ordered scale sequence at that step length. In the comparison across step lengths, the above grouping and sorting rules remain unchanged, thus directly obtaining the quantitative trend of the scale of the same type and the same arrangement position as the seepage threshold changes. The processing of isolated nodes also follows the same rules: when a node is not connected to any other node through directed edges under the current seepage threshold, the node constitutes a connected subgraph on its own, its node count is recorded as 1, and its type is still labeled and archived according to the rules for classifying by the connection state through directed edges.

[0063] S104. Based on the change of connectivity state with the seepage threshold, obtain the changing states of multiple connected subgraphs in the conflict network, and generate a stability criterion for the conflict network under all traffic flow scenarios through the changing states.

[0064] Here, based on the connectivity state and conflict index set of the conflict network obtained in step S103, the structural change characteristics of the conflict network during the increasing seepage threshold are analyzed. This allows for the identification of the evolutionary patterns of different types of connected subgraphs within the network, and based on these patterns, a stability criterion is generated that can uniformly characterize the overall stability under various pedestrian movement scenarios. This process, through the decomposition and trend tracking of the network structure, achieves a cross-scale mapping from local conflict characteristics to the macroscopic stability of the system, providing a theoretical basis for subsequent crowd stability identification and early warning.

[0065] First, in the continuous process of increasing seepage threshold, each threshold corresponds to a screening state of the conflict network. By repeatedly performing the screening operation at different thresholds, a series of conflict networks with different connectivity structures can be obtained. At this time, for each conflict network at each threshold, the system calculates the type and size of its connected subgraph. A connected subgraph is a substructure composed of mutually reachable nodes in the network, which can reflect the scope and intensity of local collaborative behavior within the population. This invention classifies connected subgraphs into two types based on the connection relationship of directed edges: strongly connected subgraphs and weakly connected subgraphs. A strongly connected subgraph indicates that there is a reachable path between any two nodes, reflecting the bidirectional and closed nature of interactions between nodes; a weakly connected subgraph maintains connectivity while ignoring the direction of directed edges, reflecting the macroscopic aggregation characteristics of the overall structure.

[0066] In the conflict network corresponding to each threshold, the number of internal nodes of the two types of connected subgraphs mentioned above is counted, and all subgraphs are sorted according to the number of nodes, forming a permutation sequence from largest to smallest. This sorting process can intuitively depict the formation and disappearance rules of connected clusters of different sizes: at low thresholds, the network is densely connected, with a large number of nodes concentrated in a single large-scale connected subgraph, and the system as a whole exhibits high connectivity; as the threshold increases, some connecting edges are removed, and the original overall structure gradually splits into multiple medium-sized subgraphs, and the node distribution tends to be discrete; when the threshold increases further, only a few small subgraphs or isolated nodes are retained, and the system structure exhibits highly fragmented characteristics.

[0067] During continuous threshold changes, this invention synchronously tracks the size change trends of strongly connected and weakly connected subgraphs. For each type of subgraph, the node number change curves of the largest and second-largest connected subgraphs are extracted. By comparing the morphological characteristics of these two curves during the threshold increase process, key stages of system structure change can be identified. When the number of nodes in the largest connected subgraph rapidly decreases, it indicates that the main structure that originally maintained the overall stability of the system has been destroyed; while when the number of nodes in the second-largest connected subgraph reaches a peak, it indicates that the original large cluster structure has been split into multiple relatively independent local clusters, and the network as a whole has entered a critical state of structural instability. These two types of change characteristics are considered typical signals of the transition from steady state to instability in percolation theory, and therefore can serve as important bases for identifying stability boundaries.

[0068] Based on the above analysis results, the system quantifies the changing states of strongly connected and weakly connected subgraphs in each scenario, extracts the structural inflection points corresponding to key seepage thresholds, and defines these thresholds as the stability critical values ​​for that scenario. To enable cross-scenario comparisons, this invention normalizes the key seepage thresholds obtained from different scenarios, forming a universal stability criterion applicable to all pedestrian movement scenarios. This stability criterion can be used to describe the robustness of the crowd structure during conflict evolution in any scenario. A higher value indicates that the system can maintain connectivity at higher thresholds, has stronger resistance to external disturbances, and exhibits higher overall stability. Conversely, a lower key seepage threshold indicates that the system is highly sensitive to conflict changes and is prone to splitting and instability at lower conflict intensities.

[0069] As can be seen above, by analyzing the dynamic evolution of connectivity state with the change of seepage threshold, and extracting the characteristics of strong and weak connected subgraphs in terms of scale and quantity changes, a quantitative mapping from local structural changes to global stability determination of conflict networks is realized. The generated stability criteria can maintain consistency and comparability in various pedestrian movement scenarios, providing a core evaluation benchmark for subsequent crowd stability identification and early warning.

[0070] Specifically, the steps for obtaining the changed states of multiple connected subgraphs in a conflicting network include: Arrange the connected subgraphs based on the number of nodes for each category; Using the seepage threshold as the variable, the trend of the number of internal nodes of the connected subgraph with the same ranking position is obtained under the category. The changing state is generated by quantitative trends.

[0071] Regarding the specific description above, at any given percolation threshold, the connected subgraphs of the conflict network are first identified. Based on the aforementioned rules for classifying categories by the connection states of directed edges between nodes within the connected subgraphs, all connected subgraphs are categorized. Subsequently, using each category as a grouping criterion, all connected subgraphs belonging to the same category are collected into the same group. Within each group, the connected subgraphs are monotonically sorted from largest to smallest according to the number of nodes they contain, forming an ordered sequence within the same category. To ensure consistency in cross-threshold comparisons, the above grouping and sorting operations are repeated at each percolation threshold. After each sorting, the sorting result is recorded as a three-element field containing the category, sequence number, and number of nodes, where the sequence number identifies the position of the connected subgraph within the current category. The function of this step is to eliminate the disordered differences in the size of connected subgraphs within the same category, allowing connected subgraphs with the same position to directly correspond one-to-one at different percolation thresholds, thus providing a strict mapping basis for subsequent comparisons of quantity trends.

[0072] After sorting each category, using the seepage threshold as the sole variable, the number of nodes within the connected subgraphs corresponding to the same arrangement position within the same category is read progressively along the increasing threshold sequence, and recorded as a trend of quantity changes with the threshold. To ensure the continuity of the trend, when a connected subgraph for a certain arrangement position of a certain category does not exist temporarily under a certain seepage threshold, a strict empty mark is used to occupy the position, without replacing it with other data; when the threshold continues to increase and a connected subgraph for that arrangement position reappears, records are added to the same trend sequence. The function of this step is to: under the premise that the category and arrangement position remain unchanged, only track the change of the number of nodes within the connected subgraph with the seepage threshold, thereby accurately reflecting the evolution characteristics of structural scale such as contraction, splitting, or disappearance during the screening process, and avoiding cross-interference between different arrangement positions.

[0073] After obtaining the quantity trends of the above-mentioned seepage threshold as the variable, within the same category and at the same rank, for each trend sequence, the overall change pattern within the preset seepage threshold range is comprehensively analyzed to generate the corresponding change state. The generation of the change state is based solely on the objective record of the quantity trend. Within the same category and at the same rank, the following verifiable quantitative characteristics are mainly used for judgment: First, whether the quantity trend shows a continuous decline during the threshold increase; second, whether the quantity trend shows a peak behavior of changing from rising to falling during the threshold increase; third, whether the quantity trend tends to zero and remains at zero after a certain threshold; fourth, whether the quantity trend shows a sudden decrease or recovery between adjacent threshold steps. Based on the above quantitative characteristics, without introducing other additional criteria, the change state corresponding to the quantity trend is fixedly recorded in the form of a three-element field of category, rank, and change state, so that the change states of the same category and the same rank can be compared horizontally and vertically within different seepage threshold value ranges. The function of this step is to transform the quantitative trend into a discrete description that can be directly used for judgment and summarization, thereby providing direct input for subsequent operations to obtain the changing states of multiple connected subgraphs in the conflict network based on the change of connectivity state with the seepage threshold, and to generate stability criteria for the conflict network in all traffic scenarios based on the changing states.

[0074] Specifically, the steps for generating a state of change through quantitative trends include: Filter by category and sort by position.

[0075] Based on the trend of the number of connected subgraphs under the filtered categories and rankings, the changing state of each flow movement scenario is generated.

[0076] Regarding the specific description above, assuming that the connected subgraphs under each category have been arranged based on the number of nodes, and the trend of the number of nodes within the connected subgraphs of the same arrangement position has been obtained using the percolation threshold as the variable, the following steps are taken: First, the category is used as the primary filter key to enter the corresponding category data domain. Within this data domain, the category is used as the secondary filter key to retain only all category entries corresponding to that category. Subsequently, under each retained category entry, the arrangement position is used as the tertiary filter key to select the corresponding entries one by one according to the predetermined arrangement position. For each successful filter result, a fixed index is established using a ternary structure of category, type, and arrangement position, and a one-to-one correspondence is established with the number trend of the connected subgraphs under this ternary structure. During implementation, to ensure the uniqueness and stability of subsequent comparisons: For all types within the same category, maintain the same range of values ​​for the order of positions; If there is no connected subgraph for the corresponding position under a certain threshold step size, the empty position is marked as a placeholder and is not replaced by other data. All filtered records are fixed with dual indexes of time order and percolation threshold order, ensuring that the mapping between category, type, and ranking position and the quantity trend of connected subgraphs can be directly traced back at any time. The function of this action is to limit subsequent processing to the truly related types and ranking positions within the target category, eliminating interference introduced by cross-category, cross-type, and cross-ranking, so that the quantity trend of each connected subgraph has a clear and unique semantic attribution.

[0077] After filtering categories and rankings according to their respective categories, the quantity trends of the connected subgraphs corresponding to the category, type, and ranking index are read one by one. The trend is then morphologically identified and summarized using the seepage threshold as the sole variable: when the trend shows a continuous decline as the threshold increases, it is recorded as a declining type; when the trend first rises and then falls with a peak, it is recorded as a peak type; when the trend approaches zero after a certain threshold and remains at zero, it is recorded as a disappearing type; when the trend shows a significant jump between adjacent steps, it is recorded as a sudden change type. The above summarization is based solely on the objective sequence of the quantity trends in the connected subgraphs, without introducing any additional indicators or substitutes. Subsequently, the identification results are solidified into a four-element field format of category, type, ranking, and change state. When there are multiple types within the same category or multiple rankings within the same type, the corresponding change states are generated and summarized one by one according to the same process, ultimately forming a set of change states for each flow movement scenario corresponding to that category. In implementation, all changing states maintain a one-to-one correspondence with the quantitative trends of connected subgraphs in their source data, and are archived using the same threshold step size index, time index, and category, type, and ranking index. This ensures that any single changing state can be traced back to its original trend sequence and selection criteria. The function of this action is to convert the quantitative trends of connected subgraphs with continuous numerical values ​​into discrete and comparable changing states. This provides a directly usable standardized output for obtaining the changing states of multiple connected subgraphs in conflict networks without altering the meaning of the feature terms, and provides structured input for subsequent generation of stability criteria and early warning strategies based on these changing states.

[0078] Specifically, the stability criterion is generated through the following steps: By using preset rules and within a preset range of changing states, a critical seepage threshold that can characterize the stability of a conflict network is obtained.

[0079] This seepage threshold is used as the key seepage threshold, and the key seepage thresholds in each flow movement scenario are used together as the stability criterion.

[0080] Regarding the specific description above, after acquiring the changing states of various connected subgraphs in the conflict network, for each flow movement scenario, a preset range is set within a unified seepage threshold value interval, and the changing states within the same scenario are judged and filtered according to preset rules. During implementation, the seepage threshold is used as the only variable, and the changing states of the scenario are read point by point in ascending order of the threshold. Judgment is performed only within the preset range: when the changing state meets the critical characteristics indicated by the preset rules, the seepage threshold corresponding to that point is extracted as a candidate value. To ensure verifiability, all candidate values ​​are recorded using a five-element field: scenario, type, ranking, seepage threshold, and changing state, and maintain a one-to-one correspondence with the quantitative trend of their source. Within the same scenario, if multiple candidate values ​​satisfy the preset rules, selection continues within the preset range according to the priority order agreed upon by the preset rules, thereby obtaining the seepage threshold that can characterize the stability criticality of the conflict network. The function of this action is to locate a single critical threshold from the summarized changing states in a unified way without changing the meaning of the preset rules, changing states and preset ranges, so that the determination of the stability critical depends only on existing records and established criteria, avoiding the introduction of any additional variables or external assumptions.

[0081] After obtaining the critical seepage threshold that characterizes the stability of the conflict network, this seepage threshold is immediately marked as a key seepage threshold in the corresponding pedestrian movement scenario, and archived as a binary field of scenario and key seepage threshold. Subsequently, the key seepage thresholds for each pedestrian movement scenario are collected to form a one-to-one correspondence set of scenarios to values ​​under the same indexing system; this set is the stability criterion for that batch of pedestrian movement scenarios. To ensure the comparability and usability of the common stability criteria, the same value scale, the same threshold step size index, and the same recording precision are used to archive the key seepage thresholds for different scenarios. When the stability criteria need to be updated in different batches or at different times, the same data structure is still used for appending records, and version differentiation and traceability are achieved through scenario identification and time indexing. The function of this action is to elevate the critical seepage threshold at the single-scene level to a unified measurement standard across scenes, enabling different pedestrian flow scenarios to be evaluated and compared under the same stability criterion. In practical use, the conflict index set of any target scenario can be directly compared with the key seepage threshold of its corresponding scenario and the thresholds of other scenarios in the set, thereby supporting subsequent stability identification and early warning triggering.

[0082] S105 determines the stability of the target pedestrian flow scenario through stability criteria and conflict index set, and selects whether to generate an alarm based on the judgment result.

[0083] Here, based on the stability criterion obtained in step S104 and the conflict index set formed in step S103, the real-time state of the target crowd movement scenario is comprehensively judged to identify whether the crowd in the current scenario is in a stable, unstable precursor, or unstable state, and accordingly, the corresponding level of warning or alarm is automatically triggered. This step realizes a closed-loop transformation from model analysis to practical application, enabling the system not only to quantitatively describe crowd stability but also to achieve risk identification and proactive intervention during on-site operation.

[0084] First, during system operation, pedestrian movement trajectory data is acquired in real time for the target pedestrian flow scenario. Following the same steps S101 to S103, the conflict network structure and conflict index set for that scenario are calculated. This conflict index set reflects the multidimensional characteristics of the crowd in terms of spatial distribution, conflict intensity, and connectivity at the current moment, including multiple parameters such as average conflict value, number of connected subgraphs, maximum connected subgraph size, and average clustering coefficient. These indicators can reveal the overall coordination and local risk status of the crowd from different perspectives.

[0085] After obtaining the conflict index set for the target scenario, the system compares and analyzes it with the established stability criteria. The stability criteria are a unified standard formed based on seepage analysis results under various typical pedestrian flow scenarios, used to determine the critical boundary of the system's transition from steady state to instability in different scenarios. Specifically, the stability criteria include key seepage thresholds and their corresponding characteristic conflict distribution patterns, where the threshold represents the maximum conflict intensity that the system can withstand while maintaining overall connectivity.

[0086] The system calculates the seepage threshold corresponding to the current conflict index set in the target scenario and compares it with the key seepage threshold in the stability criterion. When the characteristic parameters of the target scenario approach the characteristic state corresponding to the key seepage threshold, it indicates that the crowd system has entered a critical steady-state region. At this point, the size and number of local conflict clusters increase rapidly, the overall stability of the system decreases, and there is a potential risk of instability. If the conflict index further exceeds the range of the key seepage threshold, it indicates that the crowd structure has changed from a steady state to an unstable state, and the spread of conflict in local areas may trigger congestion or pushing phenomena, requiring immediate intervention.

[0087] Based on the comparison results, the system classifies the target scenario into three state levels: when the set of conflict indicators is within the stability criterion range and changes gradually, it is determined to be in a steady state; when the set of conflict indicators approaches the distribution pattern corresponding to the critical seepage threshold, it is determined to be an early sign of instability; when the conflict indicators significantly exceed the critical threshold and network connectivity shows a breaking trend, it is determined to be unstable. For different state levels, the system executes different response strategies: when in a steady state, only routine monitoring and data recording are performed; when in the early sign of instability stage, the system triggers an early warning mechanism, outputs a prompt signal, and sends a risk warning to the operation management terminal; and when entering an unstable state, the system immediately generates an alarm, alerting on-site staff through sound and light or communication commands, and can also link with station broadcasts, passenger flow guidance systems, or gate control modules to implement proactive crowd control.

[0088] As described above, through the judgment and response process, this step achieves real-time identification and dynamic early warning of crowd stability. Its core lies in using stability criteria as a unified reference standard to map conflict characteristics under different scenarios onto a unified stability evaluation scale, thereby achieving a unified assessment of various pedestrian movement states in the complex urban rail transit environment. This method can not only quickly identify potential risks during sudden changes in passenger flow, but also provide quantitative basis for optimizing passenger flow organization and making safety decisions within stations, demonstrating significant practical application value.

[0089] This invention provides a method for identifying and warning about crowd stability within urban rail transit stations. This method enables unified quantitative assessment and real-time risk warning of crowd stability in complex and variable pedestrian flow scenarios. By acquiring pedestrian movement trajectories and quantifying conflicts between neighbors, a dynamic conflict network is established with pedestrians as nodes and conflicts as directed edges, reflecting individual interaction behavior at a micro level. By setting an adjustable seepage threshold and analyzing the evolution of connectivity as the threshold changes, the method reveals the steady-state and critical instability characteristics of the crowd structure during conflict evolution. Furthermore, stability criteria are generated through the changing trends of various connected subgraphs, achieving unified stability evaluation across scenarios. Finally, by combining a real-time conflict index set, the method performs steady-state determination and outputs warnings for the target scenario, constructing a complete closed loop from data acquisition and network modeling to risk identification. Compared with existing methods, this invention overcomes the limitations of traditional analysis based on macroscopic indicators such as velocity density or pressure, and makes up for the shortcomings of focusing only on individual behavior while ignoring the overall structural evolution. It achieves a unified multi-scale representation from microscopic conflict to macroscopic steady state, and can more accurately identify the precursors of population instability and high-risk areas, providing a scientific basis and real-time decision support for passenger flow control and safety management in rail transit stations.

[0090] In any of the above embodiments, in the specific implementation process, the first step is to collect pedestrian movement trajectory data and quantify neighbor conflicts. The data used in the method includes pedestrian trajectory data in multiple typical scenarios. First, the basic motion information such as the position and speed of each pedestrian is extracted. We propose a quantitative value for neighbor conflicts to quantify neighbor conflicts. This index describes the conflict from two perspectives: space and velocity. Among them, the Volonoi diagram is used to describe spatial conflicts according to formula (1). According to formula (2), the speed obstacle characterizes the speed conflict. .

[0091] (1) (2) in, As a function, representing Point to line segment distance, pedestrian At any moment walking speed, Let represent the set of edges of the Volonoi polygon. Let the set of two sides of the velocity barrier cone be represented. This represents the Volonoi neighborhood set of pedestrians at that moment. pedestrian Compared to neighbors Speed ​​barrier cone.

[0092] Based on this, pedestrians with his neighbors exist Specific conflicts at any moment It can be calculated according to formula (3). When the pedestrian's speed is within the personal space or speed barrier cone of a neighbor, it indicates that a conflict has occurred. At this time, the quantitative value of the conflict between neighbors is negative, and its absolute value is the minimum speed adjustment required to avoid the conflict. Conversely, when the speed will not cause any conflict, the quantitative value of the conflict between neighbors is positive, indicating the maximum acceleration that the pedestrian can make without causing a conflict.

[0093] (3) In any of the above embodiments, in the specific implementation process, step two involves constructing a dynamic pedestrian conflict network. Based on the quantified neighbor conflicts from the previous step, at each time step, pedestrians are treated as nodes. If two pedestrians are in the Volonoi neighborhood (i.e., they interact geometrically), an edge is established between them, pointing from the obstructing party to the obstructed party. Subsequently, the quantified values ​​of neighbor conflicts are used to calculate the edge weights of the network; smaller weights indicate more severe conflicts between neighbors. To facilitate weighting and subsequent calculations, at each time step, the quantified values ​​of neighbor conflicts are shifted to a non-negative interval as the weights of the directed edges. , as in equation (4).

[0094] (4) In any of the above embodiments, in the specific implementation process, step three involves percolation of the network at each time step. At this time, each time step corresponds to a directed weighted graph. If an edge at a certain time step... The corresponding weight is Then we have: (5) in, A value of 0 indicates that the edge in the graph is broken, while a value of 1 indicates that the edge in the graph is connected normally. Represents a given seepage threshold, and has That is, if the edge capacity is greater than the seepage threshold, the edge is retained in the graph; otherwise, the edge is removed. It should be noted that the seepage threshold... This is a manually assigned parameter, which is increased by a step size each time. Smaller step sizes yield more accurate results. As the value increases... Edges in the network will be continuously removed until... At this point, all edges are removed, and the seepage process is complete.

[0095] In any of the above embodiments, in the specific implementation process, step four involves quantifying the steady state of the crowd. In this invention, the severity of conflict can be represented by two main perspectives: 1) conflict magnitude; 2) the diversity of conflict directions. Conflict magnitude refers to the value of any conflict itself, representing the severity of a conflict, and is directly represented by the edge weights in the network. The diversity of conflict directions refers to the direction of any conflict. If the conflict directions in the crowd are more diverse, from the perspective of network structure, the direction each person walks is more chaotic, the crowd is more disordered, and that is, the stability is worse. These two perspectives can comprehensively characterize the severity and stability of crowd conflict from the perspectives of network structure and conflict value distribution, respectively.

[0096] In order to characterize the diversity of conflict directions from the perspective of the network, the in-degree distribution of nodes is used to represent the topological structure of individual conflicts in the network at a given time, as shown in formulas (6)-(7):

[0097] (6) (7) in, yes Current Network Corresponding adjacency matrix The elements in ,if time Nodes and If connected, then ,otherwise , node At any moment The degree of ingress.

[0098] Considering the differences in conflict patterns among multiple individuals, for example, in a unidirectional flow, everyone moves in roughly the same direction, resulting in similar conflict directions and low diversity in conflict directions. However, in three-way, four-way, and even circular scenarios, each person's conflict direction differs from others, making the connection patterns between individuals more complex and resulting in higher diversity in conflict directions. Therefore, we use a weighted average clustering coefficient to characterize the conflict patterns of the entire population during movement, as shown in formulas (8)-(10): (8) (9) (10) in, yes Time Node Clustering coefficient, yes Time Node The number of directed triangles involved in the formation. time The clustering coefficient is a set of nodes. In dynamic pedestrian conflict networks, the clustering coefficient serves as an indicator of the closure of the neighborhood edges of nodes. A lower value means that the connections between neighbors are sparser and the conflict chains are more difficult to close. The clustering coefficient can represent the sparsity of local conflicts, thus more accurately characterizing the sparsity and dispersion of local interactions.

[0099] Finally, to characterize the stability of the network during the seepage process under different states and identify key thresholds, we tracked the changes in the size of the largest and second largest strong and weak connected subgraphs with the sieve threshold. We defined the thresholds corresponding to the rapid decay of the largest connected subgraph or the peak of the second largest subgraph as key seepage thresholds. The larger this threshold, the more robust the network is to disturbances and the higher its stability, indicating that it can maintain overall connectivity under more stringent sieve conditions. The largest strong connected subgraph, the second largest strong connected subgraph, the largest weak connected subgraph, and the second largest weak connected subgraph... Figure 4 The calculation of each indicator is shown in formulas (11)-(14): (11) (12) (13) (14) in, , , , Representing networks The table lists the number of nodes in the largest strongly connected subgraph, the second largest strongly connected subgraph, the largest weakly connected subgraph, and the second largest weakly connected subgraph. It's important to note that in a directed graph, a strongly connected subgraph is formed when there is a reachable path between any two nodes. A weakly connected subgraph, however, is formed by treating all directed edges in the directed graph as undirected edges before searching for a connected subgraph. Therefore, the constraints for forming a strongly connected subgraph are stronger than those for a weakly connected subgraph.

[0100] Based on the above indicators, the stability of a population can be quantified from a network perspective.

[0101] In any of the above embodiments, to verify the effectiveness of the invented crowd steady-state analysis method based on seepage theory, pedestrian trajectory experimental data from six scenarios were selected as the research object. The scenario settings are as follows: Figure 2 The diagram shows unidirectional flow (U), opposing flow (B), 120-degree tri-directional flow (T), 90-degree four-directional flow (F), room bottleneck flow (R), and circular antipodal flow (C).

[0102] First, we analyze the basic topology of the dynamic pedestrian conflict network in six scenarios. The number of nodes in scenarios U, B, T, F, and R exhibits a loading and unloading process, while the number of nodes in scenario C is constant. Because scenario T initially has low crowd density and almost no conflicts, the mean edge weights after normalization are relatively low compared to other scenarios. Within a single scenario, the in-degree and out-degree distributions of nodes are almost identical. As the system evolves over time, the state is relatively unstable at the initial and final stages, but relatively stable during the intermediate dynamic evolution. (See attached diagram) Figure 4 As shown.

[0103] Next, the directional distribution and severity of pedestrian conflicts in different scenarios were quantitatively analyzed. Conflicts in scenarios U, T, and R were concentrated in the pedestrian's target direction. Conflicts in scenario C were similar; since the pedestrian's target was evenly distributed in all directions, the conflicts were also relatively evenly distributed on the radar chart. For scenarios B and F, it was noted that the overall conflict was lowest in both scenarios, and the conflict in the pedestrian's target direction was even less severe. This may be due to the tendency for pedestrians to form queues; self-organization means that pedestrians traveling in the same direction tend to move in queues, resulting in most pedestrians traveling in the same direction, causing conflicts to occur more frequently on the sides of the pedestrian's direction of travel. The evaluation results of the above four characteristics are attached. Figure 4 As shown.

[0104] Furthermore, the percolation process of scenario U from a strongly connected perspective is demonstrated. The results show that as the network undergoes percolation transformation, it is divided into several connected subgraphs of different sizes. Regions with more severe obstruction in the network are often adjacent to the largest and second-largest connected subgraphs from a strongly connected perspective. Conversely, regions with significant obstruction tend to form more stable clusters, and the percolation process clearly illustrates the evolution of population stability. (See attached image) Figure 5 As shown.

[0105] Finally, the trends and differences in critical seepage thresholds under strongly connected and weakly connected perspectives in six scenarios were calculated (referred to as strong critical seepage thresholds and weak critical seepage thresholds). In scenarios U and R, passenger flow is relatively unidirectional, resulting in simpler conflict patterns and smaller conflicts, exhibiting significant differences in strong and weak critical seepage thresholds. In scenarios B and F, flows exist in different directions, leading to more severe conflicts between individuals. Furthermore, conflicts in four-way flows are more severe than those in two-way flows, so the difference in strong and weak critical seepage thresholds is greater in scenario B than in scenario F. Scenario T has excessive density, and scenario C has more severe passenger conflicts, thus the difference in strong and weak critical seepage thresholds is smallest in both. (See attached diagram) Figure 6 As shown.

[0106] Overall, by introducing quantitative analysis of network indicators such as clustering coefficients and seepage thresholds, this study analyzed the coupling influence mechanism of local density, conflict intensity, and individual aggressive behavior on overall homeostasis. The proposed method has advantages in identifying risk areas, quantifying conflict severity, and capturing the critical state of overall population homeostasis.

[0107] If an integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in a computer-readable medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0108] The above embodiments are only used to illustrate the technical solutions of this disclosure, and are not intended to limit it. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be included within the protection scope of this disclosure.

Claims

1. A method for identifying and warning about crowd stability within urban rail transit stations, characterized in that, Includes the following steps: Acquire action data of people in various pedestrian movement scenarios, and quantify the conflict events between each person and their neighbors and obtain quantitative features based on the pedestrian movement scenarios and the action data; Personnel are treated as nodes, and directed edges are generated between nodes based on the conflict events. The weight of each directed edge is assigned using the quantized features to construct a conflict network; An adjustable percolation threshold is set, and directed edges in the conflict network are filtered by the percolation threshold and the edge weight; During the process of increasing the seepage threshold, the connectivity status of the conflict network and the set of conflict indicators are continuously acquired; Based on the change of the connectivity state with the seepage threshold, the change states of multiple connected subgraphs in the conflict network are obtained, and the stability criteria of the conflict network in all traffic flow scenarios are generated through the change states. The stability of the target pedestrian flow scenario is determined by the stability criterion and the conflict index set, and an alarm is generated based on the result of the determination.

2. The method for identifying and warning about crowd stability in urban rail transit stations according to claim 1, characterized in that, The categories of the pedestrian flow scenarios are set by variables such as the number of pedestrian flow directions, the number of open intersections, and the flow width.

3. The method for identifying and warning about crowd stability in urban rail transit stations according to claim 2, characterized in that, The action data includes the location and speed of personnel at each moment, and the steps of quantifying conflict events between each person and their neighbor and obtaining quantified features include: Based on the location, the spatial conflict of each person in the Volonoi neighborhood is determined using a Volonoi map; Based on the stated speed, a speed obstacle is used to determine the speed conflict of each person within the speed obstacle cone; The spatial conflict and the velocity conflict are used to calculate the quantification of conflict events between each person and their neighbors in the Volonoi neighborhood.

4. The method for identifying and warning about crowd stability in urban rail transit stations according to claim 3, characterized in that, The representational meaning of the quantified features is defined by the following rules: When the quantization feature is negative, the absolute value represents the minimum speed adjustment required to avoid conflict; When the quantization feature is positive, the absolute value represents the maximum acceleration allowed without causing a conflict.

5. The method for identifying and warning about crowd stability in urban rail transit stations according to claim 2, characterized in that, For each flow movement scenario at a certain moment, the directed edges in the conflict network are filtered according to the following rules: When the weight of the directed edge is greater than the percolation threshold, the directed edge is retained in the conflict network. When the weight of the directed edge is not greater than the percolation threshold, the directed edge is deleted from the conflict network.

6. The method for identifying and warning about crowd stability in urban rail transit stations according to claim 5, characterized in that, The step of continuously acquiring the connectivity state of the conflict network and the conflict index set during the increasing seepage threshold includes: The seepage threshold is continuously increased in fixed increments within a preset range; During the enlargement process, when directed edges are deleted from the conflict network, the size of the connected subgraph is obtained, and the conflict severity of the conflict events in the conflict network is calculated. A mapping relationship between the connected subgraph and the conflict severity is generated at each step of the seepage threshold, and the conflict index set is generated based on the mapping relationship and the conflict severity.

7. The method for identifying and warning about crowd stability in urban rail transit stations according to claim 6, characterized in that, The scale includes the types of connected subgraphs and the number of nodes contained within each connected subgraph; The categories are determined by the connection states between nodes within the connected subgraph via the directed edges.

8. The method for identifying and warning about crowd stability in urban rail transit stations according to claim 7, characterized in that, The step of obtaining the changed states of multiple connected subgraphs in the conflict network includes: The connected subgraphs are arranged based on the number of nodes for each of the aforementioned categories; Using the seepage threshold as the variable, the trend of the number of internal nodes of the connected subgraph with the same rank is obtained under the category. The changing state is generated by the quantitative trend.

9. The method for identifying and warning about crowd stability in urban rail transit stations according to claim 8, characterized in that, The step of generating the change state through the quantity trend includes: Filter the categories and the ranking according to the categories; Based on the trend of the number of connected subgraphs under the filtered categories and rankings, the changing state under each of the aforementioned pedestrian movement scenarios is generated.

10. The method for identifying and warning about crowd stability in urban rail transit stations according to claim 6, characterized in that, The stability criterion is generated through the following steps: By using preset rules and the changing state within the preset range, a critical seepage threshold that can characterize the stability of the conflict network is obtained. The seepage threshold is used as the key seepage threshold, and the key seepage thresholds in each of the aforementioned pedestrian movement scenarios are used together as the stability criterion.

Images