A municipal dense flow risk control prediction system

Abstract

The application relates to the technical field of public safety of smart cities, in particular to a municipal dense people flow risk control prediction system, which comprises an information collection module, a quality checking module, a weight decision module, a situation calculation module and a risk deduction module connected in sequence. People flow data is obtained by deploying multi-source information collection equipment, the quality checking module calculates quality indexes, and the weight decision module dynamically determines a data fusion weight scheme accordingly. The situation calculation module generates a current people flow heat map and a moving trend based on the scheme, and the risk deduction module deduces the same by using a cloud group behavior simulation model, and outputs a future risk area prediction map and a potential congestion path. The system improves the accuracy of people flow monitoring and the foresight of risk early warning through dynamic data fusion and group behavior simulation.

Description

Technical Field

[0001] This invention relates to the field of smart city public safety technology, and in particular to a municipal dense crowd risk control and prediction system. Background Technology

[0002] Existing municipal dense crowd monitoring systems generally rely on single or limited data sources such as video surveillance, mobile signaling, or turnstile counting for crowd flow statistics. These systems typically use preset rules to simply overlay or complementarily fuse data from multiple sources. The weighting configuration for data fusion is static and cannot be adjusted according to real-time changes in the quality of on-site data. When a data source experiences data loss, increased noise, or systematic deviation due to equipment failure, signal interference, or scene obstruction, the static fusion scheme will treat low-quality data with high-quality data with equal weight, resulting in distorted or blind spots in the generated regional crowd flow map, affecting the accuracy of perceiving the current distribution and movement status of the crowd.

[0003] At the risk warning level, existing technologies mainly rely on setting thresholds for current crowd density to trigger alarms when these limits are exceeded, or on simple linear extrapolation of crowd movement directions. These methods can only identify areas that have reached dangerous densities or where there is obvious unidirectional aggregation; they cannot simulate the interactions between individuals within a crowd or the dynamic interaction between the group and its environment. For potential risk areas and congested paths that are rapidly evolving but have not yet formed high densities, caused by path intersections, bottleneck effects, information transmission, or abnormal events, conventional technologies lack effective prediction and detection capabilities.

[0004] A technology is needed that can adaptively assess the quality of multi-source data and dynamically optimize fusion strategies to improve the reliability of basic situational awareness. Simultaneously, a technology is needed that can go beyond simple threshold judgments and simulate complex dynamic behaviors of crowds to achieve early and accurate prediction of potential risks. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a risk control and prediction system for dense crowds in municipal areas.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a municipal dense crowd flow risk control and prediction system, comprising: The information collection module deploys a group of multi-source information collection devices at key nodes within the municipal area to continuously collect information and obtain a multi-dimensional spatiotemporal pedestrian flow data set for the target municipal area within a preset collection period. The quality verification module performs data integrity and consistency verification on the multi-dimensional spatiotemporal pedestrian flow data set, and generates trajectory continuity index, density distribution rationality index, and passenger flow matching index. The weight decision module, based on the trajectory continuity index, the density distribution rationality index, and the passenger flow matching index, performs pattern matching queries in a pre-built data fusion strategy library to determine a multi-source data fusion weight scheme suitable for the target municipal area. The situation calculation module initiates a regional pedestrian flow situation calculation task based on the multi-source data fusion weighting scheme, and fuses and calculates the multi-dimensional spatiotemporal pedestrian flow data set to generate a regional pedestrian flow heat map and a group movement trend vector at the current moment. The risk extrapolation module inputs the regional pedestrian heat map and the group movement trend vector into the cloud-based risk prediction engine. It then uses the group behavior simulation model built into the cloud-based risk prediction engine to extrapolate and analyze the regional pedestrian heat map and the group movement trend vector, generating a risk area prediction map and a set of potential congestion path predictions for future periods.

[0007] As a further aspect of the present invention, the multidimensional spatiotemporal pedestrian flow data set is subjected to data integrity and consistency verification to generate trajectory continuity indicators, density distribution rationality indicators, and passenger flow matching indicators, including: The multi-source information acquisition equipment group includes video surveillance equipment, mobile terminal signaling base stations, and vehicle card-swiping terminals. The multidimensional spatiotemporal human flow data set includes an individual movement trajectory data set, a regional population density data set, and a transportation passenger flow data set. The individual movement trajectory data set is fragmented using the spatiotemporal pane segmentation method to obtain a set of several continuous trajectory segments. Calculate the number of trajectory interruptions and the number of average velocity mutation points for each set of continuous trajectory segments, and calculate the mean interruption rate and standard deviation of mutation for all sets of continuous trajectory segments. Combine the mean interruption rate and the standard deviation of mutation to obtain the trajectory continuity index. Spatial gridding analysis is performed on the population density data set of the region to calculate the population density value of each grid and analyze the density gradient distribution between adjacent grids. The rationality index of the density distribution is generated by combining the spatial autocorrelation coefficient of the density value and the uniformity of the gradient distribution. Extract the passenger flow sequences of each route from the passenger flow data set of the transportation vehicles, and compare them with the number of trajectories originating from or going to transportation stations in the individual movement trajectory data set in the corresponding spatiotemporal pane. Calculate the passenger flow matching degree of each route, and derive the passenger flow matching index by combining the mean and variance of the passenger flow matching degree of all routes.

[0008] As a further aspect of the present invention, based on the trajectory continuity index, the density distribution rationality index, and the passenger flow matching index, a pattern matching query is performed in a pre-built data fusion strategy library to determine a multi-source data fusion weighting scheme suitable for the target municipal area, including: The trajectory continuity index, the density distribution rationality index, and the passenger flow matching index are input as three-dimensional quality vectors into the data fusion strategy library; Retrieve several historical fusion strategy records with the highest cosine similarity to the three-dimensional quality vector from the data fusion strategy library, and extract the data source weight allocation scheme corresponding to each of the several historical fusion strategy records. The data source weight allocation schemes of the aforementioned historical fusion strategy records are weighted and fused based on confidence level, and the fused weight allocation scheme is used as the multi-source data fusion weight scheme.

[0009] As a further aspect of the present invention, the construction process of the data fusion strategy library includes: We collected samples of trajectory continuity indicators, density distribution rationality indicators, and passenger flow matching indicators from multiple historical municipal area monitoring cases, along with their respective validated fusion weighting schemes. A three-dimensional strategy matching space is constructed, with trajectory continuity index as the first dimension, density distribution rationality index as the second dimension, and passenger flow matching index as the third dimension. The quality index sample of each historical case is mapped to a strategy anchor point in the three-dimensional strategy matching space, and the corresponding strategy anchor point is marked with the effective fusion weight scheme sample of the historical case to form the data fusion strategy library.

[0010] As a further aspect of the present invention, when performing pattern matching queries in a pre-built data fusion strategy library, a strategy anchor point optimization step is also included: In the three-dimensional strategy matching space, the initial query radius is defined with the query point corresponding to the current three-dimensional quality vector as the center of the sphere; Calculate the density of all strategy anchor points within the initial query radius; If the density is lower than the preset anchor density threshold, the query radius is adaptively expanded until the density reaches the threshold, and then cosine similarity-based strategy anchor retrieval is performed within the radius.

[0011] As a further aspect of the present invention, a regional pedestrian flow situation calculation task is initiated based on the multi-source data fusion weighting scheme. This involves fusing and calculating the multi-dimensional spatiotemporal pedestrian flow data set to generate a regional pedestrian flow heat map and a group movement trend vector for the current moment, including: According to the multi-source data fusion weighting scheme, calculation weights are assigned to video surveillance data, mobile signaling data, and card swipe data respectively. Based on the calculated weights, the population counts within the same geographic grid from different data sources are weighted and summed to obtain the merged population count for each grid. Based on the population data after merging all grids, a continuous distribution heat map of population flow in the region is generated using a spatial interpolation method. The overall movement direction and speed of the group in the individual movement trajectory data set are analyzed, and the movement trend vector of the group, which represents the current main direction of the flow of people, is calculated by the vector averaging method.

[0012] As a further aspect of the present invention, the group behavior simulation model built into the cloud-based risk prediction engine is used to extrapolate and analyze the regional population heat map and the group movement trend vector to generate a risk area prediction map and a set of potential congestion path predictions for future time periods, including: The heat map of pedestrian flow in the region is discretized into the initial distribution of surrogate individuals on a spatial grid; The group movement trend vector is input into the group behavior simulation model as an environmental guiding force; In the group behavior simulation model, the movement decision of each agent individual is simulated based on the principle of social force model, under the combined effects of environmental guidance, destination attraction, interpersonal rejection and obstacle avoidance. Run the group behavior simulation model to a specified future time, count the density of agent individuals in each spatial grid, and generate a predicted density distribution map; By comparing the predicted density distribution map with the preset safety capacity thresholds for each region, the areas exceeding the limit are marked, thus forming the risk area prediction map; Paths exhibiting consistently high traffic volume and speeds below a threshold during the simulation process are identified and included in the potential congestion path prediction set.

[0013] As a further aspect of the present invention, the construction and calibration process of the group behavior simulation model includes: Acquire multi-source monitoring data from historical high-density crowd scenarios to reconstruct the crowd distribution and movement vectors at historical moments; Set the initial parameters of the group behavior simulation model, including individual expected speed, interpersonal force magnitude, and field of vision; The simulation model is run using historical data as initial conditions, and the population distribution and movement status output by the simulation are compared with the historical data that actually occurred later. By optimizing the algorithm and adjusting the model parameters, the difference between the simulation results and the actual data is minimized until the model output error converges to an acceptable range.

[0014] As a further aspect of the present invention, the system further includes: The instruction execution module triggers the early warning decision core to generate a set of risk control instructions when the predicted risk level of an area in the risk area prediction map exceeds a preset risk threshold, and then sends the set of risk control instructions to the municipal execution terminal network. The risk control instruction set is parsed through the municipal execution terminal network, driving the information release device, traffic signal control system and on-site traffic control personnel to perform coordinated traffic control operations. The triggering early warning decision core generates a set of risk control instructions, including: Based on the risk area prediction map and the potential congestion path prediction set, query the municipal contingency plan database and match the applicable basic traffic management plan template. By combining real-time traffic operation status information and the availability of municipal facilities, the abstract measures in the basic traffic diversion plan template are specified and parameters are assigned to generate a draft instruction containing specific measures, execution locations and activation timing; After performing logical consistency checks and resource conflict detection on the draft instructions to ensure that the instructions are executable and free of contradictions, the final set of risk control instructions is formed.

[0015] As a further aspect of the present invention, the risk control instruction set is parsed through the municipal execution terminal network to drive the information dissemination device, traffic signal control system, and on-site traffic management personnel to perform coordinated traffic management operations, including: The risk control instruction set is analyzed and classified into information guidance instructions, signal control instructions, and personnel dispatch instructions. Based on the content and target area of ​​the information guidance instructions, control the corresponding electronic displays, broadcasts, and mobile application push terminals to release real-time pedestrian flow guidance information; Based on the parameters of the signal control instructions, adjust the traffic light timing schemes at relevant intersections to provide priority passage for evacuation flows; The personnel dispatch instructions are sent to the on-site management personnel terminal to guide the personnel to the designated location to carry out crowd diversion and order maintenance tasks, and the execution status is fed back to the early warning decision core.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: By quantitatively verifying the trajectory continuity, density distribution rationality, and passenger flow matching of multi-dimensional spatiotemporal pedestrian flow data, objective indicators reflecting the reliability of each data source are generated. Based on these real-time quality indicators, the optimal multi-source data fusion weight scheme is dynamically matched and determined from a pre-built strategy library. This process enables the system to automatically reduce the contribution of low-quality or abnormal data in the fusion calculation, while enhancing the influence of high-quality data. The resulting regional pedestrian flow heat map and group movement trend vector have a more solid and reliable data foundation, reducing misjudgments of the overall situation due to the failure or performance fluctuation of a single data source, and improving the system's robustness and perception accuracy in complex real-world environments.

[0017] By utilizing the crowd behavior simulation model built into the cloud-based risk prediction engine, the thermal distribution and movement trend vector of the crowd at the current moment are extrapolated and analyzed. This model can simulate the micro-behavioral rules of individuals during movement, such as following, avoiding, and conforming, as well as macro-effects such as panic propagation and information diffusion. By calculating the evolution of the crowd state over multiple future time steps, it can identify congestion that is about to form due to path conflicts and traffic convergence at physical bottlenecks, or predict secondary risk areas newly generated due to the diffusion of pressure from local high-density areas. This extrapolation method based on complex system simulation can reveal the hidden, nonlinear risk evolution patterns in crowd dynamics, outputting a set of potential congestion paths and a predicted map of future risk areas, realizing a shift from passively responding to high-density alarms that have already occurred to proactively warning of potential risks that have not yet formed. Attached Figure Description

[0018] Figure 1 This is a time sequence diagram of the municipal dense crowd flow risk control and prediction system described in this invention; Figure 2 A flowchart for determining the weighting scheme for multi-source data fusion; Figure 3 A flowchart for generating a heat map of regional pedestrian flow and a vector of population movement trends. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 At key nodes within the target municipal area, such as transportation hubs, commercial centers, and scenic area entrances, a multi-source information collection device cluster consisting of video surveillance equipment, mobile terminal signaling base stations, and transportation card-swiping terminals is deployed. This cluster continuously collects information, obtaining a multi-dimensional spatiotemporal pedestrian flow data set for the target municipal area within a preset collection cycle. Subsequently, the quality verification module performs data integrity and consistency verification on the multi-dimensional spatiotemporal pedestrian flow data set, generating trajectory continuity indicators, density distribution rationality indicators, and passenger flow matching indicators. Based on the above three quality indicators, the weight decision module performs pattern matching queries in a pre-built data fusion strategy library to determine a multi-source data fusion weight scheme suitable for the current target municipal area. The situation calculation module initiates a calculation task based on this weight scheme, fusing and calculating the multi-dimensional spatiotemporal pedestrian flow data set, and outputting a regional pedestrian flow heat map and a group movement trend vector at the current moment. The risk simulation module inputs heat map and trend vector into the cloud-based risk prediction engine. This engine uses a built-in group behavior simulation model to perform simulation analysis and finally generates a risk area prediction map and a set of potential congestion path predictions for future periods, providing direct basis for municipal management departments to make forward-looking decisions and proactively manage traffic.

[0022] In one embodiment of the present invention, the verification process of the quality verification module for the multi-dimensional spatiotemporal pedestrian flow data set is illustrated using a large-scale event scenario in a municipal park as an example. The information collection module deploys a multi-source information collection device group at multiple entrances, main passages, open squares, and adjacent subway and bus hubs within the municipal park. This device group includes video surveillance equipment for capturing video sequences and generating structured trajectories, mobile terminal signaling base stations for anonymously collecting mobile terminal connection events to estimate the area's population, and transportation card-swiping terminals for recording bus and subway card transactions. During a preset collection period during the event, for example, from 2:00 PM to 6:00 PM, the multi-source information collection device group continuously operates, obtaining a multi-dimensional spatiotemporal pedestrian flow data set for the target municipal park area. This multi-dimensional spatiotemporal pedestrian flow data set specifically includes an individual movement trajectory data set generated by the video surveillance equipment using target detection and tracking algorithms, an area population density data set estimated by the mobile terminal signaling base station based on signaling connection density, and a transportation passenger flow data set provided by the bus company and subway system.

[0023] In its implementation, the quality verification module employs a spatiotemporal pane segmentation method to fragment the individual movement trajectory data set. This method uses a fixed 5-minute time interval and a 10-meter by 10-meter geographic grid as the spatial unit. It segments and categorizes continuous trajectory point sequences with timestamps and geographic coordinates, resulting in several sets of continuous trajectory segments. Each set contains trajectory points of the same target within the same spatiotemporal pane. The module calculates the number of trajectory interruptions and the number of average velocity abrupt changes for each continuous trajectory segment set. The number of trajectory interruptions refers to the number of missing points between continuous trajectory points within a segment due to occlusion or recognition failure. The number of average velocity abrupt changes refers to the number of times the velocity change between adjacent trajectory point pairs within a segment exceeds a set threshold. The module also calculates the mean interruption rate and standard deviation of abrupt changes for all continuous trajectory segment sets. The mean interruption rate is the average ratio of the number of interruptions to the total duration of all segments, and the standard deviation of abrupt changes is the standard deviation of the average number of velocity abrupt changes for all segments. The mean interruption rate and standard deviation of abrupt changes are then fused to obtain the trajectory continuity index. The fusion method is a weighted summation, expressed by the formula:

[0024] in: Represents the continuity index of the trajectory. Represents the average interruption rate. Represents the standard deviation of mutations. and These are preset weighting coefficients used to balance the contributions of the two factors in the indicator.

[0025] In some embodiments, when performing quality verification on a regional population density dataset, the module performs spatial gridding analysis. The module divides the target municipal area into a 10m x 10m spatial grid, identical to the one used in trajectory processing, and calculates the population density value for each grid in each time slice within the acquisition period. The population density value is derived from the de-identified user counts of the mobile terminal signaling base station. The module analyzes the density gradient distribution between adjacent grids, calculating the mean absolute value of the difference between the population density values ​​of each grid and its eight neighboring grids, as the local density gradient for that grid. The module combines the spatial autocorrelation coefficient of the density values ​​with the gradient distribution uniformity to generate a density distribution rationality index. The spatial autocorrelation coefficient, for example, using the global Moran's index, is used to quantify the spatial clustering or dispersion pattern of population density. The gradient distribution uniformity is assessed by calculating the information entropy or coefficient of variation of the local density gradients of all grids, measuring the smoothness or drasticness of density changes. The density distribution rationality index is a weighted combination of the spatial autocorrelation coefficient and the gradient distribution uniformity index.

[0026] In some embodiments, for a transportation passenger flow dataset, the module performs passenger flow matching verification. The module extracts the passenger flow sequences for each bus or subway line within the data collection period, at each station. The temporal resolution of these sequences is consistent with the time interval of the spatiotemporal pane segmentation method. Within the corresponding spatiotemporal pane, the module counts the number of trajectories whose origins or destinations are located within a 50-meter buffer zone around the transportation station. These trajectories are considered as potential passenger flow originating from or heading to that station, as observed through video data. The module compares the absolute value of the passenger flow for each line within each spatiotemporal pane with the number of trajectories originating from or heading to that line's station within the corresponding spatiotemporal pane to calculate the passenger flow matching degree for each line. The passenger flow matching degree can be calculated as 1 minus the absolute value of the standardized difference between the two. The module synthesizes the passenger flow matching degree values ​​calculated for all lines across all spatiotemporal panes and calculates the mean and variance of these values. Ultimately, the module derives a passenger flow matching index by fusing the mean and variance of passenger flow matching. The fusing method can be a certain proportion of the mean minus the variance, so that the index can reflect the average matching level and also penalize situations where the matching degree fluctuates too much.

[0027] In practical implementation, the indicators generated by the above verification process have specific numerical meanings. For example, during the closing hours of large-scale events in municipal parks, video surveillance equipment may experience significant obstruction due to dense crowds, leading to an increase in the number of trajectory interruptions in the individual movement trajectory data set. This results in a corresponding decrease in the calculated trajectory continuity index, indicating potentially insufficient data continuity. The population density values ​​collected by mobile terminal signaling base stations at the park exit are extremely high, while the street grid density drops sharply within 50 meters of the exit. This leads to a large density gradient between adjacent grids, worsening the gradient distribution uniformity index and consequently resulting in a low density distribution rationality index value. This suggests that abrupt changes in density data in space require careful consideration. Transportation card-swiping terminal records show a surge in passenger flow entering the park's east gate subway station during closing hours, while the individual movement trajectory data set shows a simultaneous significant increase in the number of trajectories heading towards the east gate subway station. Comparing the two, the calculated passenger flow matching degree is high, and the matching degree fluctuates little across different time periods, leading to a high passenger flow matching index value. This indicates good consistency between card-swiping data and video trajectory data in this scenario. The three specific values—trajectory continuity index, density distribution rationality index, and passenger flow matching index—together constitute a three-dimensional vector describing the current multi-source data quality status, which is then input into the subsequent weight decision module.

[0028] See Figure 2 In one embodiment of the present invention, the weight decision module determines a multi-source data fusion weight scheme suitable for the current target municipal area based on the trajectory continuity index, density distribution rationality index, and passenger flow matching index output by the quality verification module. Taking the monitoring scenario of a core urban commercial street during weekend evening rush hour as an example, the trajectory continuity index value output by the quality verification module is 0.85, the density distribution rationality index value is 0.72, and the passenger flow matching index value is 0.90. The weight decision module combines these three index values ​​into a three-dimensional quality vector. The weight decision module inputs the three-dimensional quality vector into a pre-built data fusion strategy library for pattern matching query. The data fusion strategy library stores a large number of effective data fusion strategy records under different scenarios in history. Each record contains a three-dimensional vector composed of historical quality indices and its corresponding fusion weight scheme that has been verified as effective in practice. The weight decision module calculates the three-dimensional quality vector. The three-dimensional quality vector corresponding to each historical policy record in the data fusion policy library The cosine similarity between them is calculated using the following formula:

[0029] in: This represents the cosine similarity between the i-th historical strategy record and the current vector. Represents the current three-dimensional mass vector. This represents the 3D quality vector corresponding to the i-th historical strategy record in the data fusion strategy library. The module retrieves the cosine similarity. The module selects the three historical fusion strategy records with the highest similarity values. For example, these three historical records have similarity values ​​of 0.98, 0.96, and 0.94, respectively. The module extracts the data source weight allocation schemes corresponding to each of these three historical fusion strategy records, assuming they are: Record A's scheme is (video weight: 0.5, signaling weight: 0.3, card swipe weight: 0.2); Record B's scheme is (video weight: 0.6, signaling weight: 0.25, card swipe weight: 0.15); Record C's scheme is (video weight: 0.55, signaling weight: 0.35, card swipe weight: 0.10). For the data source weight allocation schemes of these three historical fusion strategy records, the weight decision module performs a confidence-based weighted fusion. The confidence of each historical record is determined by its similarity. The weight is determined by both the value and the historical application success rate. In this example, assume the calculated fusion weights for the three records are 0.5, 0.3, and 0.2, respectively. The weight decision module calculates the fusion weight for the video data as: 0.5 + 0.5 + 0.3 + 0.6 + 0.2 + 0.55 = 0.53. Similarly, the weight for the signaling data is calculated as: 0.5 + 0.3 + 0.3 + 0.25 + 0.2 + 0.35 = 0.295, and the weight for the card swipe data is calculated as: 0.5 + 0.2 + 0.3 + 0.15 + 0.2. 0.10 = 0.165. After normalization, the final multi-source data fusion weighting scheme is as follows: video data weight 0.53, mobile signaling data weight 0.30, and transportation card swipe data weight 0.17.

[0030] In some embodiments, the construction process of the pre-built data fusion strategy library involves the systematic collection and organization of historical data. The construction process includes collecting multiple historical municipal area monitoring cases, such as complete data records from scenarios like the end of sporting events, holiday visits to scenic spots, and large-scale commercial promotional activities. For each historical case, samples of trajectory continuity indicators, density distribution rationality indicators, and passenger flow matching indicators, calculated by the quality verification module, are extracted. Samples of fusion weight schemes that have been verified to produce accurate pedestrian flow trends in that case are recorded; these schemes are marked as "verified valid." A three-dimensional Cartesian coordinate system is constructed using the trajectory continuity indicator as the first dimension, the density distribution rationality indicator as the second dimension, and the passenger flow matching indicator as the third dimension. This space is the three-dimensional strategy matching space. The three quality indicator sample values ​​of each historical case are used as coordinate points and mapped to the three-dimensional strategy matching space; these coordinate points are called strategy anchor points. Each strategy anchor point is marked using its corresponding verified and valid historical fusion weight scheme sample. By aggregating hundreds or thousands of such labeled policy anchors, a data fusion strategy library covering different combinations of quality data is formed. When a new scenario generates a set of quality indicators, its corresponding three-dimensional vector forms a query point in the three-dimensional policy matching space. The weight decision module obtains the basis for fusion decisions by finding the weight schemes marked by the policy anchors adjacent to this query point.

[0031] It can be understood that the data fusion strategy library is a dynamically updated knowledge base. As the system is deployed and applied in more municipal areas and more types of scenarios, each new application cycle, along with its generated data quality indicators and the ultimately validated fusion weight scheme, can be collected as a new historical case. The quality indicators and effective weight scheme of this new case will constitute a new strategy anchor point and be added to the three-dimensional strategy matching space, thereby continuously enriching the coverage of the data fusion strategy library and the diversity of decision-making basis. This construction method enables the system to make decisions based on historical experience, rather than relying on fixed, preset weight rules.

[0032] See Figure 3 In one embodiment of the present invention, the strategy anchor point selection step in the pattern matching query process is illustrated using the decision-making process of a railway station hub area during the peak return travel period of holidays as an example. The current three-dimensional quality vector received by the weight decision module is... This vector represents the evaluation values ​​of the trajectory continuity index, density distribution rationality index, and passenger flow matching index under the given spatiotemporal conditions. The weighted decision module uses a three-dimensional quality vector within a pre-constructed three-dimensional strategy matching space. corresponding coordinates Define an initial query radius centered on the sphere. Initial query radius The value is 0.1. The weight decision module calculates the initial query radius. Spatial distribution density of all strategy anchor points within the spherical space formed The strategy anchor point is the mapping point of historical cases in the three-dimensional strategy matching space. The calculated initial spatial distribution density... The preset anchor density threshold is 2. The value is 3. This is due to the calculated initial spatial distribution density. The value is 2, which is lower than the preset anchor density threshold. The value is 3, therefore the weight decision module initiates the process of adaptively expanding the query radius. The weight decision module will adjust the query radius... With step size Gradually increase the radius, for example, by 0.05 each time, and recalculate the new radius after each increase. Spatial distribution density of strategy anchor points within When the query radius When the value is increased to 0.2, the number of strategy anchors within the radius increases, and the calculated spatial distribution density... Reaching 3.5, at this point The anchor density is higher than the preset threshold. The weight decision module stops expanding the query radius and, within the finally determined spherical space with a radius of 0.2, performs a strategy anchor point retrieval based on cosine similarity to find the current 3D quality vector. The most similar historical strategy anchors and their corresponding fusion weight schemes. This strategy anchor selection process ensures that even when querying certain sparse regions of the three-dimensional strategy matching space, the system can adaptively expand the search scope to find a sufficient number of valuable historical strategy anchors, thus supporting robust subsequent weight decisions and avoiding the problem of not being able to obtain effective references due to the scarcity of nearby historical cases.

[0033] In practical implementation, the situational awareness calculation module initiates the regional pedestrian flow situational awareness calculation task based on a determined multi-source data fusion weighting scheme. In the example of a railway station hub area, assuming the multi-source data fusion weighting scheme obtained from the weight decision module is: weighting based on video surveillance data... Mobile signaling data weight calculation Weighting of transportation card swipe data The situational awareness calculation module performs a weighted summation of population counts within the same geographic grid from different data sources. The geographic grid division is consistent with the quality verification phase, for example, 20 meters by 20 meters. For a specific grid... The population count from video surveillance data is The population estimate from mobile signaling data is The associated passenger flow from transportation card swipe data (converted to the number of people entering or exiting the station) is The module calculates the population of the merged grid according to the computational weights. The calculation is performed using the following formula:

[0034] in: Representing the grid The population size after integration , , These represent the calculation weights for video surveillance data, mobile signaling data, and transportation card swipe data, respectively. , , These represent data from the corresponding data source, specifically for the grid. The population count or estimate is obtained. The situation calculation module traverses all geographic grids within the target area, performs weighted calculations, and obtains a fused population set of all grids. Based on this set, the module uses spatial interpolation methods, such as inverse distance weighted interpolation, to interpolate the population values ​​of discrete grids, generating a continuous and smooth regional population heat map. This map uses color depth to intuitively represent the population density at different spatial locations.

[0035] In one embodiment of the present invention, the risk simulation module takes the regional pedestrian flow heat map and the group movement trend vector output by the situation calculation module as input, and sends them to the group behavior simulation model built into the cloud-based risk prediction engine for simulation analysis. Taking the New Year's Eve countdown event scenario in a city center square as an example, the regional pedestrian flow heat map is presented as a discretized grid density value, with a grid size of 5 meters by 5 meters. The group movement trend vector is obtained by the situation calculation module, with an east-west component of 1.2 meters per second and a south-north component of 0.5 meters per second, representing that the crowd is moving as a whole in the southeast direction. The risk simulation module discretizes the regional pedestrian flow heat map into the initial distribution of surrogate individuals on the spatial grid. Specifically, it converts the population in each grid into a fixed number of surrogate individuals proportionally, for example, 1 surrogate individual for every 10 people, so that the spatial pattern and heat distribution of the initial distribution are consistent with the heat distribution. Figure 1 The module inputs the group movement trend vector as an environmental guiding force into the group behavior simulation model. In the model, the environmental guiding force is manifested as a continuous force applied to the agent individual in the southeast direction, guiding its macroscopic movement direction and speed towards the direction and magnitude indicated by the trend vector.

[0036] In practical implementation, the group behavior simulation model, based on the principles of the social force model, simulates the movement decisions of each agent individual under the combined influence of environmental guidance, destination attraction, interpersonal repulsion, and obstacle avoidance. The model defines state variables such as position, velocity, and desired velocity for each agent individual, and calculates the resultant force of the four forces within each simulation time step to update the individual's acceleration and position at the next moment. The environmental guidance force is determined by the input group movement trend vector; the strength of the destination attraction force is inversely proportional to the individual's distance from the preset destination; the interpersonal repulsion force increases sharply as the distance between individuals decreases; and the obstacle avoidance force acts on individuals close to building walls, fences, or temporary stages. Agent individuals adjust their movement paths according to the direction of the resultant forces, thus forming a dynamic crowd flow pattern in the simulation space. Running the group behavior simulation model to a specified future time, such as 30 minutes after the event ends, the model outputs the number of agent individuals contained in each spatial grid at that moment, and statistically obtains a predicted density distribution map. By comparing the predicted density distribution map with preset safety capacity thresholds for each region (the safety capacity thresholds are pre-set based on grid area and per capita area), all regions where the predicted density exceeds the safety threshold are marked. These marked regions together constitute the risk area prediction map. During the simulation, the model simultaneously monitors the flow rate and velocity on each path, identifies paths with consistently high flow rates and velocities below the threshold, and includes these paths in the potential congestion path prediction set.

[0037] In some embodiments, the construction and calibration of the crowd behavior simulation model relies on multi-source monitoring data from historical high-density crowd scenarios. Multi-source monitoring data from historical high-density crowd scenarios is acquired, such as video and signaling data from the end of a stadium event, to reconstruct the crowd distribution and movement vectors at historical moments. Initial parameters of the crowd behavior simulation model are set, including individual expected speed, interpersonal force magnitude, and field of vision. Individual expected speed can be set based on the mainstream walking speed of crowds in historical data; interpersonal force magnitude is calibrated based on the perceived pressure under crowded conditions; and the field of vision limits the range of surrounding space that an individual can perceive during decision-making. The simulation model is run using historical data as initial conditions, and the simulated crowd distribution and movement status are compared with subsequent actual historical data. The model parameters are adjusted through optimization algorithms to minimize the difference between the simulation results and the actual data. The optimization objective can be set as minimizing the mean absolute error between the simulated distribution and the actual distribution, until the model output error converges to an acceptable range.

[0038] In some embodiments, model calibration can employ a combination of parameter scanning and iterative fitting. Reasonable ranges are set for the expected individual speed, interpersonal force magnitude, and field of vision. Within these ranges, multiple parameter combinations are selected and simulations are run. The simulation output for each parameter combination is compared with historical real data, and error indices are recorded. The parameter combination with the smallest error is selected as the new initial value, the range is narrowed, and the scanning is repeated iteratively until the optimal error change between two adjacent iterations is less than the set tolerance. Refer to Table 1, which lists the three sets of parameters and their corresponding error values ​​for a particular iteration during the calibration process.

[0039] Table 1: Example Table of Parameter Calibration for Group Behavior Simulation Model

[0040] Optionally, random perturbation terms can be introduced during model operation to allow individual behaviors to exhibit a certain degree of diversity while adhering to the laws of the main forces, avoiding overly smooth simulation results that deviate from the heterogeneous performance of real crowds. Optionally, the calculation of obstacle avoidance forces can be combined with precise boundary data provided by high-precision maps, enabling proxy individuals to make more realistic avoidance actions when approaching steps, slopes, or temporary structures. It can be understood that by transforming the regional pedestrian flow heat map into the initial distribution of proxy individuals and guiding the simulation with the group movement trend vector, the possible distribution and flow of crowds in future periods can be reproduced in virtual space. The resulting risk area prediction map and potential congestion path prediction set can reflect high-risk states that have not yet occurred but are foreseeable. The construction and calibration process of the group behavior simulation model enables the model to possess dynamic response capabilities that match the characteristics of real crowd movement, thereby providing more realistic prediction results in risk extrapolation.

[0041] In one embodiment of the present invention, the instruction execution module is triggered when the predicted risk level of an area in the risk area prediction map exceeds a preset risk threshold. Taking the end scene of a light show on a riverside promenade as an example, the risk area prediction map marks three grid clusters whose predicted density exceeds the safe capacity threshold. The corresponding predicted risk level is calculated to be high risk, exceeding the upper limit of the medium risk set by the preset risk threshold. The system then triggers the early warning decision core. The early warning decision core queries the municipal contingency plan database based on the risk area prediction map and the set of potential congestion path predictions, and matches an applicable basic diversion plan template. The basic diversion plan template includes general diversion principles and measure categories for different sections of the riverside promenade where people are stranded. The early warning decision-making core combines real-time traffic operation information and the availability of municipal facilities to concretize and assign parameters to the abstract measures in the basic traffic management plan template. Real-time traffic operation information includes traffic saturation at nearby intersections and bus departure frequency. The availability of municipal facilities includes variable message signs, loudspeakers, and the number of on-duty traffic management personnel. Specific measures include adding temporary directional signs to the main road in the middle section of the riverside area, adjusting bus connection frequencies at the east and west intersections, and assigning a fixed number of traffic management personnel to the vicinity of high-risk areas. A draft instruction is generated, containing specific measures, execution locations, and activation timing. For example, the draft instruction specifies that information board guidance should be activated 120 meters from the east entrance on the main road in the middle section of the riverside area, 8 minutes after the event ends, and that bus departures at the east and west bus stops should be increased 10 minutes after the event ends. The early warning decision-making core performs logical consistency verification and resource conflict detection on the draft instructions. Logical consistency verification checks whether there are temporal contradictions or conflicting objectives between the measures. Resource conflict detection verifies whether the required facilities and personnel have been occupied by other tasks at the specified time. After confirming that the instructions are executable and without contradictions, the final set of risk control instructions is formed.

[0042] In practice, the instruction execution module distributes a set of risk control instructions to the municipal execution terminal network. The municipal execution terminal network parses these instructions, classifying them into information guidance instructions, signal control instructions, and personnel dispatch instructions. Information guidance instructions specify the locations and target areas for disseminating pedestrian guidance information, while signal control instructions define the green light extension time and phase sequence for traffic lights at relevant intersections. Personnel dispatch instructions specify the target locations and arrival times for personnel. Based on the content and target areas of the information guidance instructions, the network controls corresponding electronic displays, broadcasts, and mobile applications to publish real-time pedestrian guidance information. Electronic displays are placed in prominent locations along the riverside promenade, broadcasts cover the promenade and adjacent plazas, and mobile applications send targeted notifications to users already within the geofence. Based on the parameters of the signal control instructions, the network adjusts the traffic light timing schemes at relevant intersections to provide priority passage for evacuation flows. For example, at the intersection connecting the riverside promenade to the main road at the eastern end, the east-west green light duration is extended, and the north-south left-turn phase is shortened to improve the efficiency of dispersing crowds crossing the main road. Personnel dispatch instructions are sent to the on-site management personnel's terminals, guiding them to designated locations to perform crowd diversion and order maintenance tasks. Upon receiving the instructions, the on-site management personnel's terminals generate navigation routes and display a task list. Following the instructions, the personnel arrive at the vicinity of high-risk grids at the designated time to guide the crowd away along low-density paths. The execution status is fed back to the early warning decision-making core, including whether the information guidance equipment is displaying correctly, whether the traffic light timing has been successfully switched, whether the personnel have arrived on time, and the on-site crowd response. Based on this feedback, the early warning decision-making core updates the instruction execution monitoring view and can trigger supplementary dispatching.

[0043] In some embodiments, the process of triggering the early warning decision core to generate a set of risk control instructions can introduce multi-dimensional risk level determination. Each area in the risk area prediction map is assigned a composite risk score calculated based on the proportion of predicted density exceeding the safe capacity threshold and the number of associated paths included in the potential congestion path prediction set. The composite risk score is compared with a preset risk threshold to determine whether to trigger the system. The preset risk threshold can be set in segments according to the activity type and site carrying capacity characteristics. When combining real-time traffic operation status information and the availability of municipal facilities for concretization and parameter assignment, the implementation intensity of measures can be graded. For example, the signal priority amplitude can be increased at intersections with high traffic saturation, and backup information dissemination channels can be invoked when facilities are in low availability.

[0044] In some embodiments, the parsing and driving process of the municipal execution terminal network can establish an instruction priority queue. Information guidance instructions, signal control instructions, and personnel dispatch instructions are sorted according to the risk level in the risk area prediction map and the impact range of the potential congestion path prediction set before entering the execution stage. High-priority instructions receive priority execution guarantees when resources are limited. Optionally, during the execution of information guidance instructions, real-time broadcast logs from electronic displays and broadcasts can be collected simultaneously to verify whether the guidance information covers the target area as planned. Optionally, during the execution of personnel dispatch instructions, the on-site management personnel terminal can integrate simple positioning and status reporting functions, enabling the early warning decision-making core to monitor the movement trajectory and on-duty status of personnel in real time.

[0045] It is understandable that by linking the risk area prediction map with the potential congestion path prediction set to query the municipal contingency plan database and generate an executable set of risk control instructions, the system can quickly formulate targeted traffic management plans when a high-risk situation is predicted, ensuring that the response measures are closely matched with the on-site conditions. It is also understandable that the municipal execution terminal network's classification, analysis, and collaborative driving of the risk control instruction set enables consistent action in time and space across information dissemination devices, traffic signal control systems, and on-site traffic management personnel, thereby forming a comprehensive traffic management capability covering information, traffic flow, and manpower dimensions, transforming predicted risks into controllable on-site handling.

[0046] In practical implementation, to quantify the generation and execution of risk control instruction sets, an instruction completeness rate index can be introduced, with the following formula:

[0047] in: Represents instruction completeness. This represents the number of valid instructions generated after logical consistency checks and resource conflict detection. This represents the total number of measures in the basic traffic management plan template generated by matching the risk area prediction map with the potential congestion path prediction set. The instruction completeness rate reflects the completeness of the transformation from the plan template to executable instructions, and serves as a basis for evaluating the plan's adaptability and system response capability in subsequent operations.

[0048] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A municipal dense crowd flow risk control and prediction system, characterized in that, The system includes: The information collection module deploys a group of multi-source information collection devices at key nodes within the municipal area to continuously collect information and obtain a multi-dimensional spatiotemporal pedestrian flow data set for the target municipal area within a preset collection period. The quality verification module performs data integrity and consistency verification on the multi-dimensional spatiotemporal pedestrian flow data set, and generates trajectory continuity index, density distribution rationality index, and passenger flow matching index. The weight decision module, based on the trajectory continuity index, the density distribution rationality index, and the passenger flow matching index, performs pattern matching queries in a pre-built data fusion strategy library to determine a multi-source data fusion weight scheme suitable for the target municipal area. The situation calculation module initiates a regional pedestrian flow situation calculation task based on the multi-source data fusion weighting scheme, and fuses and calculates the multi-dimensional spatiotemporal pedestrian flow data set to generate a regional pedestrian flow heat map and a group movement trend vector at the current moment. The risk extrapolation module inputs the regional pedestrian heat map and the group movement trend vector into the cloud-based risk prediction engine. It then uses the group behavior simulation model built into the cloud-based risk prediction engine to extrapolate and analyze the regional pedestrian heat map and the group movement trend vector, generating a risk area prediction map and a set of potential congestion path predictions for future periods.

2. The municipal dense population flow risk control and prediction system as described in claim 1, characterized in that, The multidimensional spatiotemporal pedestrian flow dataset is subjected to data integrity and consistency verification to generate trajectory continuity indicators, density distribution rationality indicators, and passenger flow matching indicators, including: The multi-source information acquisition equipment group includes video surveillance equipment, mobile terminal signaling base stations, and vehicle card-swiping terminals. The multidimensional spatiotemporal human flow data set includes an individual movement trajectory data set, a regional population density data set, and a transportation passenger flow data set. The individual movement trajectory data set is fragmented using the spatiotemporal pane segmentation method to obtain a set of several continuous trajectory segments. Calculate the number of trajectory interruptions and the number of average velocity mutation points for each set of continuous trajectory segments, and calculate the mean interruption rate and standard deviation of mutation for all sets of continuous trajectory segments. Combine the mean interruption rate and the standard deviation of mutation to obtain the trajectory continuity index. Spatial gridding analysis is performed on the population density data set of the region to calculate the population density value of each grid and analyze the density gradient distribution between adjacent grids. The rationality index of the density distribution is generated by combining the spatial autocorrelation coefficient of the density value and the uniformity of the gradient distribution. Extract the passenger flow sequences of each route from the passenger flow data set of the transportation vehicles, and compare them with the number of trajectories originating from or going to transportation stations in the individual movement trajectory data set in the corresponding spatiotemporal pane. Calculate the passenger flow matching degree of each route, and derive the passenger flow matching index by combining the mean and variance of the passenger flow matching degree of all routes.

3. The municipal dense population flow risk control and prediction system as described in claim 2, characterized in that, Based on the trajectory continuity index, the density distribution rationality index, and the passenger flow matching index, a pattern matching query is performed in a pre-built data fusion strategy library to determine a multi-source data fusion weighting scheme suitable for the target municipal area, including: The trajectory continuity index, the density distribution rationality index, and the passenger flow matching index are input as three-dimensional quality vectors into the data fusion strategy library; Retrieve several historical fusion strategy records with the highest cosine similarity to the three-dimensional quality vector from the data fusion strategy library, and extract the data source weight allocation scheme corresponding to each of the several historical fusion strategy records. The data source weight allocation schemes of the aforementioned historical fusion strategy records are weighted and fused based on confidence level, and the fused weight allocation scheme is used as the multi-source data fusion weight scheme.

4. The municipal dense population flow risk control and prediction system as described in claim 3, characterized in that, The construction process of the data fusion strategy library includes: We collected samples of trajectory continuity indicators, density distribution rationality indicators, and passenger flow matching indicators from multiple historical municipal area monitoring cases, along with their respective validated fusion weighting schemes. A three-dimensional strategy matching space is constructed, with trajectory continuity index as the first dimension, density distribution rationality index as the second dimension, and passenger flow matching index as the third dimension. The quality index sample of each historical case is mapped to a strategy anchor point in the three-dimensional strategy matching space, and the corresponding strategy anchor point is marked with the effective fusion weight scheme sample of the historical case to form the data fusion strategy library.

5. The municipal dense population flow risk control and prediction system as described in claim 4, characterized in that, When performing pattern matching queries in a pre-built data fusion strategy library, a strategy anchor point optimization step is also included: In the three-dimensional strategy matching space, the initial query radius is defined with the query point corresponding to the current three-dimensional quality vector as the center of the sphere; Calculate the density of all strategy anchor points within the initial query radius; If the density is lower than the preset anchor density threshold, the query radius is adaptively expanded until the density reaches the threshold, and then cosine similarity-based strategy anchor retrieval is performed within the radius.

6. The municipal dense population flow risk control and prediction system as described in claim 1, characterized in that, Based on the aforementioned multi-source data fusion weighting scheme, a regional pedestrian flow trend calculation task is initiated. This involves fusing and calculating the multi-dimensional spatiotemporal pedestrian flow data set to generate a regional pedestrian flow heat map and a group movement trend vector for the current moment, including: According to the multi-source data fusion weighting scheme, calculation weights are assigned to video surveillance data, mobile signaling data, and card swipe data respectively. Based on the calculated weights, the population counts within the same geographic grid from different data sources are weighted and summed to obtain the merged population count for each grid. Based on the population data after merging all grids, a continuous distribution heat map of population flow in the region is generated using a spatial interpolation method. The overall movement direction and speed of the group in the individual movement trajectory data set are analyzed, and the movement trend vector of the group, which represents the current main direction of the flow of people, is calculated by the vector averaging method.

7. The municipal dense population flow risk control and prediction system as described in claim 6, characterized in that, The cloud-based risk prediction engine utilizes a built-in group behavior simulation model to extrapolate and analyze the regional pedestrian heat map and the group movement trend vector, generating a risk area prediction map and a set of potential congestion path predictions for future time periods, including: The heat map of pedestrian flow in the region is discretized into the initial distribution of surrogate individuals on a spatial grid; The group movement trend vector is input into the group behavior simulation model as an environmental guiding force; In the group behavior simulation model, the movement decision of each agent individual is simulated based on the principle of social force model, under the combined effects of environmental guidance, destination attraction, interpersonal rejection and obstacle avoidance. Run the group behavior simulation model to a specified future time, count the density of agent individuals in each spatial grid, and generate a predicted density distribution map; By comparing the predicted density distribution map with the preset safety capacity thresholds for each region, the areas exceeding the limit are marked, thus forming the risk area prediction map; Paths exhibiting consistently high traffic volume and speeds below a threshold during the simulation process are identified and included in the potential congestion path prediction set.

8. The municipal dense population flow risk control and prediction system as described in claim 7, characterized in that, The process of constructing and calibrating the group behavior simulation model includes: Acquire multi-source monitoring data from historical high-density crowd scenarios to reconstruct the crowd distribution and movement vectors at historical moments; Set the initial parameters of the group behavior simulation model, including individual expected speed, interpersonal force magnitude, and field of vision; The simulation model is run using historical data as initial conditions, and the population distribution and movement status output by the simulation are compared with the historical data that actually occurred later. By optimizing the algorithm and adjusting the model parameters, the difference between the simulation results and the actual data is minimized until the model output error converges to an acceptable range.

9. The municipal dense population flow risk control and prediction system as described in claim 1, characterized in that, Also includes: The instruction execution module triggers the early warning decision core to generate a set of risk control instructions when the predicted risk level of an area in the risk area prediction map exceeds a preset risk threshold, and then sends the set of risk control instructions to the municipal execution terminal network. The risk control instruction set is parsed through the municipal execution terminal network, driving the information release device, traffic signal control system and on-site traffic control personnel to perform coordinated traffic control operations. The triggering early warning decision core generates a set of risk control instructions, including: Based on the risk area prediction map and the potential congestion path prediction set, query the municipal contingency plan database and match the applicable basic traffic management plan template. By combining real-time traffic operation status information and the availability of municipal facilities, the abstract measures in the basic traffic diversion plan template are specified and parameters are assigned to generate a draft instruction containing specific measures, execution locations and activation timing; After performing logical consistency checks and resource conflict detection on the draft instructions to ensure that the instructions are executable and free of contradictions, the final set of risk control instructions is formed.

10. The municipal dense population flow risk control and prediction system as described in claim 9, characterized in that, The risk control instruction set is parsed through the municipal execution terminal network to drive the information dissemination device, traffic signal control system, and on-site traffic management personnel to perform coordinated traffic management operations, including: The risk control instruction set is analyzed and classified into information guidance instructions, signal control instructions, and personnel dispatch instructions. Based on the content and target area of ​​the information guidance instructions, control the corresponding electronic displays, broadcasts, and mobile application push terminals to release real-time pedestrian flow guidance information; Based on the parameters of the signal control instructions, adjust the traffic light timing schemes at relevant intersections to provide priority passage for evacuation flows; The personnel dispatch instructions are sent to the on-site management personnel terminal to guide the personnel to the designated location to carry out crowd diversion and order maintenance tasks, and the execution status is fed back to the early warning decision core.

Images