Sports event audience flow intelligent prediction method and device, equipment and medium

Abstract

The invention provides a sports event audience flow intelligent prediction method and device, equipment and a medium, and the method comprises the steps: collecting the real-time people flow data of a venue through a sensing device, carrying out the space-time grid processing of the real-time people flow data through combining with an event flow of an event process, and generating a space-time grid data set; performing spatio-temporal feature fusion on the spatio-temporal grid data set through a deep neural network model, calculating a mutation coefficient of each grid, and determining the grid of which the mutation coefficient meets a preset threshold value as a core region; obtaining a building database of the venue, and constructing a path impedance matrix based on the building database; constructing a conduction model according to the core area and the path impedance matrix, predicting a people flow pressure diffusion path, and generating a diffusion path set; and based on the diffusion path set and the path impedance matrix, calculating a people flow prediction value of each region in a future time period through the residence willingness weight. By adopting the method, the flow core area in the venue can be automatically identified, and the safety management level of sports events is improved.

Description

Technical Field

[0001] The present invention belongs to the field of sports event management, and in particular relates to a method, device, equipment and medium for intelligently predicting spectator flow of sports events. Background Art

[0002] Intelligent prediction of crowd flow at sports events is a crucial research area in modern large-scale event management, directly impacting venue safety, spectator experience, and operational efficiency. With the expansion of event scale and the surge in spectator numbers, accurate prediction and effective management of crowd flow have become crucial to ensuring the smooth running of events. However, existing crowd flow prediction methods, which rely primarily on static data analysis or simple real-time monitoring, struggle to adapt to the dynamic changes in spectator behavior during events and are inadequate for identifying traffic concentrations in localized areas within venues. This often results in managers reacting slowly to sudden congestion events and lacking targeted mitigation measures.

[0003] The core challenges in this area stem from several interrelated technical factors: First, spectator traffic is highly dynamic and heterogeneous at different stages of an event. For example, during peak periods, crowd concentration areas can rapidly shift, making it difficult for traditional models to capture this spatiotemporal dynamic. Second, due to the lack of an automatic mechanism for identifying core traffic areas, existing systems are unable to accurately distinguish which areas are most prone to congestion during specific periods, limiting the accuracy of predictions. This lack of identification capability further hinders the system's ability to effectively predict the duration and impact of traffic hotspots, making it difficult for managers to formulate traffic diversion plans in advance.

[0004] However, existing technologies are unable to automatically identify the core traffic areas within the event venues due to sudden changes in audience behavior in the dynamic spatiotemporal dimension, and simultaneously predict the spatiotemporal boundaries of the congestion in this area, resulting in operational deviations in emergency responses, affecting the level of event management, and creating potential safety risks. Summary of the Invention

[0005] Based on this, it is necessary to provide a method, device, equipment and medium for intelligent prediction of audience flow for sports events to address the above technical problems. It can automatically identify the core flow area in the venue and predict its impact range and duration, providing intelligent decision-making support for the safety management of large-scale events.

[0006] In a first aspect, the present application provides a method for intelligently predicting audience flow for sports events, comprising:

[0007] Real-time crowd flow data of the venue is collected through sensor equipment, and the real-time crowd flow data is processed into spatiotemporal grids based on the event flow of the event process to generate a spatiotemporal grid dataset.

[0008] The deep neural network model is used to fuse the spatiotemporal features of the spatiotemporal grid dataset, calculate the mutation coefficient of each grid, and identify the grids whose mutation coefficient meets the preset threshold as the core area;

[0009] Obtain the building database of the venue and construct a path resistance matrix based on the building database;

[0010] Construct a conduction model based on the core area and path resistance matrix to predict the diffusion path of pedestrian pressure and generate a diffusion path set;

[0011] Based on the diffusion path set and the path resistance matrix, the predicted value of the passenger flow in each area in the future time period is calculated using the residence intention weight, where the residence intention weight is used to quantify the audience's tendency to stay in the target area.

[0012] In a second aspect, the present application also provides an intelligent prediction device for audience flow of sports events, comprising:

[0013] The data processing module is used to collect real-time crowd flow data in the venue through sensor equipment, and perform spatiotemporal grid processing on the real-time crowd flow data in combination with the event stream of the event process to generate a spatiotemporal grid dataset;

[0014] The mutation coefficient calculation module is used to perform spatiotemporal feature fusion on the spatiotemporal grid dataset through a deep neural network model, calculate the mutation coefficient of each grid, and determine the grids whose mutation coefficients meet the preset threshold as the core area;

[0015] The path resistance construction module is used to obtain the architectural database of the venue and construct the path resistance matrix based on the architectural database;

[0016] The diffusion path generation module is used to build a conduction model based on the core area and the path resistance matrix, predict the diffusion path of pedestrian pressure, and generate a diffusion path set;

[0017] The crowd flow prediction module is used to calculate the crowd flow forecast value of each area in the future time period based on the diffusion path set and the path resistance matrix, using the residence intention weight, where the residence intention weight is used to quantify the audience's tendency to stay in the target area.

[0018] In a third aspect, the present application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-mentioned method for intelligently predicting the audience flow of sports events when executing the computer program.

[0019] In a fourth aspect, the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for intelligently predicting the audience flow of sports events.

[0020] The above-mentioned intelligent prediction method, device, equipment, and medium for sports event spectator flow utilizes sensors to collect real-time venue crowd flow data and integrates it with event flow data for spatiotemporal gridding. This creates a multi-dimensional, correlated spatiotemporal grid dataset, providing a structured data foundation for dynamic perception of spectator behavior. A deep neural network model is used to integrate the spatiotemporal features of the grid dataset, accurately calculating the mutation coefficient of each grid and automatically identifying core areas of spectator concentration based on preset thresholds. A path resistance matrix is constructed based on the venue architectural database to quantify the effect of physical paths on crowd flow transmission. A transmission model is constructed based on the core areas and the path resistance matrix to simulate the diffusion paths of crowd flow pressure and generate a set of diffusion paths to clarify the impact range of the core areas. Predicted crowd flow values for each area in the future time period are calculated using retention intention weights, quantifying the dynamic impact of spectator retention tendency on the duration of pressure, thereby providing a real-time decision-making basis for venue flow warnings and diversion strategy generation. This technical solution achieves automated identification of venue flow core areas and accurate prediction of diffusion paths and crowd flow, effectively addressing the technical challenges of dynamic monitoring and decision support in the safety management of large-scale events. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 A schematic diagram of a flow chart of a method for intelligently predicting spectator traffic for sports events provided by an embodiment of the present invention;

[0023] Figure 2 A schematic diagram of a process for generating a spatiotemporal grid dataset provided by an embodiment of the present invention;

[0024] Figure 3 A schematic diagram of the structure of an intelligent prediction device for sports event spectator flow provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0026] First, a brief introduction is given to the terms involved in the embodiments of this application.

[0027] The mutation coefficient is a key indicator used to quantify the intensity of sudden changes in crowd behavior within a specific area. It comprehensively reflects the dynamic trends of crowd gathering, evacuation, or sudden changes in flow direction by analyzing the rate of change of crowd density and movement direction vectors between adjacent time periods in spatiotemporal data. This coefficient transforms complex crowd behavior characteristics into a scalar value that can be quantified using thresholds, providing an objective basis for identifying core areas. It addresses the technical limitations of traditional methods, which rely on manual judgment and have difficulty capturing instantaneous traffic flow changes in real time.

[0028] The path resistance matrix is a quantitative model that characterizes the degree of obstruction to pedestrian flow across various paths within a venue. It is constructed by coupling physical path characteristics (such as path length and aisle width) with environmental factors (such as the attractiveness of commercial facilities). This matrix defines the resistance between paths using a grid as its unit. It quantifies the combined resistance to spectator movement caused by both physical path properties (such as narrow aisles increasing navigation difficulty) and external factors (such as the distribution of shops guiding crowd flow). This provides a multidimensional data foundation for dynamically predicting crowd flow direction and congestion risks, addressing the technical issues of traditional path analysis methods, which often ignore environmental dynamics and fail to reflect actual traffic trends.

[0029] Based on the above explanations, the implementation environment of a method for intelligently predicting spectator traffic at sports events provided in an embodiment of the present application is described. Schematically, the implementation environment includes: a terminal, a sensor device, a processor, and a memory. The terminal is connected to the processor, sensor device, and memory via a network signal; the sensor device includes, but is not limited to, a lidar, a WiFi probe, an infrared thermal imager, and a video surveillance camera; the processor can be a central processing unit (CPU), a graphics processing unit (GPU), or an artificial intelligence chip (such as an NPU or TPU); and the memory can be a distributed cloud storage system or a local server cluster, which is not limited here.

[0030] In combination with the above explanations of terms and implementation environments, the application scenarios of the embodiments of this application are explained. The intelligent prediction method for sports event audience flow provided in the embodiments of this application can be applied to, but not limited to, the following scenarios:

[0031] At international sporting events, peak spectator traffic can reach hundreds of thousands, creating immense pressure for instantaneous crowd accumulation and evacuation. This solution collects real-time crowd flow data from entrance gates, stands, and aisles, dynamically predicts core gathering areas based on event progress (e.g., goals, timeouts, and the end of the match), and generates multi-level evacuation routes based on a path resistance matrix. Based on the evacuation instructions generated by the system, security personnel can pre-deploy emergency channels and adjust security checkpoint diversion strategies to effectively prevent stampedes. Furthermore, a retention intention model is used to reduce congestion in commercial areas and enhance the spectator experience.

[0032] Sports stadiums contain numerous functional areas, such as dining, rest areas, and souvenir sales areas. Their layout and resource allocation must be dynamically adjusted based on crowd flow. By monitoring and predicting crowd flow in each area in real time, combined with a weighted analysis of visitor retention, venue operators can anticipate spectator retention trends and optimize staffing, cargo replenishment plans, and advertising placement strategies. For example, if sustained high crowds are predicted in the core competition area, surrounding dining areas can prepare for peak hours and allocate supplies to areas expected to see subsequent surges, thereby improving overall venue operational efficiency.

[0033] During sporting events, traffic pressure around venues surges. The concentrated outflow of private cars and public transportation can paralyze regional traffic. This method can predict the diffusion paths of spectators leaving the venues in advance, generating a collection of these paths and feeding it into the city's traffic management system. Based on this information, traffic management departments can dynamically adjust traffic light timings around the venues, implement temporary traffic controls, or increase public transportation services. This effectively alleviates traffic congestion caused by the mass outflow after the event and ensures the overall efficiency of urban transportation.

[0034] Illustratively, the method for intelligently predicting the audience flow of sports events provided in the embodiments of the present application can also be applied to other application scenarios. This is only provided as an example and is not limited to a specific application scenario.

[0035] In an exemplary embodiment, Figure 1 As shown, a method for intelligently predicting the audience flow of sports events is provided. This embodiment uses the method applied to a terminal in the aforementioned implementation environment as an example. It is understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps 101 to 105:

[0036] Step 101: collect real-time crowd flow data in the venue through sensor equipment, and perform spatiotemporal grid processing on the real-time crowd flow data in combination with the event stream of the event process to generate a spatiotemporal grid dataset.

[0037] For example, real-time crowd flow data in a venue can be collected through sensor equipment. There are many types of sensor equipment, such as infrared thermal imaging equipment installed in key locations such as the main passages, entrances and exits, and stands. They can sense the heat distribution of the human body and thus obtain information on the flow of people passing through the corresponding areas; or surveillance cameras with video analysis functions installed in the venue can use image recognition algorithms to count and track people in the video footage, thereby obtaining real-time crowd flow data; combined with the event stream of the event process, that is, according to the real-time situation of the event, classification and labeling are carried out according to the event characteristics of different stages, such as the peak period of spectator entry before the start of the game, the moments when the exciting moments of the game trigger large-scale audience interaction, the period of game pauses, and the period when the spectators leave the venue after the game, and the collected real-time crowd flow data is processed in time and space. Specifically, the space of the venue is divided into several grid cells with a certain accuracy, and each grid cell corresponds to a specific area in the venue; in the time dimension, the entire event duration is divided into multiple time periods at preset time intervals. The real-time crowd flow data is assigned to the corresponding space-time grid to generate a space-time grid dataset. This dataset completely records the crowd flow distribution in various areas of the venue at different time points during the event, providing basic data support for further analysis and prediction.

[0038] Step 102: Use a deep neural network model to fuse the spatiotemporal features of the spatiotemporal grid dataset, calculate the mutation coefficient of each grid, and determine the grids whose mutation coefficients meet a preset threshold as core areas.

[0039] Specifically, after obtaining the generated spatiotemporal grid dataset, it is input into a deep neural network model for spatiotemporal feature fusion. The deep neural network model can employ a convolutional neural network (CNN) combined with a long short-term memory (LSTM) network architecture. The CNN extracts spatial features from the grid data, such as identifying areas with relatively high pedestrian density and the clustering patterns of pedestrian flows. The LSTM excels at capturing long-term dependencies in time series data, thereby analyzing trends in pedestrian flow over different time periods, such as the rate of increase or decrease in pedestrian density. Through the synergistic effect of these two networks, the spatiotemporal features of each grid in the spatiotemporal grid dataset are fused. The fused feature data is used to calculate the mutation coefficient for each grid, which reflects the magnitude of changes in pedestrian flow within a grid within a specific time and spatial range. The calculated mutation coefficient is compared with a preset threshold. If the mutation coefficient exceeds the threshold, the grid is identified as a core area. The core area is the key area in the venue where the flow of people gathers at a high level, changes rapidly and may have a greater impact on the safety of the event. By accurately identifying the core area, management personnel can focus on these areas and take corresponding safety management measures in a timely manner.

[0040] Step 103: Obtain the building database of the venue, and construct a path resistance matrix based on the building database.

[0041] Specifically, the venue's architectural structure information is obtained from its architectural database, including the layout of various areas within the venue, the width and length of passageways, connectivity between different areas, and the distribution of obstacles. For example, a network analysis algorithm can be used to construct a path resistance matrix: the venue's space is abstracted into a network model consisting of nodes and edges. Nodes represent key locations within the venue, such as entrances and exits, stands, rest areas, and dining areas; edges represent passages between nodes. Based on the physical properties of the passages, such as width, length, slope, and whether there are obstacles, each edge is assigned a resistance value. This resistance value represents the amount of resistance encountered by spectators moving through the passage. By calculating the shortest path resistance between each node, a complete path resistance matrix is constructed. This matrix intuitively reflects the connectivity between different areas within the venue, providing a key basis for subsequently predicting crowd pressure diffusion paths and helping to analyze spectator movement trends and potential gathering areas within the venue.

[0042] Step 104 : construct a conduction model based on the core area and the path resistance matrix, predict the diffusion path of the pedestrian flow pressure, and generate a diffusion path set.

[0043] Specifically, a conduction model is constructed based on the identified core areas and the constructed path resistance matrix. This model is based on the principle of diffusion in physics. The core area is considered the source of crowd pressure, and its inherent crowd pressure diffuses to surrounding areas through the venue's channels. In the conduction model, the intensity of crowd pressure in the core area is correlated with the area's mutation coefficient. A higher mutation coefficient indicates greater crowd pressure in the core area and a stronger potential for outward diffusion. Furthermore, the resistance value in the path resistance matrix determines the diffusion rate and path selection of crowd pressure along different channels. Channels with lower resistance values are more likely to become the primary paths for crowd pressure diffusion. By simulating the conduction model, possible crowd pressure diffusion paths are predicted and a diffusion path set is generated. This diffusion path set details the channels along which audience flow from the core area is likely to flow to other areas, as well as the order and coverage of diffusion. This allows management personnel to proactively identify potential areas of crowd gathering risk and provides strong support for developing scientific and reasonable evacuation strategies.

[0044] Step 105 , based on the diffusion path set and the path resistance matrix, the predicted value of the flow of people in each area in the future time period is calculated by the stay intention weight, wherein the stay intention weight is used to quantify the audience's tendency to stay in the target area.

[0045] Specifically, based on the diffusion path set and the path resistance matrix, the flow of visitors to each area in the future time period is predicted by calculating the retention intention weight. The retention intention weight reflects the spectator's tendency to stay in different target areas within the venue. Its calculation method comprehensively considers various factors, such as the type of target area, the impact of the event flow on spectator behavior, and individual spectator attributes, and assigns a corresponding retention intention weight to each area within the venue. By combining the information in the diffusion path set with the path resistance matrix, the flow trends and possible retention areas of spectators along different diffusion paths are analyzed. The retention intention weight is then incorporated into this analysis process, and by establishing a mathematical model, such as a prediction model based on probability statistics or a regression model in machine learning, the predicted flow of visitors to each area in the future time period is calculated. This prediction value can provide accurate spectator flow information to event organizers, venue operators, and security departments, helping them make various preparations in advance, such as rationally arranging staff, optimizing resource allocation, formulating personalized service strategies, and strengthening safety precautions, thereby effectively improving the operational management level and safety assurance capabilities of sports events, ensuring the smooth progress of the events and a good experience for the audience.

[0046] The above-mentioned intelligent prediction method for sports event spectator flow uses real-time sensor data to collect venue crowd flow data and integrates it with event flow data for spatiotemporal gridding. This creates a multi-dimensional, correlated spatiotemporal grid dataset, providing a structured data foundation for dynamic perception of spectator behavior. A deep neural network model is used to integrate the spatiotemporal features of the grid dataset, accurately calculating the mutation coefficient of each grid and automatically identifying core areas of spectator concentration based on preset thresholds. A path resistance matrix is constructed based on the venue architectural database to quantify the effect of physical paths on crowd flow transmission. A conduction model is constructed based on the core areas and the path resistance matrix to simulate the diffusion paths of crowd flow pressure and generate a set of diffusion paths to clarify the impact range of the core areas. Predicted crowd flow values for each area in the future time period are calculated using retention intention weights. This quantifies the dynamic impact of spectator retention tendency on the duration of pressure, providing a real-time decision-making basis for venue flow warnings and diversion strategy generation. This technical solution achieves automated identification of venue flow core areas and accurate prediction of diffusion paths and crowd flow, effectively addressing the technical challenges of dynamic monitoring and decision support in the safety management of large-scale events.

[0047] like Figure 2 As shown, in a possible embodiment, real-time crowd flow data of the venue is collected by sensing equipment, and the real-time crowd flow data is processed into a spatiotemporal grid in combination with the event stream of the event process to generate a spatiotemporal grid dataset, including:

[0048] Step 201 : Divide the venue into monitoring grids according to preset sizes, collect real-time coordinate data through laser radar, and generate an initial crowd distribution matrix, wherein the initial crowd distribution matrix is used to record the real-time crowd density of each grid.

[0049] Specifically, the stadium is divided into monitoring grids of preset dimensions. These dimensions are determined based on the venue's size, expected monitoring accuracy, and data processing capabilities. LiDAR equipment is deployed at key locations within each monitoring grid, such as the grid center or at intersections of major corridors. LiDAR emits laser beams and receives reflected signals, acquiring real-time three-dimensional coordinate data of objects within the monitoring range. After preprocessing these coordinates through operations such as denoising and filtering, an initial crowd distribution matrix is generated. This matrix is presented as a two-dimensional table, with rows and columns corresponding to the rows and columns of the monitoring grids within the stadium, respectively. The value of each element represents the crowd density within the corresponding grid, calculated by counting the number of people per unit area or by estimating the density based on the density of point cloud data. For example, a counting method can be used to directly count the number of valid reflection points within each grid, converting this number into an estimated number of people, and then calculating the density. Alternatively, a clustering algorithm can be used to cluster the point cloud data into crowd groups, and density can be calculated based on group size and distribution. The resulting initial crowd distribution matrix clearly displays the real-time crowd density of each area within the stadium, providing basic and accurate quantitative data for subsequent crowd analysis.

[0050] In step 202 , the device movement trajectory is analyzed by the WiFi probe, the movement direction vector of each grid is calculated, and movement direction distribution data is generated. The movement direction distribution data is used to characterize the dynamic changes in the movement direction of the audience in each grid.

[0051] Specifically, WiFi probe devices are deployed in the venue to scan and collect information such as the MAC addresses, signal strengths, and appearance times of surrounding WiFi devices (such as smartphones and tablets carried by spectators). The WiFi probe parses the collected device movement trajectories and infers the device's movement path by analyzing the signal strength change sequence and timestamp information of the device at different locations. For each monitoring grid, the moving direction vector of the grid is calculated based on the parsed movement path. The specific calculation method can adopt the vector superposition method. First, the moving direction vector of each WiFi device in the grid is determined, and then the vectors of all devices in the same grid are superimposed to obtain a comprehensive moving direction vector. The direction of the vector represents the mainstream movement direction of the audience in the grid, and the length of the vector reflects the intensity or consistency of the movement. For example, in a grid, if most spectators move in the northeast direction, the vector direction is biased to the northeast and the length is longer; if the spectators' movement directions are scattered, the vector length is shorter. The generated movement direction distribution data, combined with the initial crowd flow distribution matrix, not only provides crowd flow information within each grid, but also reveals the dynamic movement trends of the crowd, enabling managers to understand the audience's flow behavior more intuitively, such as discovering that the crowd in a certain area is rapidly gathering or evacuating.

[0052] Step 203 : Based on the initial crowd flow distribution matrix and movement direction distribution data, the event flow of the competition process is accessed, the event types are weighted and coded, and a spatiotemporal grid dataset with competition event tags is generated.

[0053] Specifically, the event stream of a sports event is connected to the data processing system. The event stream covers key events during the game, such as the start, goal, pause, substitution, halftime, and end of the game. These event types are weighted, with weights determined based on their potential impact on spectator behavior and emotions. For example, a goal event, likely to trigger strong emotional reactions and behavioral changes, may be weighted higher; while a pause event, with its relatively small impact on spectator behavior, may be weighted lower. Based on the initial crowd flow distribution matrix and movement direction distribution data, the weighted event information from the event stream is fused to generate a spatiotemporal grid dataset with event event markers. During the fusion process, the temporal and spatial consistency and correspondence of each data dimension are ensured. Specifically, LiDAR data, WiFi probe analysis data, and event event marker information within the same monitoring grid at the same time point are precisely correlated and integrated. The resulting spatiotemporal grid dataset not only contains crowd flow density and movement direction information for each monitoring grid within the venue at different times but also incorporates the influencing factors of event events, thereby more accurately reflecting the dynamic changes in spectator flow. For example, when exciting moments in a match (high-weight events) occur, the grid data at the corresponding time point in the dataset can reflect the audience's possible behavioral characteristics, such as standing up and cheering, and gathering in specific areas. This provides comprehensive and context-relevant data support for subsequent crowd flow prediction and management, helping managers to predict potential crowd flow peaks and risk areas in advance, formulate more targeted management strategies, and improve the safety and efficiency of event operations.

[0054] In a possible embodiment, a deep neural network model is used to fuse spatiotemporal features of a spatiotemporal grid dataset, calculate the mutation coefficient of each grid, and screen grids whose mutation coefficients meet a preset threshold to determine them as core areas, including:

[0055] For the crowd density data in the spatiotemporal grid dataset, time series features are extracted through the long short-term memory network to generate a potential state vector.

[0056] Specifically, the crowd density data in the spatiotemporal grid dataset is processed, and a long short-term memory (LSTM) network is used to extract time series features. An LSTM network consists of an input layer, a hidden layer, and an output layer. The hidden layer contains memory cells, which effectively capture long-term dependencies in time series. The crowd density data for each grid is input into the LSTM network in chronological order. The network learns and processes the data through its internal gating mechanisms (input gate, forget gate, and output gate). For example, the input gate determines which new information is stored in the memory cell, the forget gate controls which information in the memory cell is discarded, and the output gate determines which information in the memory cell is output. The trained LSTM network can extract the latent time series features in the crowd density data and generate a latent state vector. This latent state vector is a fixed-length vector that contains the time series features of the crowd density of the corresponding grid, such as the changing trend and periodicity of the crowd density. It provides a temporal feature representation for subsequent spatiotemporal feature fusion.

[0057] Based on the venue topology map of the building database, the spatial features of the potential state vector and movement direction distribution data are aggregated through the graph neural network to generate spatiotemporal fusion features.

[0058] Specifically, based on the venue topology map in the building database, a graph neural network (GNN) is used to perform spatial feature aggregation on the latent state vector and movement direction distribution data. The venue topology map describes the spatial relationship between each grid within the venue, including information such as adjacency and distance. The GNN treats each grid as a node in the graph, and the connection relationship between nodes is determined by the venue topology map. The latent state vector and movement direction distribution data are input into the GNN as the initial feature vector of the node. The GNN propagates and aggregates feature information between nodes through a message passing mechanism. For example, during the aggregation process, each grid node receives feature information from its neighboring grid nodes and updates it based on its own feature information. After multiple iterations, the GNN generates spatiotemporal fusion features that comprehensively consider time series features and spatial relationships. This feature can comprehensively reflect the pedestrian flow status of each grid and its mutual influence with surrounding grids, providing a richer information basis for subsequent mutation coefficient calculation.

[0059] The mutation coefficient of the spatiotemporal fusion feature is calculated using the following formula to obtain the mutation coefficient matrix:

[0060] α=σ(w d ΔD+w v ΔV)

[0061] Among them, α is the mutation coefficient, ΔD is the difference in the flow density of adjacent time windows, ΔV is the rate of change of the moving direction vector, and w dis the preset crowd density weight coefficient, w v is the preset speed weight coefficient, and σ is the activation function.

[0062] Specifically, the mutation coefficient reflects the degree of mutation of the crowd flow state within the grid; the difference in crowd density between adjacent time windows indicates the drastic change in crowd density in a short period of time; the rate of change of the movement direction vector reflects the speed of change in the audience's movement direction; the preset crowd density weight coefficient and movement speed weight coefficient are used to adjust the relative importance of the two in the calculation; the activation function can be a si gmo id function, which maps the calculation results to the (0,1) interval to make the mutation coefficient more interpretable and comparable.

[0063] The mutation coefficient matrix is threshold-screened to obtain the core area, where the core area is used to characterize the instantaneous aggregation hotspots formed by sudden changes in audience behavior.

[0064] Specifically, a mutation coefficient threshold is set, and grids with mutation coefficients exceeding the threshold are screened. Spatially adjacent screening results are then merged to form core regions, representing instantaneous spectator gathering hotspots triggered by event events or environmental factors. This core region, identified by this method, has clear spatial location and mutation level information. This helps managers quickly locate potential risk areas and implement timely measures, such as increasing security personnel and conducting crowd control. This effectively improves the efficiency and accuracy of event safety management, ensuring the smooth progress of the event and the safety of spectators.

[0065] In a possible embodiment, obtaining a building database of a venue and constructing a path resistance matrix based on the building database includes:

[0066] The path length and channel width of the grid in the core area are extracted from the building database to generate basic resistance parameters.

[0067] Specifically, the path lengths and channel widths of the grids in the core area are extracted from the venue's architectural database to generate basic impedance parameters. Furthermore, for channels between grids with a winding path, the centerline of the channel can be fitted using coordinate points, and the equivalent path length between adjacent grids can be calculated based on this centerline. The channel width can be the width at its narrowest point or the average width, depending on the venue's actual traffic conditions. These path length and channel width data are compiled into basic impedance parameters between corresponding grids, providing a direct and accurate data source for generating these basic impedance parameters and ensuring the accuracy of the subsequent path impedance matrix construction.

[0068] Based on the commercial facility distribution data in the building database, the commercial attraction coefficient of each grid is calculated to generate an attraction coefficient matrix.

[0069] Specifically, based on the commercial facility distribution data in the building database, the number, type, and area share of shops within each grid are counted, and a weighted calculation is performed to generate a commercial attraction coefficient. For example, a weight of 0.7 is assigned to restaurants and 0.3 to retail shops. The grid's attractiveness value is then calculated based on the area share. The same calculation is performed for all grids to form an attraction coefficient matrix, which quantifies the effectiveness of commercial facilities in guiding pedestrian flow.

[0070] The path resistance matrix is constructed based on the basic resistance parameters and the attraction coefficient matrix using the following formula:

[0071]

[0072] Among them, R ij is the path impedance matrix, L ij is the path length, W ij is the channel width, β is the commercial attractiveness adjustment coefficient, S j is the commercial attractiveness coefficient.

[0073] Specifically, the basic resistance parameters are combined with the attraction coefficient matrix to calculate the path resistance between each grid using a resistance formula. In this formula, path length is positively correlated with resistance, while channel width is negatively correlated with attraction coefficient. The commercial attraction adjustment coefficient is used to balance the impact of commercial facilities on pedestrian flow. For example, if a path's terminal grid contains a highly attractive store, its resistance value will decrease, indicating a tendency for pedestrian flow to spread toward that area. This ultimately generates a path resistance matrix, describing the comprehensive traffic resistance of each path within the venue.

[0074] In a possible embodiment, a conduction model is constructed based on the core area and the path resistance matrix to predict the diffusion path of pedestrian flow pressure and generate a diffusion path set, including:

[0075] Taking the core area as the pressure source and based on the path resistance matrix, the pressure conduction process is initialized and simulated through the fluid mechanics model to generate the initial pressure distribution map.

[0076] Specifically, the core area is used as the initial pressure source, and the path resistance matrix is input into a fluid dynamics model to simulate the pressure transmission process. Based on the resistance distribution, the fluid dynamics model calculates the rate and direction of pressure diffusion in the venue's path network, generating an initial pressure distribution map. This map plots pressure intensity on a grid-like basis, reflecting the initial trend of pedestrian flow spreading outward from the core area. For example, high-resistance paths correspond to areas of slow pressure transmission, while low-resistance paths indicate rapid pressure release.

[0077] According to the initial pressure distribution map, the pressure gradient between adjacent grids is calculated by the finite difference method to generate a set of candidate diffusion paths.

[0078] Specifically, based on the initial pressure distribution map, the finite difference method is used to calculate the pressure gradient between adjacent grids to determine the direction and strength of pressure conduction. Based on the gradient direction, a set of candidate diffusion paths extending outward from the core region is generated, with each path consisting of consecutive grid nodes. For example, a path extending in the direction of the fastest pressure gradient decreases, passing through multiple low-resistance grids, forms the primary conduction direction.

[0079] For each path in the candidate diffusion path set, the total impedance value is calculated based on the path impedance matrix, and the paths whose total impedance exceeds a preset threshold are eliminated to generate a preliminary diffusion path set.

[0080] Specifically, for each path in the candidate diffusion path set, the impedance values of the paths passing through the grid are accumulated to calculate the total impedance. A total impedance threshold is set, and paths exceeding the threshold are eliminated. Paths with high conduction efficiency are retained to form the preliminary diffusion path set. For example, a path with a high total impedance due to passing through a narrow channel is identified as an inefficient path and removed.

[0081] Physical conflict detection is performed on the preliminary diffusion path set to generate a diffusion path set, wherein the diffusion path set includes a main conduction path and at least one secondary conduction path.

[0082] Specifically, the initial diffusion path set is checked for physical conflicts to verify compliance with the venue's channel capacity constraints and topological connectivity rules. If channel capacity exceeds the limit or a path cross-conflict is detected, the path's direction is automatically adjusted or split into primary and secondary paths to generate the final diffusion path set. For example, if a primary path triggers a conflict due to insufficient channel capacity, a secondary path detouring around the service area is generated as a supplementary solution.

[0083] In a possible embodiment, based on the diffusion path set and the path resistance matrix, the predicted value of the flow of people in each area in the future period is calculated by the residence intention weight, including:

[0084] The commercial attraction coefficient of the terminal grid in the diffusion path set is extracted to generate the attraction parameter of the target area.

[0085] Specifically, the commercial attraction coefficient of the terminal grid is extracted from the diffusion path set and combined with the commercial facility distribution data in the building database to generate a target area attraction parameter. This parameter quantifies the strength of the terminal grid's attraction to visitors due to the distribution of commercial facilities. For example, if a path terminal grid contains a restaurant and souvenir shop, its commercial attraction coefficient is high. The generated target area attraction parameter reflects the tendency of visitors to stay in this area.

[0086] Based on the weight of the event marker and the target area attraction parameter, the preset commercial attraction ratio coefficient and event weight ratio coefficient are weighted and summed to generate the audience's stay intention weight.

[0087] Specifically, based on the weights of event markers (e.g., a goal event weight of 0.8 and a game-ending event weight of 0.5) and the target area's attractiveness parameter, a linear weighted summation of the preset commercial attractiveness ratio and event weight ratio is performed to generate a spectator's intended stay weight. For example, when a goal event is triggered in a certain area, its event weight and commercial attractiveness parameter are weighted at a ratio of 0.6 and 0.4, respectively, to generate a comprehensive intended stay indicator.

[0088] The path conduction pressure values are averaged according to the path impedance matrix to generate a reference conduction pressure value.

[0089] Specifically, the path conduction pressure values are averaged according to the path resistance matrix to generate a benchmark conduction pressure value. The path resistance matrix contains the resistance information of different paths in the venue to reflect the difficulty of the audience moving on each path. The path conduction pressure value is obtained in the pressure conduction model based on the conduction of crowd pressure on the path. When calculating the mean of the path conduction pressure value, a simple arithmetic average method can be used, that is, all path conduction pressure values are added together and divided by the number of paths; or a weighted average method can be used, which assigns different weights according to factors such as the importance of the path or the density of crowds and then performs the calculation. The generated benchmark conduction pressure value represents the average intensity of crowd pressure conduction in the venue, provides a basic pressure reference value for subsequent crowd flow prediction, and helps to evaluate the overall crowd pressure level.

[0090] The baseline conduction pressure value and the residence intention weight are dynamically modified to generate the predicted value of the passenger flow in each area in the future period.

[0091] Specifically, the dynamic correction model inputs the baseline transmission pressure value and the retention intention weight into the model. The baseline value is adjusted using linear weighting to generate a predicted flow rate for each area over the next period. For example, if the baseline transmission pressure value for a certain area is 100 people / minute and the retention intention weight is 0.3, the corrected prediction value is adjusted to 130 people / minute, accurately reflecting the cumulative flow effect caused by spectator retention.

[0092] In a possible embodiment, the method further includes:

[0093] Based on the predicted value of passenger flow, a dynamic weight decision tree is constructed to perform multi-dimensional weight allocation on the number of paths in the diffusion path set, the remaining capacity of the grid, and the weight of the willingness to stay, and generate a comprehensive diversion priority. Among them, the remaining capacity of the grid is the difference between the maximum carrying capacity of the area defined by the venue database and the real-time passenger flow data.

[0094] Specifically, based on predicted passenger flow, a decision tree algorithm is used to assign multi-dimensional weights to the number of paths in the diffusion path set, the remaining grid capacity, and the intended visitor weight, generating a comprehensive diversion priority. The number of paths reflects potential congestion risk, the remaining grid capacity is calculated by subtracting the real-time passenger flow from the maximum capacity of the area defined in the venue database, and the intended visitor weight is trained using historical behavioral data.

[0095] When the comprehensive diversion priority exceeds the preset response threshold, the total impedance of the candidate diversion path is calculated through the path impedance matrix to generate the optimal diversion path solution.

[0096] Specifically, when the comprehensive diversion priority exceeds the preset response threshold, it indicates that the current flow of people in the venue requires emergency diversion. At this time, the total impedance of the candidate diversion path is calculated through the path resistance matrix. The path resistance matrix contains the impedance information of each path in the venue, reflecting the difficulty of the audience passing through each path. The total impedance calculation method is to perform a weighted summation of the impedance values of each segment of the path included in the candidate diversion path. The weight can be determined based on factors such as path length and channel width. The path with the smallest total impedance is selected as the optimal diversion path solution. Such a path usually has a relatively wide channel, a short path length and a smaller commercial attraction impedance, which can divert the flow of people more quickly and efficiently, reduce the pressure of the flow of people in the venue, improve evacuation efficiency, and ensure the safety of the audience.

[0097] Based on the optimal evacuation path plan, dynamic guidance instructions are generated. The dynamic guidance instructions are used to instruct the path indication information of the electronic guidance screen to be updated in real time.

[0098] Specifically, based on the optimal evacuation path plan, dynamic guidance instructions are generated that include path direction identification and evacuation rate recommendations. The instructions are transmitted to the electronic guidance screen through a wireless communication protocol, and the display content is updated in real time.

[0099] In summary, the intelligent prediction method for spectator flow at sports events provided in the embodiment of the present application collects real-time crowd flow data of the venue through sensing equipment, and performs spatiotemporal grid processing in combination with the event flow of the event process to generate a spatiotemporal grid data set. The spatiotemporal feature fusion of the spatiotemporal grid data set is performed using a deep neural network model, and the mutation coefficient of each grid is calculated, thereby automatically identifying the core areas of crowd flow in the venue. A path resistance matrix is constructed based on the venue building database, and a conduction model is constructed in combination with the core area information to predict the diffusion path of crowd pressure and generate a diffusion path set. Furthermore, the predicted value of crowd flow in each area in the future time period is calculated by the weight of the willingness to stay. In addition, the method can also construct a dynamic weight decision tree based on the predicted value of crowd flow, generate a comprehensive diversion priority, and calculate the optimal diversion path plan through the path resistance matrix when necessary, and generate dynamic guidance instructions to update the path indication information of the electronic guidance screen. The above technical solution realizes real-time monitoring of audience flow in the venue, automatic identification of core areas, and accurate prediction of crowd diffusion trends, effectively reducing the need for manual intervention. It can automatically identify the core areas of traffic in the venue and accurately predict their impact range and duration, providing comprehensive, scientific and intelligent decision-making support for the safety management of large-scale events.

[0100] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0101] Based on the same inventive concept, the present application also provides an intelligent sports event spectator flow prediction device 10 for implementing the aforementioned intelligent sports event spectator flow prediction method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the intelligent sports event spectator flow prediction device 10 provided below can be found in the above-mentioned limitations of the intelligent sports event spectator flow prediction method, and will not be repeated here.

[0102] In an exemplary embodiment, Figure 3 As shown, a sports event spectator flow intelligent prediction device 10 is provided, comprising:

[0103] The data processing module 11 is used to collect real-time crowd flow data in the venue through sensor equipment, and perform spatiotemporal grid processing on the real-time crowd flow data in combination with the event flow of the event process to generate a spatiotemporal grid data set.

[0104] The mutation coefficient calculation module 12 is used to perform spatiotemporal feature fusion on the spatiotemporal grid dataset through a deep neural network model, calculate the mutation coefficient of each grid, and determine the grid whose mutation coefficient meets the preset threshold as the core area.

[0105] The path resistance construction module 13 is used to obtain the building database of the venue and construct a path resistance matrix based on the building database.

[0106] The diffusion path generation module 14 is used to construct a conduction model based on the core area and the path resistance matrix, predict the diffusion path of the pedestrian pressure, and generate a diffusion path set.

[0107] The crowd flow prediction module 15 is used to calculate the crowd flow prediction value of each area in the future time period based on the diffusion path set and the path resistance matrix by the residence intention weight, wherein the residence intention weight is used to quantify the audience's tendency to stay in the target area.

[0108] In a possible embodiment, the data processing module 11 includes:

[0109] The traffic grid division unit is used to divide the monitoring grid into preset sizes within the venue, collect real-time coordinate data through lidar, and generate an initial crowd distribution matrix, where the initial crowd distribution matrix is used to record the real-time crowd density of each grid.

[0110] The movement direction analysis unit is used to analyze the device movement trajectory through the WiFi probe, calculate the movement direction vector of each grid, and generate movement direction distribution data, where the movement direction distribution data is used to characterize the dynamic changes in the audience's movement direction in each grid.

[0111] The event marking unit is used to access the event flow of the event process based on the initial crowd distribution matrix and movement direction distribution data, perform weight encoding on the event type, and generate a spatiotemporal grid dataset with event markers.

[0112] In a possible embodiment, the mutation coefficient calculation module 12 includes:

[0113] The feature extraction unit is used to extract time series features from the crowd density data in the spatiotemporal grid dataset through the long short-term memory network to generate a potential state vector.

[0114] The feature aggregation unit is used to aggregate the spatial features of the potential state vector and movement direction distribution data based on the venue topology map of the building database through the graph neural network to generate spatiotemporal fusion features.

[0115] The mutation coefficient generation unit is used to calculate the mutation coefficient of the spatiotemporal fusion feature using the following formula to obtain the mutation coefficient matrix:

[0116] α=σ(w d ΔD+w v ΔV)

[0117] Among them, α is the mutation coefficient, ΔD is the difference in the flow density of adjacent time windows, ΔV is the rate of change of the moving direction vector, and w d is the preset crowd density weight coefficient, w v is the preset speed weight coefficient, and σ is the activation function.

[0118] The core area screening unit is used to perform threshold screening on the mutation coefficient matrix to obtain the core area, where the core area is used to characterize the instantaneous aggregation hotspot formed by the sudden change of audience behavior.

[0119] In a possible embodiment, the path impedance building module 13 includes:

[0120] The basic resistance unit is used to extract the path length and channel width of the grid in the core area from the building database to generate basic resistance parameters.

[0121] The attraction calculation unit is used to calculate the commercial attraction coefficient of each grid based on the commercial facility distribution data in the building database and generate an attraction coefficient matrix.

[0122] The resistance matrix generation unit is used to construct the path resistance matrix based on the basic resistance parameters and the attraction coefficient matrix using the following formula:

[0123]

[0124] Among them, R ij is the path impedance matrix, L ij is the path length, W ij is the channel width, β is the commercial attractiveness adjustment coefficient, S j is the commercial attractiveness coefficient.

[0125] In a possible embodiment, the diffusion path generation module 14 includes:

[0126] The pressure simulation unit is used to use the core area as the pressure source, based on the path resistance matrix, and initialize the simulation of the pressure conduction process through the fluid mechanics model to generate an initial pressure distribution map.

[0127] The gradient calculation unit is used to calculate the pressure gradient between adjacent grids through the finite difference method according to the initial pressure distribution map and generate a set of candidate diffusion paths.

[0128] The path screening unit is used to calculate the total impedance value of each path in the candidate diffusion path set based on the path impedance matrix, eliminate the paths whose total impedance exceeds a preset threshold, and generate a preliminary diffusion path set.

[0129] The conflict detection unit is used to perform physical conflict detection on the preliminary diffusion path set to generate a diffusion path set, wherein the diffusion path set includes a main conduction path and at least one secondary conduction path.

[0130] In a possible embodiment, the crowd flow prediction module 15 includes:

[0131] The attraction parameter unit is used to extract the commercial attraction coefficient of the terminal grid in the diffusion path set and generate the attraction parameter of the target area.

[0132] The residence weight unit is used to calculate the weighted sum of the preset commercial attraction ratio coefficient and event weight ratio coefficient based on the weight of the event mark and the target area attraction parameter, and generate the audience's residence intention weight.

[0133] The reference pressure unit is used to calculate the mean of the path conduction pressure values according to the path resistance matrix to generate a reference conduction pressure value.

[0134] The dynamic correction unit is used to dynamically correct the baseline conduction pressure value and the residence intention weight to generate the predicted value of the passenger flow in each area in the future period.

[0135] In a possible embodiment, the device further includes:

[0136] The priority generation unit is used to construct a dynamic weight decision tree based on the predicted value of passenger flow, perform multi-dimensional weight allocation on the number of paths in the diffusion path set, the remaining capacity of the grid, and the weight of the willingness to stay, and generate a comprehensive diversion priority. Among them, the remaining capacity of the grid is the difference between the maximum carrying capacity of the area defined by the venue database and the real-time passenger flow data.

[0137] The path screening unit is used to calculate the total impedance of the candidate diversion path through the path impedance matrix when the comprehensive diversion priority exceeds the preset response threshold, and generate the optimal diversion path solution.

[0138] The guidance instruction generation unit is used to generate dynamic guidance instructions based on the optimal diversion path plan. The dynamic guidance instructions are used to instruct the path indication information of the electronic guidance screen to be updated in real time.

[0139] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the aforementioned method for intelligently predicting the audience flow of sports events are implemented.

[0140] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0141] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0142] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A method for intelligently predicting the audience flow of sports events, characterized in that: The method comprises: Collect real-time crowd flow data in the venue through sensor equipment, and perform spatiotemporal grid processing on the real-time crowd flow data in combination with the event flow of the event process to generate a spatiotemporal grid dataset; Performing spatiotemporal feature fusion on the spatiotemporal grid dataset through a deep neural network model, calculating the mutation coefficient of each grid, and determining the grid whose mutation coefficient meets a preset threshold as the core area; Acquire a building database of the venue, and construct a path resistance matrix based on the building database; Constructing a conduction model based on the core area and the path resistance matrix, predicting the diffusion path of pedestrian flow pressure, and generating a diffusion path set; Based on the diffusion path set and the path resistance matrix, the predicted value of the flow of people in each area in the future time period is calculated by the residence intention weight, wherein the residence intention weight is used to quantify the audience's tendency to stay in the target area.

2. The method according to claim 1, characterized in that The method of collecting real-time crowd flow data of the venue through sensing equipment and performing spatiotemporal grid processing on the real-time crowd flow data in combination with the event stream of the event process to generate a spatiotemporal grid data set includes: Divide the venue into monitoring grids of preset sizes, collect real-time coordinate data through a laser radar, and generate an initial crowd distribution matrix, wherein the initial crowd distribution matrix is used to record the real-time crowd density of each grid; Analyzing the device movement trajectory through a WiFi probe, calculating the movement direction vector of each grid, and generating movement direction distribution data, wherein the movement direction distribution data is used to characterize the dynamic changes in the movement direction of the audience in each grid; Based on the initial crowd flow distribution matrix and the movement direction distribution data, the event flow of the competition process is accessed, the event type is weighted and coded, and the spatiotemporal grid dataset with competition event tags is generated.

3. The method according to claim 2, characterized in that The step of fusing spatiotemporal features of the spatiotemporal grid dataset using a deep neural network model, calculating a mutation coefficient of each grid, and screening the grids whose mutation coefficients meet a preset threshold to determine as core areas includes: For the crowd density data in the spatiotemporal grid dataset, time series feature extraction is performed through a long short-term memory network to generate a potential state vector; Based on the venue topology map of the building database, spatial feature aggregation is performed on the potential state vector and the movement direction distribution data through a graph neural network to generate a spatiotemporal fusion feature; The mutation coefficient of the spatiotemporal fusion feature is calculated using the following formula to obtain a mutation coefficient matrix: α=σ(w d ΔD+w v (V) Among them, α is the mutation coefficient, ΔD is the difference in the density of people flow in adjacent time windows, ΔV is the rate of change of the moving direction vector, and w d is the preset crowd density weight coefficient, w v is the preset speed weight coefficient, σ is the activation function; Threshold screening is performed on the mutation coefficient matrix to obtain the core area, wherein the core area is used to characterize the instantaneous aggregation hotspot formed due to the sudden change of audience behavior.

4. The method according to claim 2, characterized in that The acquiring of the building database of the venue and constructing a path resistance matrix based on the building database includes: Extracting the path length and channel width of the grid in the core area from the building database to generate basic resistance parameters; Calculating the commercial attraction coefficient of each grid based on the commercial facility distribution data in the building database to generate an attraction coefficient matrix; The path resistance matrix is constructed based on the basic resistance parameters and the attraction coefficient matrix using the following formula: Among them, R ij is the path impedance matrix, L ij is the path length, W ij is the channel width, β is the commercial attractiveness adjustment coefficient, S j is the commercial attractiveness coefficient.

5. The method according to claim 1, wherein The step of constructing a conduction model based on the core area and the path resistance matrix, predicting the pedestrian pressure diffusion path, and generating a diffusion path set includes: Taking the core area as the pressure source and based on the path resistance matrix, performing an initial simulation of the pressure conduction process through a fluid mechanics model to generate an initial pressure distribution map; According to the initial pressure distribution map, the pressure gradient between adjacent grids is calculated by a finite difference method to generate a set of candidate diffusion paths; For each path in the candidate diffusion path set, a total impedance value is calculated based on the path impedance matrix, and paths whose total impedance exceeds a preset threshold are eliminated to generate a preliminary diffusion path set; Physical conflict detection is performed on the preliminary diffusion path set to generate the diffusion path set, wherein the diffusion path set includes a main conduction path and at least one secondary conduction path.

6. The method according to claim 4, characterized in that Based on the diffusion path set and the path resistance matrix, the predicted value of the flow of people in each area in the future period is calculated by the residence intention weight, including: Extracting the commercial attractiveness coefficient of the terminal grid in the diffusion path set to generate a target area attractiveness parameter; Based on the weight of the event marker and the target area attraction parameter, a weighted sum calculation is performed on the preset commercial attraction ratio coefficient and the event weight ratio coefficient to generate the audience's stay intention weight; Calculating the mean of the path conduction pressure values according to the path resistance matrix to generate a reference conduction pressure value; The baseline conduction pressure value and the residence intention weight are dynamically modified to generate the predicted value of the pedestrian flow in each area in the future time period.

7. The method according to claim 6, characterized in that The method further comprises: Based on the predicted passenger flow value, a dynamic weight decision tree is constructed to perform multi-dimensional weight allocation on the number of paths in the diffusion path set, the remaining capacity of the grid, and the weight of the willingness to stay, to generate a comprehensive diversion priority, wherein the remaining capacity of the grid is the difference between the maximum carrying capacity of the area defined by the venue database and the real-time passenger flow data; When the comprehensive grooming priority exceeds a preset response threshold, the total impedance of the candidate grooming paths is calculated using the path impedance matrix to generate an optimal grooming path solution; According to the optimal evacuation path plan, a dynamic guidance instruction is generated, and the dynamic guidance instruction is used to instruct the path indication information of the electronic guidance screen to be updated in real time.

8. An intelligent prediction device for audience flow of sports events, characterized in that: The device comprises: A data processing module is used to collect real-time crowd flow data of the venue through sensor equipment, and perform spatiotemporal grid processing on the real-time crowd flow data in combination with the event stream of the event process to generate a spatiotemporal grid data set; A mutation coefficient calculation module is used to perform spatiotemporal feature fusion on the spatiotemporal grid dataset through a deep neural network model, calculate the mutation coefficient of each grid, and determine the grid whose mutation coefficient meets a preset threshold as the core area; A path impedance construction module, configured to obtain a building database of the venue and construct a path impedance matrix based on the building database; A diffusion path generation module is used to construct a conduction model based on the core area and the path resistance matrix, predict the diffusion path of pedestrian pressure, and generate a diffusion path set; The crowd flow prediction module is used to calculate the crowd flow prediction value of each area in the future time period based on the diffusion path set and the path resistance matrix through the residence intention weight, wherein the residence intention weight is used to quantify the audience's tendency to stay in the target area.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images