Big data-based intelligent campus safety management early warning method and system

Abstract

The present application belongs to the technical field of smart campus, and relates to a smart campus safety management early warning method and system based on big data. The method comprises: collecting registration information, real-time behavior data and state information of the environment of an external vehicle; constructing a dynamic space-time graph, and generating an edge attribute reflecting the relationship between the deployed nodes in the dynamic space-time graph; calculating a reasonable path set of the external vehicle, and analyzing trajectory deviation data of the external vehicle; performing abnormal scoring on the external vehicle; generating a corresponding intervention control strategy; performing causal influence analysis and generating a decision basis; training a local graph neural network model, and centrally aggregating and distributing updates on shared model parameters in the local graph neural network model training result. The present application introduces a dynamic behavior modeling strategy based on a heterogeneous graph neural network, realizes structured modeling and time sequence behavior evolution expression of multi-source data of a campus mobile target, enables the system to capture minor abnormalities in individual trajectory evolution, and improves abnormal behavior recognition sensitivity.

Description

Technical Field

[0001] This invention belongs to the field of smart campus technology, and more specifically, relates to a smart campus security management and early warning method and system based on big data. Background Technology

[0002] With the continuous advancement of smart campus construction, the complexity of various activities such as personnel flow and vehicle traffic on campus is high, and campus safety management faces multiple challenges: On the one hand, the frequency of external vehicles entering the campus is increasing, and the uncertainty of their driving trajectories, stopping areas, and behavioral patterns may cause traffic congestion, collisions, or safety hazards; on the other hand, the complex road network on campus and the many densely populated areas (such as teaching buildings, canteens, and playgrounds) and the coupling effect of environmental factors (such as construction areas, weather changes, and lighting conditions) with the behavior of traffic participants further exacerbate the difficulty of safety management.

[0003] Existing campus security management systems use video surveillance equipment to achieve visual monitoring of key areas, combine facial recognition technology to verify the identity of personnel, and rely on access control systems to record the entry and exit information of vehicles and personnel. Some systems introduce rule engines or simple machine learning models (such as decision trees and support vector machines) to judge abnormal behavior based on preset thresholds (such as speeding detection and intrusion into restricted areas detection) and trigger alarm prompts.

[0004] However, existing systems mostly rely on fixed rules or static features (such as single speed thresholds or fixed restricted areas) for anomaly detection, making it difficult to adapt to the dynamic changes in the campus environment. In campus traffic systems, there are complex spatial relationships and interactions between vehicles, roads, and people. For example, vehicle speed is constrained by pedestrian density and road construction affects traffic paths. Existing technologies lack structured modeling of these relationships. Some systems use models (such as complex neural network models) to output anomaly detection results, but do not clarify the logical basis for the results, making it difficult for managers to trace the causes of abnormal behavior and verify the effectiveness of intervention strategies. Traditional methods are weak at detecting minor anomalies (such as vehicles slowly deviating from reasonable paths and gradual speed changes), often triggering alarms only when risks accumulate to a significant level, thus missing the opportunity for intervention.

[0005] Therefore, there is an urgent need for a smart campus security management and early warning system that can integrate multi-source data, dynamically model entity relationships, be interpretable, and support cross-regional collaboration, so as to improve the accuracy of identifying abnormal behavior in complex scenarios and the efficiency of intervention, and ensure campus safety. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a smart campus security management and early warning method and system based on big data.

[0007] In a first aspect, the present invention provides a smart campus security management and early warning method based on big data, comprising:

[0008] Collect registration information, real-time behavior data, and environmental status information of external vehicles;

[0009] A dynamic spatiotemporal graph containing deployment nodes is constructed using the collected data, and edge attributes reflecting the relationships between deployment nodes are generated in the dynamic spatiotemporal graph; deployment nodes include vehicles, roads, and people;

[0010] The set of reasonable routes for external vehicles is calculated based on the dynamic spatiotemporal graph, and the trajectory deviation data of external vehicles is analyzed based on the spatial relationship between the actual driving trajectory of external vehicles and the set of reasonable routes.

[0011] Based on the trajectory deviation data and changes in the real-time behavior data of external vehicles, anomaly scores are assigned to external vehicles.

[0012] Based on the anomaly score and the current status information of the external vehicle, a corresponding intervention and control strategy is generated.

[0013] Conduct causal impact analysis on intervention and control strategies and generate decision-making basis;

[0014] The local graph neural network model is trained on multiple deployment nodes, and the shared model parameters in the training results of the local graph neural network model are centrally aggregated and distributed for updating.

[0015] Secondly, this invention provides a smart campus security management and early warning system based on big data, including an external vehicle information collection unit, a dynamic spatiotemporal map construction unit, a path modeling and deviation analysis unit, an abnormal behavior measurement unit, an intervention strategy decision-making unit, an analysis and generation unit, and a federated learning collaboration unit.

[0016] The external vehicle information collection unit is used to collect the registration information, real-time behavior data and environmental status information of external vehicles.

[0017] The dynamic spatiotemporal graph construction unit is used to construct a dynamic spatiotemporal graph containing deployment nodes using the collected data, and to generate edge attributes in the dynamic spatiotemporal graph that reflect the relationships between deployment nodes; deployment nodes include vehicles, roads, and people;

[0018] The path modeling and deviation analysis unit is used to calculate the set of reasonable paths for external vehicles based on a dynamic spatiotemporal map, and to analyze the trajectory deviation data of external vehicles based on the spatial relationship between the actual driving trajectory of external vehicles and the set of reasonable paths.

[0019] The abnormal behavior measurement unit is used to score the abnormality of external vehicles based on trajectory deviation data and changes in real-time behavior data of external vehicles.

[0020] The intervention strategy decision unit is used to generate corresponding intervention control strategies based on the anomaly score and the current status information of the external vehicle.

[0021] The analysis and generation unit is used to perform causal impact analysis on intervention and control strategies and generate decision-making basis.

[0022] The federated learning collaboration unit is used to train local graph neural network models on multiple deployment nodes, and to centrally aggregate and distribute the shared model parameters in the training results of local graph neural network models for updates.

[0023] Based on the above technical solution, the present invention can be further improved as follows.

[0024] Furthermore, registration information for external vehicles will be collected, including license plate number, vehicle type, entry time, entry location, and destination.

[0025] The collection of real-time behavioral data of external vehicles includes collecting their GPS trajectories, speeds, and accelerations within the campus.

[0026] The collection of environmental status information for external vehicles includes information on road conditions, pedestrian density, light intensity, and construction area distribution at the vehicle's location.

[0027] Furthermore, edge-to-edge attributes of a dynamic spatiotemporal graph are generated based on the spatial distance, speed difference, and risk value between deployed nodes.

[0028] Furthermore, the reasonable path set for external vehicles is calculated based on the dynamic spatiotemporal graph, including: calculating the shortest path between the entry node and the target node of the external vehicle in the dynamic spatiotemporal graph as the basic path; setting the tolerance parameter of the preset basic path; and constructing the reasonable path set according to the path length of the basic path and the preset tolerance parameter.

[0029] Furthermore, based on the spatial relationship between the actual driving trajectories of external vehicles and the set of reasonable paths, the trajectory deviation data of external vehicles are analyzed, including:

[0030] Calculate the path deviation of external vehicles by comparing their actual driving trajectory with the path in the selected road segment from the set of reasonable paths.

[0031] Based on the historical data of each external vehicle, a historical speed distribution model corresponding to each external vehicle is constructed; the current speed distribution of the external vehicle is compared with the historical speed distribution of the external vehicle to obtain the speed distribution difference of the external vehicle.

[0032] Based on the differences in the path deviation and speed distribution of the incoming vehicle, the trajectory deviation data of the incoming vehicle is obtained.

[0033] Furthermore, based on the path deviation and speed distribution differences of the incoming vehicle, trajectory deviation data of the incoming vehicle is obtained, including: calculating the anomaly score of the incoming vehicle based on the path deviation and speed distribution differences.

[0034] Let the anomaly score be The average spatial offset between the actual driving trajectory of external vehicles and the reasonable paths in the reasonable path set is , and The predefined non-negative real weighting coefficients are used, and the Kullback-Leibler divergence is... The current speed distribution of incoming vehicles is as follows: The historical speed distribution of incoming vehicles is as follows The anomaly score is then:

[0035] .

[0036] Furthermore, the location information of external vehicles, graph embedding representation, anomaly scoring, and environmental state information are used to construct a state vector;

[0037] Intervention and control strategies are generated based on state vectors using a reinforcement learning model.

[0038] Intervention and control strategies include voice prompts, route replanning, security clearances, and road restrictions.

[0039] Furthermore, a causal impact analysis of the intervention and control strategies is conducted to generate a basis for decision-making, including:

[0040] Construct a causal graph structure among intervention behavior variables, vehicle response behavior variables, and environmental variables;

[0041] The expected value of vehicle response behavior variables under intervention behavior variables is calculated based on the causal graph structure as the result of causal inference.

[0042] Generate intervention decision explanation information based on the results of causal reasoning as the basis for decision generation.

[0043] Furthermore, the local graph neural network model is trained on multiple deployment nodes, and the shared model parameters in the training results of the local graph neural network model are centrally aggregated, distributed, and updated, including:

[0044] A local model is obtained by training a graph neural network model based on local data on each deployment node.

[0045] Upload the shared model parameters from the local model to the central server;

[0046] Perform a weighted average calculation on the shared model parameters uploaded by multiple deployment nodes and generate globally shared model parameters;

[0047] Distribute the globally shared model parameters to each deployment node to update the local model.

[0048] The beneficial effects of this invention are:

[0049] (1) This invention introduces a dynamic behavior modeling strategy based on heterogeneous graph neural networks, which realizes structured modeling and temporal behavior evolution expression of multi-source data of mobile targets on campus. This enables the system to capture small anomalies in the evolution of individual trajectories, improves the sensitivity of abnormal behavior identification, and avoids the problem that traditional methods, which mostly rely on static rule matching or simple sequence statistics, are difficult to adapt to the diversity and evolution of behavior patterns in complex environments.

[0050] (2) By integrating causal reasoning calculation, this invention can quantitatively analyze the effectiveness of intervention measures and trace the causal logic behind the strategy trigger, achieving the dual goals of credible control explanation and reasonable strategy optimization. Compared with the existing solutions that generally use black box models to output control commands but lack reasoning transparency, this invention effectively solves the system trust problem.

[0051] (3) This invention constructs a federated learning collaboration mechanism without transmitting the original data, enabling different campus nodes to share graph neural network knowledge, ensuring the consistency of model updates and the collaborative learning capability of the system. This mechanism overcomes the bottlenecks of data leakage risk and insufficient cross-regional model generalization capability that are easily generated in the existing centralized training, and significantly enhances the adaptability of the system to cross-campus deployment. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the smart campus security management and early warning method based on big data provided in Embodiment 1 of the present invention;

[0053] Figure 2 This is a flowchart illustrating a specific implementation of Embodiment 1 of the present invention;

[0054] Figure 3 This is a schematic diagram of a smart campus security management and early warning system based on big data, provided in Embodiment 2 of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0056] Example 1

[0057] As an example, see the attached document. Figure 1 As shown, to solve the above-mentioned technical problems, this embodiment provides a smart campus security management and early warning method based on big data, including:

[0058] Collect registration information, real-time behavior data, and environmental status information of external vehicles;

[0059] A dynamic spatiotemporal graph containing deployment nodes is constructed using the collected data, and edge attributes reflecting the relationships between deployment nodes are generated in the dynamic spatiotemporal graph; deployment nodes include vehicles, roads, and people;

[0060] The set of reasonable routes for external vehicles is calculated based on the dynamic spatiotemporal graph, and the trajectory deviation data of external vehicles is analyzed based on the spatial relationship between the actual driving trajectory of external vehicles and the set of reasonable routes.

[0061] Based on the trajectory deviation data and changes in the real-time behavior data of external vehicles, anomaly scores are assigned to external vehicles.

[0062] Based on the anomaly score and the current status information of the external vehicle, a corresponding intervention and control strategy is generated.

[0063] Conduct causal impact analysis on intervention and control strategies and generate decision-making basis;

[0064] The local graph neural network model is trained on multiple deployment nodes, and the shared model parameters in the training results of the local graph neural network model are centrally aggregated and distributed for updating.

[0065] As attached Figure 2 The flowchart shown illustrates a specific implementation method for a smart campus security management and early warning system based on big data.

[0066] Optionally, registration information of external vehicles can be collected, including license plate number, vehicle type, entry time, entry location, and destination.

[0067] The collection of real-time behavioral data of external vehicles includes collecting their GPS trajectories, speeds, and accelerations within the campus.

[0068] The collection of environmental status information for external vehicles includes information on road conditions, pedestrian density, light intensity, and construction area distribution at the vehicle's location.

[0069] By collecting multi-dimensional and multi-modal data on external vehicles entering the campus area, and closely integrating this data with subsequent dynamic spatiotemporal graphs, the system achieves seamless data and logical flow integration. Typically, upon receiving a new external vehicle entry event, the system can initiate data collection for the external vehicle, including registration information, real-time behavioral data, and environmental status information.

[0070] As a typical application process, when an external vehicle enters the controlled area of ​​the campus, the system will initiate the registration and collection of the vehicle, trajectory collection, and real-time detection of the environmental status through the multi-source sensing devices deployed at the school gate. The data will be timestamped and stored in a structured cache queue as the basic data fragments for subsequent dynamic graph construction.

[0071] The registration information of external vehicles is integrated with the campus access control platform and bound to the visitor reservation system via high-definition license plate recognition cameras. The system can synchronously acquire information including but not limited to:

[0072] The license plate number of the visiting vehicle; the vehicle type, such as a sedan or a minivan; the actual timestamp of entry to the campus, in Unix timestamp format; the geographical location of entry to the campus, which can be represented by static location codes; and the target location to be visited, usually provided by the campus reservation system.

[0073] In one alternative implementation, the registration information can also record the vehicle's expected departure time from school and the contact person's information for subsequent trajectory deviation and abnormal stay analysis.

[0074] The system continuously tracks real-time vehicle behavior data within the campus. Typically, it deploys several Global Navigation Satellite System (GNSS) differential positioning base stations on campus, combining them with an inertial navigation module and an onboard GPS terminal to achieve high-precision sampling. In some embodiments, the trajectory point sampling frequency is set to 1Hz, and the collected data includes: location information, set to... Recorded in WGS-84 or projected coordinate system format; instantaneous velocity Acceleration is calculated from the velocity difference between two adjacent time points. Let... For a moment The vehicle acceleration, in m / s² 2 , The vehicle speed at time is , The vehicle speed at time is , Given a sampling time interval, typically 1 second, the vehicle acceleration is expressed as:

[0075] .

[0076] It is worth noting that, in order to suppress the influence of GPS noise, Kalman filters are often introduced in actual deployments to smooth the trajectory.

[0077] When collecting status information about the environment of external vehicles, scene perception and context modeling of the vehicle's surroundings are performed.

[0078] In one possible design approach, the environment state is obtained from the following data source:

[0079] Road slipperiness: Real-time rainfall and humidity are obtained through micro-meteorological monitoring equipment deployed on key road sections; Pedestrian density: Instantaneous number of people in hotspot areas is detected using facial recognition cameras or Wi-Fi probes; Illumination intensity: Detected using ambient light sensors, measured in Lux, and often used to determine nighttime driving behavior; Area construction information: Usually represented by a map-based code provided by the campus management system or building information modeling platform.

[0080] As an alternative, to improve the accuracy of scene representation, the system performs block-based mapping processing on the above environmental information, that is, divides the campus into multiple perception grids, and each grid maintains a set of environmental state vectors; let the environmental state vectors be... , The road surface slippage index. Population density, in people per square meter. Light level, measured in Lux. This is the code for construction signs; it is set to 1 if construction is present, and 0 otherwise. It also represents the road surface slippage index. Normalized to [0,1], the environment state vector is:

[0081] .

[0082] These structured environment vectors are mapped to road and crowd node attributes during the dynamic graph construction phase. In some embodiments, to reduce sensor load and communication pressure, the system adopts an edge computing node deployment strategy. Each collection point completes data preprocessing, compression, and encoding locally, uploading only key feature values ​​to the central node. For example, the trajectory point sequence is compressed into a five-dimensional state summary, including the start point, end point, maximum speed, average acceleration, and dwell time. Furthermore, in practical applications, to enhance adaptability to special types of vehicles (such as delivery trucks and construction vehicles), the system supports preliminary classification and labeling of the collected data using a rule base. For example, if the vehicle trajectory is clearly concentrated between the cafeteria and the teaching building, and the dwell time is short, it is initially labeled as "short-term delivery"; if the trajectory lingers around non-pre-arranged destinations, the system will label it with "high deviation potential" to trigger warning conditions.

[0083] Optionally, edge-to-edge attributes of a dynamic spatiotemporal graph can be generated based on the spatial distance, velocity difference, and risk value between deployed nodes.

[0084] Based on the collected information on incoming vehicles, a dynamic spatiotemporal graph containing vehicle, road, and pedestrian nodes is constructed, and edge attributes reflecting the relationships between nodes are generated. Specifically, the heterogeneous data of incoming vehicle information is transformed into a unified graph data structure and then input into the dynamic spatiotemporal graph for subsequent graph computation tasks such as path modeling and behavior analysis. The key task in the overall system architecture is to organize multimodal raw information into a dynamic spatiotemporal graph adapted to the graph neural network processing flow using core components of data structure abstraction and graph encoding.

[0085] This invention introduces a dynamic behavior modeling strategy based on heterogeneous graph neural networks, which realizes structured modeling and temporal behavior evolution expression of multi-source data of mobile targets on campus. This enables the system to capture subtle anomalies in the evolution of individual trajectories, improves the sensitivity of abnormal behavior identification, and avoids the problem that traditional methods, which mostly rely on static rule matching or simple sequence statistics, are difficult to adapt to the diversity and evolution of behavior patterns in complex environments.

[0086] Generally, after the data collection process for external vehicles entering the school is completed, the system immediately triggers graph structure encoding of the spatial, behavioral, and environmental data involved in the current vehicle behavior cycle. During the dynamic graph construction process, the graph nodes, edge attributes, and their temporal embeddings are dynamically updated to ensure that the constructed graph can effectively reflect the changes in the spatiotemporal environment and potential risk situations of the vehicle.

[0087] Dynamic spatiotemporal graph construction includes node construction, edge attribute generation, and graph embedding extraction, which are implemented as follows:

[0088] Node construction is used to model elements related to vehicle activity within the campus as nodes in a graph. Node types include at least three categories: external vehicle nodes, representing a single vehicle entering the campus, whose attributes may include vehicle identification, current trajectory coordinates, speed, and acceleration; campus road nodes, used to represent the spatial location, directionality, and traffic status of roads within the campus; and densely populated area nodes, representing specific areas within the campus, such as high-traffic hotspots like canteens, teaching buildings, and libraries.

[0089] In some embodiments, to enhance the expressive power of nodes, the system defines each node as a set of vehicle node attribute vectors, assuming... For vehicles at any time The horizontal coordinate information, For vehicles at any time The ordinate information is in meters, using the WGS-84 coordinate system or a projected coordinate system. The vehicle speed is expressed in meters per second (m / s). Vehicle acceleration is expressed in meters per second squared (m / s²). The sampling timestamp is in Unix timestamp format, and the vehicle node attribute vector is... , is represented as:

[0090] .

[0091] Alternatively, each node in the graph can be assigned a node category code to identify the semantic role the node plays in the graph, in order to support subsequent embedding learning.

[0092] Edge attributes define the connection relationships and attributes between different types of nodes. Edges are used to model the structural relationships and spatiotemporal interactions between nodes. The main edge types include: navigation edges between vehicle and road nodes, indicating whether a vehicle is traveling on a specific road; proximity edges between vehicle and crowd area nodes, indicating the spatial proximity between vehicle and crowd hotspots; and topological edges between road nodes, which form the basic framework of the transportation network.

[0093] Specifically, let For nodes and The spatial Euclidean distance between them, in meters. If a vehicle is connected to a road node, the speed difference is expressed in m / s. The risk weight value reflects the potential safety risk of the connection path. It can be obtained by comprehensively evaluating historical accident records, pedestrian density, and lighting conditions, and normalized to [0,1]. The attribute vector of each edge is... ,but Represented as:

[0094] .

[0095] In one possible implementation, to enhance the model's ability to model complex relationships, the system can also adopt a multi-dimensional edge attribute structure, where the risk weight values ​​are further refined into multiple dimensions, such as congestion index, slipperiness index, and illumination confidence.

[0096] Graph embedding extraction encodes the constructed dynamic graph using graph neural network methods, extracting a low-dimensional embedding representation for each node to support the input requirements of subsequent path modeling and behavior discrimination models. Generally, this module uses spectral graph convolution to encode the graph structure. Assume the current graph... ,set up For the first Layered graph embedding input, the adjacency matrix of the graph is unit array is , Add an identity matrix to the adjacency matrix of the graph; for The degree matrix, where the diagonal elements are the node degree values; For the first Layer trainable weight matrix, For the activation function, ReLU or ELU activation functions are commonly used. The formula for calculating single-layer graph convolution is as follows:

[0097] .

[0098] Alternatively, considering dynamic graph changes, this embodiment supports sliding updates of the graph based on time windows, i.e., at fixed time intervals. Reconstruct the current dynamic graph to adapt to the dynamic evolution of vehicle behavior.

[0099] In some embodiments, to enhance the temporal modeling capability of dynamic graphs, the system further introduces a temporal encoding mechanism, mapping the temporal attributes of nodes to temporal vectors and concatenating them as input into the node features. Let the sampling timestamp be... The time embedding vector is , Represents the dimension index in the vector. The total number of dimensions of the temporal embedding vector, and the sine and cosine position encoding method of the temporal embedding vector are represented as follows: The cosine position encoding of the time embedding vector is represented as follows: ,but:

[0100] ; .

[0101] The final node feature input is represented as the concatenation result of the original features and the temporal embedding, i.e.:

[0102] .

[0103] As an extended implementation, the system also supports a graph layering mechanism. Different types of nodes are constructed into a heterogeneous graph model, and different attention weights are applied to different types of nodes and edges through a graph attention mechanism to improve the model's ability to express multiple relationships.

[0104] Optionally, the reasonable path set for external vehicles is calculated based on the dynamic spatiotemporal graph, including: calculating the shortest path between the entry node and the target node of the external vehicle in the dynamic spatiotemporal graph as the basic path; setting the tolerance parameter of the preset basic path; and constructing the reasonable path set according to the path length of the basic path and the preset tolerance parameter.

[0105] Specifically, path modeling and deviation analysis follow the construction of the dynamic spatiotemporal graph. It inherits the dynamic graph structure containing road nodes, vehicle nodes, and crowd nodes generated by the latter. Its main function is to establish a reasonable path model for external vehicles from the school entrance to the target area, and to calculate the path deviation degree by combining its real-time trajectory information, so as to provide the spatial deviation dimension input basis for the abnormal behavior measurement module.

[0106] Generally, after completing the initial map modeling, the system will call this module to perform path modeling and real-time comparative analysis based on the entrance to the school and the target node accessed by the external vehicle in the registration information. It can be used for initial screening of behavioral trends as well as for determining the rationality of the trajectory.

[0107] In this embodiment, path modeling and deviation analysis includes at least path generation, reasonable path selection, and deviation analysis.

[0108] Path generation is primarily used to calculate the shortest path from the vehicle's entry point to its destination based on the road node topology in a dynamic graph. Specifically, the system first uses the entry location provided by the registration information collection unit in the external vehicle information collection module. Based on the target location, determine the corresponding start and end nodes in the graph structure.

[0109] The system employs either Dijkstra's algorithm or the A* heuristic search algorithm in the graph. The path with the minimum cost is calculated above, and this path is used as the basic path for the vehicle. Let... Let be the path length, and the minimum cost path be... , Nodes in the path With nodes Spatial Euclidean distance between them, path length It is obtained by summing the distance attributes of all edges on the path:

[0110] .

[0111] The reasonable path selection process constructs a set of reasonable paths based on the basic paths. Let the set of reasonable paths be denoted as . , Indicates the first Candidate paths; Representing a path The total length; This represents the maximum acceptable path length range in meters. The preset value can be dynamically set based on road density or traffic complexity. Generally, the system sets an adjustable path tolerance parameter to represent the maximum acceptable path gain. During the path selection phase, the system iterates through all feasible paths and retains those that meet the following constraints:

[0112] .

[0113] As an alternative, to avoid excessive path numbers affecting computational efficiency, path structure constraints can be introduced, such as restrictions on the number of node visits or the number of path loops, to further reduce the size of the path set.

[0114] Optionally, the trajectory deviation data of external vehicles can be analyzed based on the spatial relationship between the actual driving trajectory of the external vehicle and the set of reasonable paths, including:

[0115] Calculate the path deviation of external vehicles by comparing their actual driving trajectory with the path in the selected road segment from the set of reasonable paths.

[0116] Based on the historical data of each external vehicle, a historical speed distribution model corresponding to each external vehicle is constructed; the current speed distribution of the external vehicle is compared with the historical speed distribution of the external vehicle to obtain the speed distribution difference of the external vehicle.

[0117] Based on the differences in the path deviation and speed distribution of the incoming vehicle, the trajectory deviation data of the incoming vehicle is obtained.

[0118] Deviation analysis is used to compare the actual driving trajectories of external vehicles in real time. With reasonable path set Calculate the path deviation index in the path. .

[0119] In one possible implementation, the system discretizes the actual driving trajectory of the external vehicle into a sequence of trajectory points:

[0120] .

[0121] The system selects from the set of reasonable paths and The most similar path Matching can be performed using Hausdorff distance or weighted DTW (Dynamic Time Warping) algorithms.

[0122] set up This is an index of average spatial deviation, in meters. The actual number of trajectory points. For the actual trajectory of the first Coordinates of a point, Indicates a reasonable path The coordinates of each path point and the path deviation index are defined as follows:

[0123] .

[0124] The higher the path deviation index value, the more obvious the vehicle deviates from the expected path. This index is usually passed to the abnormal behavior measurement module to participate in the abnormality score calculation.

[0125] In some embodiments, to adapt to the uncertainty of multi-source location data, the system can set confidence intervals for trajectory points and introduce the deviation index calculation process into a probabilistic modeling mechanism. For example, each trajectory point can be regarded as the center of a two-dimensional Gaussian distribution, and the expected offset value can be calculated using the expected distance, thereby enhancing the robustness of the algorithm in location drift scenarios.

[0126] In some embodiments, to adapt to the uncertainty of multi-source location data, the system can set confidence intervals for trajectory points and introduce the deviation index calculation process into a probabilistic modeling mechanism. For example, each trajectory point can be regarded as the center of a two-dimensional Gaussian distribution, and the expected offset value can be calculated using the expected distance, thereby enhancing the robustness of the algorithm in location drift scenarios.

[0127] As an extended implementation method, the system can also classify deviation behaviors, such as "short-term path deviation", "long-term deviation and lingering" and "reverse passage" as behavioral labels, to assist the logic of subsequent intervention strategy selection.

[0128] Optionally, based on the difference between the path deviation and speed distribution of the external vehicle, the trajectory deviation data of the external vehicle can be obtained, including: calculating the anomaly score of the external vehicle based on the difference between the path deviation and speed distribution.

[0129] Let the anomaly score be The average spatial offset between the actual driving trajectory of external vehicles and the reasonable paths in the reasonable path set is , and The predefined non-negative real weighting coefficients are used, and the Kullback-Leibler divergence is... The current speed distribution of incoming vehicles is as follows: The historical speed distribution of incoming vehicles is as follows The anomaly score is then:

[0130] .

[0131] Abnormal behavior measurement is based on the path deviation index calculated by the path modeling and deviation analysis module and the changes in the behavior patterns of external vehicles, and an abnormal score is given to external vehicles. Specifically, in the system architecture of this invention, abnormal behavior measurement is located after path modeling and deviation analysis. It is mainly used to comprehensively measure the spatial behavior deviation and speed behavior changes of vehicles, and to build a quantifiable behavior anomaly scoring system, which serves as an important input state source for the subsequent intervention strategy decision module.

[0132] By adopting a speed behavior deviation analysis model based on Kullback-Leibler divergence, the actual operating status of vehicles can be accurately compared and evaluated with historical behavior baselines, thereby achieving more quantitative anomaly scoring calculations. Compared with traditional solutions that use a single threshold or manually set deviation rules, this solution is more scalable and has greater discrimination stability, making it particularly suitable for risk screening needs in high-density traffic scenarios.

[0133] Generally, after acquiring spatial trajectory deviation indicators of incoming vehicles, the system combines these indicators with historical and current speed pattern changes to determine whether their behavior exhibits an abnormal trend. This module combines structured path deviation data with time-series speed data to form a unified scoring model.

[0134] In this embodiment, the abnormal behavior measurement module mainly includes a behavior pattern modeling unit, a behavior change comparison unit, and an abnormal score generation unit.

[0135] The behavior pattern modeling unit is used to establish a reference model of the speed behavior of external vehicles.

[0136] Specifically, the system first extracts the vehicle's speed sequence data from its past several entries into the school. The speed values ​​in each trajectory are normalized, and statistical modeling is performed on all samples to form the vehicle's historical speed distribution. .

[0137] In one possible implementation, the system uses kernel density estimation or histogram estimation to model the probability density of the velocity data, outputting a continuous probability distribution function:

[0138] ;

[0139] in: For velocity variables, the unit is meters per second (m / s). This indicates that the speed value is Historical probability density value at time, This indicates the historical velocity distribution in velocity The probability density function value at a given location is typically obtained by histogram estimation or kernel density smoothing estimation of historical sampling data. This speed distribution is used as a baseline behavioral feature, reflecting the speed habits of vehicles during normal traffic.

[0140] The behavior change comparison process is used to calculate the difference between current behavior and historical behavior. During the vehicle's current entry into the school, the system records its latest speed sequence in real time and generates a current speed distribution model based on this. This distribution can use the same modeling method as the historical speed distribution to ensure comparability.

[0141] To measure the degree of variation between distributions, the system uses Kullback-Leibler divergence to calculate the difference, as shown in the following formula:

[0142] ;

[0143] in: This represents the KL divergence between the actual velocity distribution and the historical velocity distribution. The total number of discrete velocity intervals represents the number of segments in the velocity space. Actual velocity distribution In the Probability values ​​over a speed range Historical velocity distribution In the Probability values ​​over a speed range; It is a logarithmic function, usually with the natural logarithm as its base.

[0144] Generally, to ensure the stability of divergence calculation, the system smooths all interval probabilities, for example, by adding 1 or setting a minimum threshold to avoid division by zero. Anomaly score generation is used to jointly model path deviation indicators and speed behavior offset, forming a unified anomaly score.

[0145] In a standard implementation, the system introduces a linear weighted combination function as the scoring model. Let... This indicates an abnormal rating for outside vehicles. and Let the pre-defined non-negative real number weighting coefficients be:

[0146] .

[0147] Generally, the system determines [the outcome] based on past security incident samples through supervised learning or expert calibration. and The appropriate values ​​should be selected to ensure that the scoring results are sensitive to abnormal behavior. Alternatively, the system can incorporate a dynamic adjustment mechanism to adaptively adjust the weighting coefficients during peak periods or under complex weather conditions, giving the scoring system environmental adaptability.

[0148] In some embodiments, the system supports setting multi-level thresholds for abnormal scoring results. Based on different scoring ranges, labels such as "low risk," "medium risk," and "high risk" are generated and fed back to the intervention strategy decision-making module via confidence scores, forming a tiered response mechanism. Furthermore, to enhance the traceability and interpretability of the scoring, the system can record the original input data, model parameters, and intermediate calculation results used for each scoring, supporting subsequent calls to the causal inference module.

[0149] Optionally, the location information of the external vehicle, graph embedding representation, anomaly score, and environmental state information can be used to construct a state vector;

[0150] Intervention and control strategies are generated based on state vectors using a reinforcement learning model.

[0151] Intervention and control strategies include voice prompts, route replanning, security clearances, and road restrictions.

[0152] The intervention control strategy is trained by setting a reward function: Let... Indicates the intervention strategy at time. The reward value below; Indicates at time Safety ratings for vehicles from outside the area; Indicates at time The false alarm loss value; Indicates the delay time of the intervention response; , and Let be the non-negative real-valued parameters used for weighting in the reward function, then the reward function can be expressed as:

[0153] .

[0154] The intervention strategy decision-making process generates corresponding intervention control strategies based on the abnormal behavior measurement's anomaly score and the current state information of the incoming vehicle. Specifically, the intervention strategy decision-making process, as the control execution process in the system structure of this invention, follows the abnormal behavior measurement. Using the vehicle's abnormal score as the core triggering factor and combining it with its current state vector, it dynamically generates the optimal intervention response strategy with the spatiotemporal feature support provided by the graph neural network. This is a key logical node for decision generation, strategy evaluation, and intervention command output, and is crucial for achieving proactive safety control. Generally, the system initiates the strategy generation process of this module upon receiving an abnormal score from an incoming vehicle. Based on the state space encoding, the system calls a reinforcement learning model to select a strategy and sends the selected control action to the control system or security personnel in real time.

[0155] The intervention strategy decision-making process includes state vector construction, policy generation, and control output, as follows: The state vector construction process integrates the multi-source state information of the external vehicle at the current moment and transforms it into an input format acceptable to the reinforcement learning algorithm. Specifically, the state vector includes at least the following elements: the vehicle's current position information. The unit is meters; the graph embedding representation vector is... The result is derived from the node embedding results in the dynamic spatiotemporal graph construction module; the anomaly score is... The parameters are from the abnormal behavior measurement module; the set of environmental state parameters is... This includes factors such as population density, road surface slippage index, light levels, and construction status; the time tag is... , representing the timestamp of state sampling. Therefore, the state vector can be written as: .

[0156] As an alternative, to improve the contextual consistency of policy generation, the system can also introduce a state trajectory stack composed of state sequences from several past moments to achieve time series dependency modeling.

[0157] The policy generation process employs a reinforcement learning model to calculate the intervention policy decision. In a standard implementation, the system uses a Deep Q-Network (DQN) framework to construct the policy function, realizing the mapping between the state vector and the action space. The system defines a discrete action space. This includes: voice prompts, route replanning, security notifications, and road closure suggestions. To construct the reinforcement learning objective, the system defines the following reward function:

[0158] ;

[0159] in: The strategy is represented at time 10. The instant reward value below, This indicates the improvement in safety score due to the vehicle's behavior under the current policy intervention. This represents the false positive loss metric generated after the strategy is executed. This represents the time delay between policy execution and the generation of a response, in seconds. , and These are non-negative weighting coefficients used to adjust the contribution of the three types of indicators to the reward function. Generally, the system trains the reinforcement learning network weights by combining them with historical event sample replay data, and uses the experience replay mechanism to stabilize the training process.

[0160] In some embodiments, the policy generation network may adopt a dual-network structure, that is, the main Q network and the target Q network are used to update parameters alternately to avoid the problem of Q-value overestimation.

[0161] Based on the decision output generated by the strategy, the system executes intervention actions. According to the strategy results, the system calls different execution interface modules, such as voice broadcasting modules, in-vehicle navigation command modules, security alarm interfaces, or traffic management API interfaces, to complete the external action of the intervention strategy. As one possible deployment method, some intervention strategies can be implemented in real time at the vehicle entry control terminal (such as the gate system and intersection broadcast), while some operations involving management permissions need to be uploaded to the central platform for confirmation by security personnel before execution. In some embodiments, the system generates a unique control number for each intervention and records the issued action, execution time, response result, and strategy score for subsequent causal reasoning and tracing in the interpretability analysis module.

[0162] Optionally, causal impact analysis of intervention and control strategies can be conducted to generate decision-making basis, including:

[0163] Construct a causal graph structure among intervention behavior variables, vehicle response behavior variables, and environmental variables;

[0164] The expected value of vehicle response behavior variables under intervention behavior variables is calculated based on the causal graph structure as the result of causal inference.

[0165] Generate intervention decision explanation information based on the results of causal reasoning as the basis for decision generation.

[0166] The interpretability analysis module is used to perform causal impact reasoning on the control strategies output by the intervention strategy decision-making module and generate decision-making basis;

[0167] Specifically, the interpretability analysis module is located in the decision feedback stage of the system architecture of this invention, immediately following the intervention strategy decision module. The main function of this module is to perform causal analysis on the implemented or recommended intervention control strategies, clarify whether the intervention measures have truly led to improvements in vehicle behavior, and provide a verifiable and traceable logical basis for system control decisions through a structured reasoning process.

[0168] Generally, once an intervention strategy has been implemented by the system or submitted for implementation confirmation, the interpretability analysis module initiates the modeling task. By constructing a causal relationship structure diagram among intervention variables, vehicle response behavior variables, and environmental covariates, the system estimates and interprets the actual utility of the behavioral intervention.

[0169] In this embodiment, the interpretability analysis process mainly includes causal structure modeling, causal reasoning calculation, and decision interpretation generation, as detailed below.

[0170] The causal structure modeling unit is used to define and construct the causal graph structure of intervention behavior variables, vehicle response behavior variables, and key environmental factors.

[0171] Specifically, the system first categorizes the input variables into three types:

[0172] Intervention variables This refers to the specific control strategies executed or recommended by the system, such as "voice alerts," "route replanning," and "triggering security linkages"; vehicle response behavior variables. : Refers to the observable behavioral changes following vehicle intervention, often including the magnitude of the decrease in anomaly scores, the degree of mitigation of path deviation, and speed stability; environmental covariates Factors such as the vehicle's current road type, pedestrian density, time of day, and weather conditions influence the response but are not directly affected by the control strategy. Generally, the system trains a structured causal graph based on historical decision samples. The relationships between each variable node are modeled using a probabilistic graphical structure, marking directions and conditional dependencies. In one possible implementation, the system uses a structured learning algorithm (such as the Peter-Clarke algorithm or acyclic constraint algorithm) to determine the existence of edges between variables in a data-driven manner, thereby constructing a complete directed acyclic graph.

[0173] The causal inference calculation process is responsible for estimating the expected changes in vehicle response behavior variables under given intervention conditions, for example, by using... The computational principle models the intervention effect. Let: Indicating mandatory intervention Rear vehicle response behavior variables The conditional probability; Denotes a set of environmental covariates, with a range of 1. ; Indicates the current distribution of environment variables in the system. The probability of; Indicates that in a given and Under the premise of observation Conditional probability is calculated using the following formula:

[0174] .

[0175] In general, the system constructs a joint distribution model based on training samples. The system then performs integration or summation based on the above formula to output the expected effect of a control strategy in the current environment. In some embodiments, the device selects a strategy as follows: The system can also compare and numerical values, evaluation strategies Alternative strategies The relative advantages and disadvantages under the same environmental conditions support strategy selection and multi-strategy evaluation. The decision interpretation generation unit is used to format the above causal reasoning results and generate explanatory labels and auxiliary explanations for the intervention behavior. Specifically, the system organizes the following structured information into the decision report output: selected intervention strategy. Description and numbering; actual environmental conditions Sampled values, such as crowd density, population risk level, time labels, etc.; vehicle response behavior variables. The system provides the direction and estimated value of changes, such as a decrease in abnormal scores; and causal effect explanations, such as "In the current high-traffic environment, triggering voice prompts can reduce abnormal scores by approximately 18%." Alternatively, the system can archive each decision statement along with the original state vector, model parameters, and graph structure information for subsequent review or access by security personnel. In some embodiments, to facilitate visualization analysis, the system supports displaying the inference links between variables in the form of a causal path graph, such as "voice prompts," "driving behavior deceleration," "decrease in deviation," and "decrease in abnormal scores." Furthermore, it supports feeding back the causal graph results to the intervention strategy module for refining the reward function structure during reinforcement learning, constructing a causal-policy closed-loop feedback chain.

[0176] This invention, by integrating causal reasoning calculations, can quantitatively analyze the effectiveness of intervention measures and trace the causal logic behind the triggering of strategies, achieving the dual goals of credible control explanations and reasonable optimization of strategies. Compared with existing solutions that generally use black-box models to output control commands but lack reasoning transparency, this invention effectively solves the system trust problem.

[0177] Optionally, the local graph neural network model can be trained on multiple deployment nodes, and the shared model parameters in the training results of the local graph neural network model can be centrally aggregated and distributed for updating, including:

[0178] A local model is obtained by training a graph neural network model based on local data on each deployment node.

[0179] Upload the shared model parameters from the local model to the central server;

[0180] Perform a weighted average calculation on the shared model parameters uploaded by multiple deployment nodes and generate globally shared model parameters;

[0181] Distribute the globally shared model parameters to each deployment node to update the local model.

[0182] The federated learning collaborative process is used to train local graph neural network models on multiple deployment nodes and to centrally aggregate and distribute the shared model parameters in the training results of local graph neural network models for updates.

[0183] Specifically, a federated learning collaboration module is set up in the model training layer of the smart campus security management and early warning system. Its main function is to enable distributed graph neural network model training among different campus deployment nodes, thereby achieving knowledge sharing and model generalization without exchanging original data.

[0184] Normally, each campus or subsystem will independently run a local version of the system architecture, receiving local vehicle information and completing the anomaly identification and intervention process. This module, through periodic training tasks, enables systems in different locations to share global model parameters under secure conditions, improving the generalization ability and convergence efficiency of their respective models. In this embodiment, the federated learning collaboration process includes at least local model training, model parameter uploading, global model aggregation, and model update distribution. Specifically, the local model training unit is used to train the graph neural network model locally within a single campus node. Specifically, on each deployment node... Above, the system constructs local graph data. And train a graph neural network model with local feature extraction capabilities. The model's internal weights are divided into two parts: private parameters. Updated locally only, not shared; shared parameters Participate in the global synchronization process. During local training, the system uses local data to perform forward propagation and backward updates, with the optimization objective typically being anomaly rating prediction error or node classification loss.

[0185] The model parameter upload process is used to submit shared weights to the central coordination server; typically, the system does this at fixed time intervals. Initiate a synchronization request. After completing several rounds of local training, each node will share the current version's shared parameters. Upload to the server. During the upload process, the system will also send sample quantity information. This is used for subsequent calculations of aggregate weight allocation.

[0186] The global model aggregation process is deployed on a central server to perform weighted fusion of shared parameters from various nodes. In a standard implementation, a federated averaging algorithm is used to perform model aggregation. Let: This is the shared weight matrix after global aggregation. For the first Shared parameters uploaded by each node; For the first The number of training samples per node The total number of training samples for all nodes. If the total number of deployment nodes participating in this round of aggregation is:

[0187] .

[0188] As an alternative, the system can also set an aggregation weight adjustment mechanism based on factors such as node performance and data representativeness to dynamically adjust the contribution ratio of parameter fusion.

[0189] The model update and distribution process is responsible for distributing the aggregated shared model parameters to all participating nodes. Specifically, after completing the aggregation calculation, the system will... The data is simultaneously distributed to all campus subsystems, with each node replacing the old local shared parameters. This forms a new round of hybrid models. The local private parameters remain unchanged, ensuring that the models in each region are still adaptable to local scene characteristics. In some embodiments, to improve convergence speed and stability, the system can use a moving average method for parameter fusion. The fusion weighting coefficient has a value range of [0,1]. For the local shared parameters before the update, The updated local shared parameters are:

[0190] .

[0191] In some embodiments, the system supports multiple rounds of validation of the global aggregation results. By deploying a validation graph dataset at the central side, the system can evaluate the performance metrics of the current global model in typical scenarios, such as anomaly detection accuracy, recall, and scoring error, thereby assisting in the optimization of subsequent aggregation strategies. Furthermore, to protect the data privacy of each node, the system introduces parameter encryption mechanisms and differential privacy processing during federated communication to prevent data leakage through parameter inference.

[0192] Example 2

[0193] Based on the same principle as the method shown in Embodiment 1 of the present invention, as illustrated in the appendix. Figure 3 As shown, the embodiments of the present invention also provide a smart campus security management and early warning system based on big data, including an external vehicle information collection unit, a dynamic spatiotemporal map construction unit, a path modeling and deviation analysis unit, an abnormal behavior measurement unit, an intervention strategy decision-making unit, an analysis and generation unit, and a federated learning collaboration unit;

[0194] The external vehicle information collection unit is used to collect the registration information, real-time behavior data and environmental status information of external vehicles.

[0195] The dynamic spatiotemporal graph construction unit is used to construct a dynamic spatiotemporal graph containing deployment nodes using the collected data, and to generate edge attributes in the dynamic spatiotemporal graph that reflect the relationships between deployment nodes; deployment nodes include vehicles, roads, and people;

[0196] The path modeling and deviation analysis unit is used to calculate the set of reasonable paths for external vehicles based on a dynamic spatiotemporal map, and to analyze the trajectory deviation data of external vehicles based on the spatial relationship between the actual driving trajectory of external vehicles and the set of reasonable paths.

[0197] The abnormal behavior measurement unit is used to score the abnormality of external vehicles based on trajectory deviation data and changes in real-time behavior data of external vehicles.

[0198] The intervention strategy decision unit is used to generate corresponding intervention control strategies based on the anomaly score and the current status information of the external vehicle.

[0199] The analysis and generation unit is used to perform causal impact analysis on intervention and control strategies and generate decision-making basis.

[0200] The federated learning collaboration unit is used to train local graph neural network models on multiple deployment nodes, and to centrally aggregate and distribute the shared model parameters in the training results of local graph neural network models for updates.

[0201] Optionally, registration information of external vehicles can be collected, including license plate number, vehicle type, entry time, entry location, and destination.

[0202] The collection of real-time behavioral data of external vehicles includes collecting their GPS trajectories, speeds, and accelerations within the campus.

[0203] The collection of environmental status information for external vehicles includes information on road conditions, pedestrian density, light intensity, and construction area distribution at the vehicle's location.

[0204] Optionally, edge-to-edge attributes of a dynamic spatiotemporal graph can be generated based on the spatial distance, velocity difference, and risk value between deployed nodes.

[0205] Optionally, the reasonable path set for external vehicles is calculated based on the dynamic spatiotemporal graph, including: calculating the shortest path between the entry node and the target node of the external vehicle in the dynamic spatiotemporal graph as the basic path; setting the tolerance parameter of the preset basic path; and constructing the reasonable path set according to the path length of the basic path and the preset tolerance parameter.

[0206] Optionally, the trajectory deviation data of external vehicles can be analyzed based on the spatial relationship between the actual driving trajectory of the external vehicle and the set of reasonable paths, including:

[0207] Calculate the path deviation of external vehicles by comparing their actual driving trajectory with the path in the selected road segment from the set of reasonable paths.

[0208] Based on the historical data of each external vehicle, a historical speed distribution model corresponding to each external vehicle is constructed; the current speed distribution of the external vehicle is compared with the historical speed distribution of the external vehicle to obtain the speed distribution difference of the external vehicle.

[0209] Based on the differences in the path deviation and speed distribution of the incoming vehicle, the trajectory deviation data of the incoming vehicle is obtained.

[0210] Optionally, based on the difference between the path deviation and speed distribution of the external vehicle, the trajectory deviation data of the external vehicle can be obtained, including: calculating the anomaly score of the external vehicle based on the difference between the path deviation and speed distribution.

[0211] Let the anomaly score be The average spatial offset between the actual driving trajectory of external vehicles and the reasonable paths in the reasonable path set is , and The predefined non-negative real weighting coefficients are used, and the Kullback-Leibler divergence is... The current speed distribution of incoming vehicles is as follows: The historical speed distribution of incoming vehicles is as follows The anomaly score is then:

[0212] .

[0213] Optionally, the location information of the external vehicle, graph embedding representation, anomaly score, and environmental state information can be used to construct a state vector;

[0214] Intervention and control strategies are generated based on state vectors using a reinforcement learning model.

[0215] Intervention and control strategies include voice prompts, route replanning, security clearances, and road restrictions.

[0216] Optionally, causal impact analysis of intervention and control strategies can be conducted to generate decision-making basis, including:

[0217] Construct a causal graph structure among intervention behavior variables, vehicle response behavior variables, and environmental variables;

[0218] The expected value of vehicle response behavior variables under intervention behavior variables is calculated based on the causal graph structure as the result of causal inference.

[0219] Generate information explaining the intervention decision based on the results of causal reasoning as the basis for decision generation.

[0220] Optionally, the local graph neural network model can be trained on multiple deployment nodes, and the shared model parameters in the training results of the local graph neural network model can be centrally aggregated and distributed for updating, including:

[0221] A local model is obtained by training a graph neural network model based on local data on each deployment node.

[0222] Upload the shared model parameters from the local model to the central server;

[0223] Perform a weighted average calculation on the shared model parameters uploaded by multiple deployment nodes and generate globally shared model parameters;

[0224] Distribute the globally shared model parameters to each deployment node to update the local model.

[0225] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart campus security management and early warning method based on big data, characterized in that, include: Collect registration information, real-time behavior data, and environmental status information of external vehicles; The collected data is used to construct a dynamic spatiotemporal graph containing deployment nodes, and edge attributes reflecting the relationships between deployment nodes are generated in the dynamic spatiotemporal graph. Deployment nodes include vehicles, roads, and people; The set of reasonable routes for external vehicles is calculated based on the dynamic spatiotemporal graph, and the trajectory deviation data of external vehicles is analyzed based on the spatial relationship between the actual driving trajectory of external vehicles and the set of reasonable routes. Anomaly scores are assigned to external vehicles based on trajectory deviation data and real-time behavior data changes. Anomaly scores are calculated based on path deviation and speed distribution differences, denoted as follows: The average spatial offset between the actual driving trajectory of external vehicles and the reasonable paths in the reasonable path set is , and The predefined non-negative real weighting coefficients are used, and the Kullback-Leibler divergence is... The current speed distribution of incoming vehicles is as follows: The historical speed distribution of incoming vehicles is as follows The anomaly score is then: ； Based on the anomaly score and the current status information of the external vehicle, a corresponding intervention and control strategy is generated. Conduct causal impact analysis on intervention and control strategies and generate decision-making basis; In multiple deployment nodes, the dynamic spatiotemporal graph is used as input, and the graph embedding representation vector of each node is used as output to train the local graph neural network model. The shared model parameters in the training results of the local graph neural network model are then centrally aggregated and distributed for updating.

2. The smart campus security management and early warning method based on big data according to claim 1, characterized in that, Collect registration information of external vehicles, including license plate number, vehicle type, entry time, entry location and destination. The collection of real-time behavioral data of external vehicles includes collecting their GPS trajectories, speeds, and accelerations within the campus. The collection of environmental status information for external vehicles includes information on road conditions, pedestrian density, light intensity, and construction area distribution at the vehicle's location.

3. The smart campus security management and early warning method based on big data according to claim 1, characterized in that, The edges and edge attributes of a dynamic spatiotemporal graph are generated based on the spatial distance, speed difference, and risk value between deployed nodes.

4. The smart campus security management and early warning method based on big data according to claim 1, characterized in that, The calculation of a reasonable set of routes for external vehicles based on a dynamic spatiotemporal graph includes: calculating the shortest path from the entrance node to the destination node of the external vehicle in the dynamic spatiotemporal graph as the basic path; setting a tolerance parameter for the preset basic path; and constructing a reasonable set of routes based on the path length of the basic path and the preset tolerance parameter.

5. The smart campus security management and early warning method based on big data according to claim 1, characterized in that, Based on the spatial relationship analysis between the actual driving trajectory of external vehicles and the set of reasonable paths, the trajectory deviation data of external vehicles is analyzed, including: Calculate the path deviation of external vehicles by comparing their actual driving trajectory with the path in the selected road segment from the set of reasonable paths. Based on the historical data of each external vehicle, a historical speed distribution model corresponding to each external vehicle is constructed; the current speed distribution of the external vehicle is compared with the historical speed distribution of the external vehicle to obtain the speed distribution difference of the external vehicle. Based on the differences in the path deviation and speed distribution of the incoming vehicle, the trajectory deviation data of the incoming vehicle is obtained.

6. The smart campus security management and early warning method based on big data according to claim 1, characterized in that, The location information of external vehicles, graph embedding representation, anomaly scoring and environmental state information are used to construct a state vector; Intervention and control strategies are generated based on state vectors using a reinforcement learning model. Intervention and control strategies include voice prompts, route replanning, security clearances, and road restrictions.

7. The smart campus security management and early warning method based on big data according to claim 1, characterized in that, Conduct causal impact analysis on intervention and control strategies and generate decision-making basis, including: Construct a causal graph structure among intervention behavior variables, vehicle response behavior variables, and environmental variables; The expected value of vehicle response behavior variables under intervention behavior variables is calculated based on the causal graph structure as the result of causal inference. Generate intervention decision explanation information based on the results of causal reasoning as the basis for decision generation.

8. The smart campus security management and early warning method based on big data according to claim 1, characterized in that, The local graph neural network model is trained on multiple deployment nodes, and the shared model parameters in the training results are centrally aggregated, distributed, and updated, including: A local model is obtained by training a graph neural network model based on local data on each deployment node. Upload the shared model parameters from the local model to the central server; Perform a weighted average calculation on the shared model parameters uploaded by multiple deployment nodes and generate globally shared model parameters; Distribute the globally shared model parameters to each deployment node to update the local model.

9. A smart campus security management and early warning system based on big data, characterized in that: It includes a unit for collecting information on external vehicles, a dynamic spatiotemporal map construction unit, a path modeling and deviation analysis unit, an abnormal behavior measurement unit, an intervention strategy decision-making unit, an analysis and generation unit, and a federated learning collaboration unit. The external vehicle information collection unit is used to collect the registration information, real-time behavior data and environmental status information of external vehicles. The dynamic spatiotemporal graph construction unit is used to construct a dynamic spatiotemporal graph containing deployment nodes using the collected data, and to generate edge attributes in the dynamic spatiotemporal graph that reflect the relationships between deployment nodes; deployment nodes include vehicles, roads, and people; The path modeling and deviation analysis unit is used to calculate the set of reasonable paths for external vehicles based on a dynamic spatiotemporal map, and to analyze the trajectory deviation data of external vehicles based on the spatial relationship between the actual driving trajectory of external vehicles and the set of reasonable paths. The abnormal behavior measurement unit is used to score the abnormality of external vehicles based on changes in trajectory deviation data and real-time behavior data. It calculates the abnormality score of external vehicles based on path deviation and speed distribution differences, assuming the abnormality score is... The average spatial offset between the actual driving trajectory of external vehicles and the reasonable paths in the reasonable path set is , and The predefined non-negative real weighting coefficients are used, and the Kullback-Leibler divergence is... The current speed distribution of incoming vehicles is as follows: The historical speed distribution of incoming vehicles is as follows The anomaly score is then: ； The intervention strategy decision unit is used to generate corresponding intervention control strategies based on the anomaly score and the current status information of the external vehicle. The analysis and generation unit is used to perform causal impact analysis on intervention and control strategies and generate decision-making basis. The federated learning collaborative unit is used to train local graph neural network models in multiple deployment nodes by taking a dynamic spatiotemporal graph as input and the graph embedding representation vector of each node as output, and to centrally aggregate and distribute the shared model parameters in the training results of local graph neural network models.

Images