Event early warning efficient management method and system based on park management

Abstract

The invention discloses an event early warning efficient management method and system based on park management, and relates to the technical field of safety management, and the method comprises the steps: constructing a hexagonal cellular unit layout, generating an equipment registry and a network topology structure, and forming a structured data set through data alignment and sliding window verification; the edge server loads an initial weight, calculates a dynamic weight matrix based on the LSTM network, and generates a high-order feature vector set in combination with the graph neural network; inputting the high-order feature vector set into a space-time prediction model to generate an environment evolution trend prediction value, fusing network topology and BIM model parameters to calculate a dynamic risk index, and triggering graded early warning; the command center analyzes the early warning signal to generate a fusion visual interface, historical cases are matched through federal learning, a resource allocation scheme is optimized, and an optimal disposal strategy is output through Monte Carlo simulation. By dynamically adjusting the weight of the sensor, the weight of the infrared sensor in the high-temperature-sensitive area is increased, and the abnormality capturing efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of security management, and in particular to an efficient event early warning management method and system based on park management. Background Art

[0002] In recent years, campus security management technology has gradually evolved towards multimodal sensor fusion and edge intelligence. Existing solutions generally deploy heterogeneous sensors (temperature and humidity, infrared thermal imaging, video surveillance, and gas concentration sensors), combining low-power wide-area networks (such as LoRaWAN) with cellular communication protocols for data transmission. Basic data aggregation is performed through edge computing nodes. For anomaly detection, methods based on threshold triggering and static weight assignment have become mainstream. Some research has introduced Kalman filtering for time series alignment and improved data consistency. Furthermore, risk prediction models often rely on linear regression analysis of single-dimensional indicators (such as temperature or gas concentration) and combine historical data to construct risk indices.

[0003] However, the sensor weight allocation strategy is rigid and does not integrate real-time network status and historical contribution parameters, making it difficult to balance anomaly detection sensitivity and resource efficiency. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an efficient management method for event warning based on park management to solve the imbalance problem between anomaly detection sensitivity and resource efficiency caused by inaccurate dynamic weight allocation.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides an efficient event warning management method based on campus management, which includes deploying multiple types of sensors, constructing a hexagonal honeycomb unit layout, generating a device registry and network topology, and forming a structured data set through data alignment and sliding window verification; The edge server loads the initial weights and calculates the dynamic weight matrix based on the LSTM network, and generates a high-order feature vector set in combination with the graph neural network; The high-order feature vector set is input into the spatiotemporal prediction model to generate the predicted value of environmental evolution trend, and the network topology and BIM model parameters are integrated to calculate the dynamic risk index and trigger graded warnings; The command center analyzes warning signals to generate a fusion visualization interface, matches historical cases through federated learning, optimizes resource allocation plans, and outputs the optimal response strategy through Monte Carlo simulation. Real-time data collection and processing generates dynamic adjustment instructions, updates network parameters, and reconstructs the knowledge graph through federated learning to complete blockchain evidence storage.

[0007] As a preferred solution of the method for efficient event warning management based on park management of the present invention, the deployment of multiple types of sensors includes the following steps: Deploy wireless temperature and humidity sensors at grid coordinate points, infrared thermal imaging sensors at building entrances and exits, video surveillance sensors on top of streetlight poles, and gas concentration sensors in underground pipe corridors and distribution rooms. All sensors have built-in Beidou positioning chips to generate device fingerprint data packets; A hexagonal cellular unit layout is constructed based on the geographical location and a self-organizing network configuration table is generated.

[0008] As a preferred solution of the efficient management method for event warning based on park management of the present invention, the forming of the structured data set includes the following steps: Encapsulate the data collected by the sensor into raw data packets and transmit them to the edge computing node through dual channels; The Kalman filter algorithm is used to align the time axis of asynchronous data streams within the same grid, construct a multidimensional tensor data stream, and verify the data integrity to form a structured data set.

[0009] As a preferred solution of the event warning efficient management method based on park management of the present invention, wherein: the generating of the high-order feature vector set includes the following steps: The LSTM network processes the historical data contribution sequence and outputs a contribution coefficient matrix. This matrix is ​​weighted and summed based on the real-time network bandwidth and the remaining power of the device to generate a dynamically adjusted weight update vector. Performing Hadamard product operation on the dynamically adjusted weight update vector and the initial weight, combined with network topology normalization and weight upper limit constraint, generates an enhanced weight matrix; Multimodal graph neural networks are used to aggregate node features and edge features to generate a set of high-order feature vectors.

[0010] As a preferred solution of the efficient management method of event warning based on park management of the present invention, wherein: the triggering of graded warning includes the following steps: The predicted value of environmental evolution trend is combined with the grid vulnerability coefficient and the time decay factor is superimposed to generate a dynamic risk index; According to the continuous period of the risk index exceeding the critical value, the third-level warning is triggered and a warning message is generated; The early warning message is attached with Beidou coordinates and risk index change curve, and synchronized to mobile terminals and service endpoints through the MQTT protocol.

[0011] As a preferred solution of the event warning efficient management method based on park management of the present invention, the output of the optimal disposal strategy includes the following steps: Analyze BIM model coordinates and render real-time data into a 3D scene to generate a fusion visualization interface; Match historical case feature vectors through federated learning, filter similar cases, and calculate the shortest evacuation path; Construct a resource matrix and a time-path weight mapping table, and use the simplex method to solve the solution that maximizes resource coverage; The loss expectation is evaluated through Monte Carlo simulation and the optimal disposal strategy is output.

[0012] As a preferred solution of the event warning efficient management method based on park management of the present invention, wherein: the reconstruction of the knowledge graph includes the following steps: Parse the operation records and path offsets in the disposal data, generate dynamic adjustment instructions and update the digital twin engine; Desensitize and encrypt incremental data sets, optimize spatiotemporal prediction model parameters, and write them back to edge nodes; Extract feature importance reports to construct historical event feature matrices, and generate knowledge graphs with spatiotemporal attributes through graph neural networks; Write the hash value of the knowledge graph with time and space attributes into the blockchain distributed ledger to complete traceability and evidence storage.

[0013] In a second aspect, the present invention provides an efficient event warning management system based on campus management, including a data perception module, deploying multiple types of sensors, and constructing a hexagonal honeycomb unit layout, generating a device registry and network topology, and forming a structured data set through data alignment and sliding window verification; Dynamic calculation module: The edge server loads the initial weights and calculates the dynamic weight matrix based on the LSTM network, and generates a high-order feature vector set in combination with the graph neural network; The risk warning module inputs high-order feature vector sets into the spatiotemporal prediction model to generate environmental evolution trend prediction values, integrates network topology and BIM model parameters to calculate dynamic risk indexes and trigger graded warnings; In the strategy optimization module, the command center analyzes warning signals to generate a fusion visualization interface, matches historical cases through federated learning, optimizes resource allocation plans, and outputs the optimal response strategy through Monte Carlo simulation. The dynamic update module collects data in real time during the processing process to generate dynamic adjustment instructions, update network parameters, and reconstruct the knowledge graph through federated learning to complete blockchain evidence storage.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the efficient event warning management method based on park management as described in the first aspect of the present invention.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the efficient event warning management method based on park management as described in the first aspect of the present invention.

[0016] The beneficial effects of the present invention are as follows: based on the contribution analysis of the LSTM network, the real-time network bandwidth and device power parameters are integrated, and the sensor weight is dynamically adjusted, which solves the resource waste and sensitivity imbalance problems caused by static allocation, increases the weight of infrared sensors in high thermal-sensitive areas, and improves the efficiency of abnormal capture; according to the characteristics of heterogeneous data streams, a differentiated timing correction strategy is designed to reduce the time axis alignment error of multi-dimensional data to the millisecond level, ensuring the spatiotemporal consistency of the fused data set; through the spatiotemporal prediction model coupling the grid vulnerability coefficient, a risk index calculation engine is constructed to realize the joint deduction of environmental evolution trends and network topology, thereby improving the accuracy of early warning; the knowledge graph update mechanism driven by the federated learning framework, combined with blockchain evidence storage and Monte Carlo simulation evaluation, optimizes the dynamic adaptability of emergency plans, improves the efficiency of resource scheduling path planning, and shortens the collaborative response time. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 The flowchart of the efficient management method of event warning based on park management.

[0019] Figure 2 Schematic diagram for dynamic weight calculation and feature generation.

[0020] Figure 3 Schematic diagram of risk warning triggering and graded processing.

[0021] Figure 4 Generate schematics for federated learning optimization and strategies. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0025] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides an efficient event warning management method based on park management, comprising the following steps: S1. Deploy wireless temperature and humidity sensors at grid coordinate points within the campus' physical space. Use expansion bolts to secure a waterproof housing sensor to each coordinate point. Additionally, deploy infrared thermal imaging sensors at building entrances and exits and at corridor intersections. Video surveillance sensors, equipped with 360-degree pan / tilt heads, are deployed atop streetlight poles. Gas concentration sensors are deployed in underground utility corridors and power distribution rooms. All sensors have built-in Beidou positioning chips. Upon power-up, they automatically acquire longitude and latitude coordinates and write them to the non-volatile storage area of ​​the sensor firmware, generating a device fingerprint data packet containing the sensor ID, geographic coordinates, and deployment time.

[0026] Start the long-range low-power communication protocol stack of the deployed sensor nodes, configure it in the low-frequency working band, and set the physical layer parameters. Simultaneously activate the high-speed wireless communication protocol stack, configure it in the high-frequency band and define the waveform parameters. Construct a hexagonal cellular unit layout based on the geographical location of all sensors. With the central node of each cellular unit as the center, calculate all sensor nodes within the coverage area and generate a self-organizing network configuration table containing the central node ID, neighbor node list, and channel allocation plan. The central node selection criteria are nodes with received signal strength and energy status conditions.

[0027] To further illustrate, physical layer parameters include setting the spreading factor, bandwidth, and coding rate.

[0028] After all sensor nodes complete the network protocol configuration, they send the device fingerprint data packet to the edge computing node via the LoRaWAN protocol. The data packet format is a JSON structure containing the sensor ID, latitude, longitude, device type, and deployment time. After the edge computing node verifies the integrity of the data packet, it establishes a device registry containing the sensor ID, geographic coordinates, device type, and network role, and stores it as a SQLite database file. Based on the geographic coordinate data in the device registry, the Delaunay triangulation algorithm is used to generate a network topology diagram, and the output is a topology description file containing the location coordinates, connection relationships, and signal strength parameters of all sensor nodes.

[0029] The dynamic weight algorithm is called to parse the device type field in the topology description file, and differentiated initial weights are assigned to different types of sensors according to environmental monitoring requirements: wireless temperature and humidity sensors are assigned basic monitoring weights, infrared thermal imaging sensors are assigned high thermal sensitivity monitoring weights, video surveillance sensors are assigned dynamic tracking weights, and gas concentration sensors are assigned safety threshold monitoring weights. A configuration list of device identifiers and weight values ​​is formed and sent to each sensor node in the Extensible Markup Language format.

[0030] To further explain, the safety threshold refers to the concentration critical value parameter established by the gas safety standard of the target monitoring area, which can be obtained as follows: Obtain the gas safety industry standard documents corresponding to the target area, extract the harmful gas concentration limit data specified therein, combine the concentration fluctuation characteristics in the historical monitoring data, and determine the safe boundary value of the concentration change through probabilistic statistical analysis; perform weighted fusion of the industry standard limit and the safety boundary value to form a dynamic safety threshold that adapts to the specific application scenario.

[0031] After completing the initial weight configuration, the sensor network starts a fixed-period data collection mechanism to collect data, as follows: The wireless temperature and humidity sensor outputs double-precision floating-point numbers containing Celsius temperature and percentage humidity values. The infrared thermal imaging sensor generates a thermal radiation intensity matrix. The video surveillance sensor captures video streams. The gas concentration sensor measures a three-channel floating-point array containing methane, carbon monoxide, and oxygen concentrations.

[0032] Double-precision floating-point numbers, thermal radiation intensity matrices, video streams, and three-channel floating-point arrays are encapsulated into raw data packets and transmitted to edge computing nodes in parallel via the LoRaWAN protocol and 5G NR-U protocol dual channels.

[0033] After receiving the original data packet, the edge computing node extracts the geographic grid coordinates of the sensor from the device registry. It then uses the Kalman filter algorithm to perform time axis alignment on the asynchronous data streams from different sensors within the same grid. The specific processing steps are as follows: Five iterations of prediction and correction are performed on the periodic data of the wireless temperature and humidity sensor, two iterations are performed on the periodic data of the infrared thermal imaging sensor, linear interpolation is performed on the periodic data of the video surveillance sensor, and cubic spline interpolation is performed on the periodic data of the gas concentration sensor. Finally, all the processed data are time-aligned.

[0034] The aligned dataset is divided into data blocks of equal length according to the time window to construct a multi-dimensional tensor data stream.

[0035] To further illustrate, each data block contains temperature value, humidity value, thermal radiation matrix mean, video key frame feature vector and gas concentration maximum value.

[0036] The sliding window mechanism is started to perform integrity check on the multidimensional tensor data stream within the continuous time window, and the window slides forward with a fixed step size.

[0037] Among them, the data missing rate of each sensor channel in the window is counted each time it slides. When the number of missing data points of the wireless temperature and humidity sensor exceeds the number of consecutive missing times allowed in a single sampling cycle, the number of complete data frames in the infrared thermal imaging sensor window is lower than the effective coverage rate of the sensor's nominal frame rate, the missing data of the video surveillance sensor exceeds the fault tolerance threshold at a specific frame rate, and the number of effective sampling points of the gas concentration sensor is lower than the data integrity baseline specified by the sampling interval, the failure judgment condition is triggered, and the fluctuation range of the numerical values ​​of each channel is calculated at the same time. The historical data of each sensor is retrieved from the device registry to calculate the benchmark standard deviation. When the data fluctuation range in the current window exceeds three times the standard deviation and is marked as abnormal, the abnormally marked invalid data segment is removed from the multidimensional tensor data stream, and a special identification code is filled in the original data position to form a structured data set.

[0038] To further explain, the setting of the fault tolerance threshold is based on the security level of the deployment area and the distribution of missing key frames of the video under historical normal working conditions. For example, the core area security level A requires a fault tolerance threshold of less than 5%, and the ordinary area level B threshold is less than 10%.

[0039] Structured datasets are encapsulated into standard data frames, with the temperature and humidity channels classified as high-priority traffic, the thermal radiation intensity and gas concentration as medium-priority traffic, and the video keyframe feature vectors as basic-priority traffic. Standard header information, including the virtual local area network (VLAN) ID, priority code point, and timestamp fields, is added to each data frame. A time-aware shaper generates a time-gating list based on the network configuration, allocating transmission time proportions to different priority levels within periodic time intervals. Encapsulated data frames are forwarded through a multi-priority queue transmission mechanism, ensuring that data transmission meets end-to-end latency requirements and jitter control limits, ultimately reaching edge data aggregation nodes.

[0040] S2. The structured dataset output by the edge data aggregation node is transmitted to the edge server via the 5G NR-U protocol. The initial weights in the XML configuration file are loaded, and the sensor ID and initial weights are mapped as key-value pairs and stored in the memory cache. The dynamic weight adjustment model based on the attention mechanism reads the initial weights from the memory cache and accesses the device registry to obtain the historical data contribution records of each sensor node.

[0041] The contribution records of historical data are processed using the time series method of the LSTM network. The input format is a triple sequence of [timestamp, sensor ID, contribution value], and each time step corresponds to a fixed time interval. The LSTM network has 128 hidden layer units and uses the tanh activation function. The gating mechanism calculates the effective contribution of each sensor to abnormal event detection in the past several cycles and outputs a contribution coefficient matrix. The contribution coefficient matrix is ​​weighted and summed with the real-time network bandwidth value and the remaining battery percentage of the device extracted from the device registry. The network bandwidth weight factor is set to 0.4, the remaining battery weight factor is set to 0.3, and the contribution coefficient weight factor is set to 0.3 to form a dynamically adjusted weight update vector.

[0042] At each cycle, the dynamic weight adjustment algorithm traverses all sensor nodes, performs a Hadamard product operation on the adjusted weight update vector and the initial weight to generate a temporary weight matrix; the temporary weight matrix is ​​combined with the connection relationship in the topology description file, and normalization is performed on the central node in each hexagonal honeycomb unit; the normalized temporary weight matrix is ​​cross-validated with the device type field in the device registry to form an updated sensor weight matrix.

[0043] It should be noted that the weighted summation of the contribution coefficient matrix, the real-time network bandwidth value extracted from the device registry, and the percentage of remaining power of the device refers to the construction of a dynamic weight system from the three dimensions of data value (contribution), transmission reliability (bandwidth), and device reliability (power). The contribution coefficient ensures that high-value data (such as anomaly capture) is processed first, the real-time bandwidth avoids network congestion and the loss of key data (such as video stream interruption), and the remaining power ensures the continuous operation of key nodes (such as gas sensors in the distribution room). After the weighted fusion of the three, resource allocation can be adaptively adjusted. For example, when the bandwidth is insufficient, the weight of low-priority sensors is reduced to ensure the transmission of critical data; when the device power is low, its weight is gradually reduced and the backup node is activated to maintain the integrity of the network topology.

[0044] The updated sensor weight matrix is ​​injected into the data stream processing engine in real time. When the multidimensional tensor data stream detects that the abnormal temperature rise rate in a local area exceeds the standard, the sliding window mechanism locates the abnormal grid coordinates, and the linked device registry retrieves the wireless temperature and humidity sensor cluster in the corresponding area. The reinforcement factor is dynamically generated based on the nonlinear function of the temperature rise slope, and the target sensor weight is proportionally amplified. The preset weight upper limit constraint is used to prevent numerical overflow. The updated result is written back to the XML configuration file to form the reinforcement weight matrix.

[0045] It is further explained that the preset weight upper limit is a constraint condition set to prevent the sensor weight value from exceeding the valid range during the dynamic adjustment process, and is the engineering safety boundary of the device weight value.

[0046] The enhanced weight matrix drives the multimodal graph neural network to load structured data sets and extract the multidimensional feature vectors of sensor nodes from the device registry, including environmental parameters, video features and gas concentration extremes; the Euclidean distance between nodes is converted into standardized edge features through the Gaussian kernel function, and the device type is one-hot encoded to form the modal feature vector; the graph attention layer concatenates the multidimensional feature vectors, standardized edge features and modal feature vectors of the nodes into three-dimensional tensors, calculates the cross-modal association weights through the multi-head attention mechanism, and generates a high-order feature vector set containing spatiotemporal associations through multi-layer graph convolution aggregation.

[0047] The high-order feature vector set is input into the fully connected neural network for forward propagation. The input layer dimension is aligned with the high-order feature vector set. The hidden layer extracts the high-order representation through nonlinear transformation. The output layer uses a linear activation function to generate dimension importance scores and performs a standardized transformation to generate a feature importance report with a confidence score.

[0048] It is further explained that the feature importance report with confidence score is associated with the device registry to generate a structured data file, and at the same time triggers the hash algorithm to generate a unique data fingerprint, which is combined with the timestamp and edge server identifier to be encapsulated into an evidence data packet, and written into the distributed ledger through the blockchain consensus mechanism to form an unalterable traceability record.

[0049] S3. Input the high-order feature vector set in the feature importance report into the Transformer-based spatiotemporal prediction model, and slice the high-order feature vector set according to the time granularity to form a multi-time scale input sequence; the encoder layer of the Transformer-based spatiotemporal prediction model performs position encoding on the input sequence of each time granularity separately, uses a sine function to generate a time-correlated position embedding vector, and adds it element-by-element to the high-order feature vector set to form a time-aware input tensor.

[0050] The time-aware input tensor enters the multi-head attention mechanism. Each attention head calculates the association weight matrix between different time steps and calculates the attention score through scaled dot product attention. The query vector, key vector and value vector all come from the input tensor of the same time granularity. The output of the multi-head attention mechanism is normalized by the layer and weighted aggregated with the high-order feature vector set to generate a spatiotemporal associated attention feature tensor.

[0051] The spatiotemporal attention feature tensor is input into the feedforward neural network layer of the encoder, and the local spatiotemporal pattern of the input tensor is extracted through nonlinear transformation, and the encoded spatiotemporal feature tensor is output; the encoded spatiotemporal feature tensor is combined with the historical risk index sequence for autoregressive prediction, and the cross-attention layer of the decoder aligns the time steps of the encoder output and the decoder input, calculates the attention distribution across time granularity, and generates a context vector that integrates multi-scale information; the context vector is mapped to the predicted value of the environmental evolution trend through the fully connected layer.

[0052] It is further explained that the predicted values ​​of environmental evolution trends include temperature change rate, gas concentration accumulation and thermal radiation intensity gradient.

[0053] The sensor deployment density, device type distribution, and historical fault records within the target grid are extracted from the device registry, and the building structure parameters in the BIM model are simultaneously loaded. The average failure interval and repair time of sensor devices are calculated using historical fault records to generate the device reliability coefficient. The frequency of extreme value occurrence is statistically analyzed using temperature and humidity monitoring data to construct an environmental sensitivity coefficient. The network topology diagram generated by Delaunay triangulation is used to analyze node connectivity and path redundancy to form a topological redundancy index.

[0054] The hierarchical analysis method is used to determine the weight distribution of equipment reliability, environmental sensitivity and topological redundancy. Equipment reliability is given a higher weight due to its significant impact on the overall risk, environmental sensitivity is given a medium weight reflecting the sensitivity of risk transmission, and topological redundancy is given a basic weight due to its ability to inhibit risk diffusion. The equipment reliability, environmental sensitivity and topological redundancy indicators are linearly combined according to the weights to output the grid vulnerability coefficient.

[0055] The predicted value of the environmental evolution trend is input into the risk index calculation engine. The three core indicators of temperature change rate, gas concentration accumulation and thermal radiation intensity gradient are multiplied by the grid vulnerability coefficient stored in the equipment registry using weighted linear combination, and the time attenuation factor is superimposed to generate a dynamic risk index.

[0056] It is further explained that the risk index calculation engine is refreshed at fixed intervals. When the risk index exceeds the risk level classification critical value for the first time, the risk index continuous monitoring mechanism is activated. If the risk index remains above the risk level classification critical value for three consecutive calculation cycles, the warning condition is confirmed to be met.

[0057] After confirming that the warning conditions have been met, the hierarchical warning engine is triggered. A level 1 warning (medium risk range) corresponds to a risk index between the basic safety threshold and the major risk threshold, indicating the presence of controllable potential risks. This indicates the need to initiate preventive measures and generate a JSON-formatted work order message containing grid coordinates, risk warning type, and recommended inspection routes. This message is written to the work order database via the REST API, triggering the automatic dispatch of inspection tasks. A level 2 warning (high risk range) corresponds to a risk index exceeding the major risk threshold but not reaching the load limit, indicating that the risk is entering a rapid evolution phase and requiring immediate proactive intervention. Voice warning text is bound to the risk coordinates, converted into an audio stream, and then injected into the emergency broadcast system's playback queue, activating control signals for the audio and visual alarm devices. A level 3 warning (extreme risk range) corresponds to a risk index exceeding the safety tolerance, indicating imminent irreversible damage. The highest level of emergency control is required, sending a hexadecimal control command containing the device code to the PLC controller of the fire sprinkler system and simultaneously activating the Modbus RTU protocol communication mechanism of the smoke exhaust system.

[0058] It is further explained that the structured alarm messages generated by all warning events follow the time format standard, and are attached with the latitude and longitude coordinates of Beidou positioning and the risk index change curve. They are transmitted to the mobile terminal through the topic subscription mechanism of the MQTT protocol, and at the same time, compressed binary notification data packets are sent to the push channel of the registered mobile terminal to complete end-to-end warning information synchronization.

[0059] S4. The command center receives structured alarm messages containing grid coordinates, risk warning types, and risk index change curves. The digital twin engine parses the Beidou positioning latitude and longitude coordinates in the alarm messages, matches the unique identifier of the building information model in the BIM model database, and loads the three-dimensional grid data of the BIM model of the corresponding area; the real-time temperature values, gas concentration extremes, and thermal radiation intensity gradients in the multidimensional tensor data stream are mapped to the device node coordinates of the BIM model, and superimposed on the corresponding positions of the three-dimensional scene through the WebGL rendering engine to generate a fusion visualization interface with dynamic parameter annotations.

[0060] The fusion visualization interface triggers the federated learning framework to extract historical case feature vectors that match the current risk warning type from the historical disposal case library. The historical case feature vectors contain the evacuation path length, resource allocation type, and coordinated response time in historical events; the cosine similarity algorithm is used to calculate the similarity score between the current environmental evolution trend prediction value and the historical case feature vector, and case records with similarity higher than the similarity threshold are screened; the screened case records are input into the path planning algorithm, and the shortest path from the risk grid to the safe area is calculated based on the topological structure of the BIM model, and an evacuation path planning plan containing priority labels is generated in combination with real-time personnel distribution.

[0061] It is further explained that the setting of the similarity threshold is to perform dimensionality reduction on the feature vectors of multiple event records in the historical disposal case library, generate a two-dimensional distribution map, and calculate the cosine similarity distribution between similar event cases. The similarity of normal working condition cases is concentrated in the range of 0.3-0.6, and the similarity of effective disposal cases is concentrated in the range of 0.7-0.95. The ROC curve shows that when the similarity is 0.8, the true positive rate is the highest and the false positive rate is the lowest. Therefore, the similarity threshold is set to 0.8.

[0062] Combining the evacuation path planning scheme with the resource allocation optimization mechanism, the resource allocation optimization mechanism calls the emergency material inventory data recorded in the equipment registry to construct a resource matrix containing material type, storage location and inventory quantity; receives the time constraint parameters in the evacuation path planning scheme and generates a time-path weight mapping table; defines the decision variable as the number of each material type allocated to each rescue point, and sets the optimization objective function to maximize the number of rescue points that meet resource requirements within a limited time window; loads the time-path weight mapping table, converts the transportation time into a path cost coefficient, and superimposes the inventory restrictions to form a linear inequality constraint group, including that the total amount of each type of material allocated does not exceed the inventory upper limit; calls the simplex method solver to iteratively optimize the objective function and the inequality constraint group, terminates the calculation when the continuous iterative change rate of the objective function value reaches the preset convergence condition, and outputs the resource allocation scheme; merges the evacuation path planning scheme and the resource allocation scheme, calls the process mining engine to analyze the task dependencies in the historical collaborative response records, and generates a multi-party collaborative process logic diagram containing the trigger conditions and handover nodes of the security, fire and medical tasks.

[0063] Further explanation: emergency supplies inventory data includes the number of firefighting equipment, the location of medical supplies, and available transportation vehicles.

[0064] Time constraint parameters include the maximum travel time of each route segment and the average speed of the transportation tool.

[0065] The allocation plan that maximizes resource coverage includes material dispatch routes, personnel allocation lists, and time node sequences.

[0066] The evacuation route planning scheme, the resource allocation scheme that maximizes resource coverage, and the multi-party collaborative process logic diagram are combined to form a draft emergency plan; the decision support mechanism analyzes the time nodes, resource allocation parameters, and path length variables in the draft emergency plan, and constructs a probability distribution model of the input parameters of the Monte Carlo simulation, in which the time delay follows the normal distribution, the resource consumption follows the Poisson distribution, and the path effectiveness follows the binomial distribution; multiple random sampling simulations are performed to statistically analyze the expected casualty probability, property loss valuation, and task completion rate indicators of different draft emergency plans, and output a Monte Carlo simulation evaluation report containing the loss value expectation and variance of each draft emergency plan.

[0067] The draft emergency plan with the lowest expected loss value in the Monte Carlo simulation assessment report is marked as the optimal disposal strategy. The electronic document of the optimal disposal strategy is transmitted to the multi-party collaborative communication platform through a standardized communication protocol. The multi-party collaborative communication platform parses the task assignment list in the electronic document and automatically creates an encrypted communication channel containing the identification of the participants from the security, fire and medical departments. The electronic document is encrypted and transmitted to the terminal devices of each participant through a secure transmission protocol. The terminal devices parse the electronic document and superimpose a visual data layer showing the task geographic coordinates, time nodes and resource requirement forms in the interactive interface.

[0068] Mobile terminals in the security, firefighting, and medical departments receive mission instructions through the visual data layer, activating the mission instruction confirmation process. Mission instructions are digitally signed using an asymmetric encryption system. The digital signature is then bound to the mission number, timestamp, and operator identity information to generate an electronic mission order. The electronic mission order is transmitted back to the command center via a secure transmission protocol, where it is associated with the coordinates of the device nodes in the BIM model, triggering the indoor positioning beacon scanning mechanism.

[0069] The indoor positioning beacon scanning mechanism uses an array of ultra-wideband positioning devices deployed in buildings to capture the radio frequency signals of rescue personnel and equipment carrying positioning identifiers in real time; it uses a geometric positioning algorithm to calculate the identifier coordinates and pushes them to the digital twin engine through a real-time communication protocol; the digital twin engine maps the real-time position coordinates of personnel and equipment to the three-dimensional scene of the BIM model, forming a visual monitoring view with dynamic trajectories and task progress.

[0070] S5. Continuously collect video streams during the incident handling process and transmit them to the edge computing node through video encoding compression. The edge computing node parses the video stream, identifies the operating status of the emergency equipment, personnel location marks, and equipment action parameters, and generates a structured operation record table with a timestamp. The operation record table is compared with the task timing in the draft emergency plan to obtain the task execution time deviation rate and path offset. When the task execution time deviation rate exceeds the time deviation rate threshold or the path offset exceeds the path offset threshold, a dynamic adjustment instruction set including task priority adjustment, resource allocation optimization, and path correction is generated.

[0071] The time deviation rate threshold is based on the standard deviation of historical task records and is set to the sum of the average deviation rate and three times the standard deviation; the path offset threshold is set according to the BIM model channel width and is set to 0.2.

[0072] The dynamic adjustment instruction set is pushed to the display layer of the augmented reality device through the low-power wide area network protocol, triggering the visual interface to update the navigation path and warning area, and synchronously updating the task status of the corresponding area in the digital twin engine. The operation records, adjustment instruction sets and positioning data generated during the adjustment process are encapsulated into event handling data packets, and a unique summary is generated by the hash algorithm. It is bound with the geographic positioning information and timestamp and submitted to the blockchain consensus node for verification; after verification, it is written into the distributed ledger to form an unalterable evidence chain with a chain structure.

[0073] The system extracts structured alarm messages and multi-dimensional tensor data streams stored in the immutable evidence chain, parses the environmental evolution trend prediction values ​​and video key frame feature vectors, aligns the environmental evolution trend prediction values ​​with the video key frame feature vectors in time series, and calculates the mean square error (MSE). It also performs a distribution consistency check on the risk index, path offset, and task execution time deviation rate, and outputs a performance evaluation report containing error distribution characteristics and significance levels. When the error index exceeds the historical average error or fails the distribution consistency check in consecutive cycles, the federated learning framework is activated, and incremental datasets are extracted. After anonymization, sensitive information is removed and data generalization is performed to generate a desensitized dataset. The desensitized dataset is converted into ciphertext blocks using an encryption algorithm and transmitted to the central parameter storage node of the federated learning framework via a secure transmission protocol. After decryption by the central parameter storage node, the dataset is divided into training and validation sets and input into the spatiotemporal prediction model for gradient descent optimization of the attention weight matrix and position encoding parameters. When the descent optimization index reaches stability, an update instruction containing the updated attention weight matrix and position encoding parameters is generated and written back to the edge service configuration, synchronously replacing the weight parameters of the graph attention layer and the edge feature encoding parameters in the weight adjustment algorithm.

[0074] It is further explained that the incremental data set includes time series data of environmental parameters, records of changes in equipment operating status, and coordinate sequences of personnel movement trajectories.

[0075] The updated network parameters are sent to the edge nodes, triggering the update mechanism of the emergency plan knowledge base, as follows: Extract feature importance reports with confidence scores from blockchain evidence records, analyze the three core features contained therein, including sensor type weights, environmental parameter extreme values, and disposal response time, and construct a historical event feature vector matrix; use the event record graph structure data in the historical disposal case library, with each case as a node in the graph structure data, and the node attributes include evacuation path length, resource configuration type, and coordinated response time; merge the historical event feature vector matrix with the case node data, and calculate the semantic similarity between nodes through the edge feature generation logic of the multimodal graph neural network; reconstruct the graph structure based on the semantic similarity weight, call the graph attention layer multi-head attention mechanism for iterative optimization, encode the optimized graph structure data into RDF triple format, and attach the metadata tag of the device registry to form a knowledge graph with spatiotemporal attributes; write the cryptographic hash value of the knowledge graph with spatiotemporal attributes into the distributed ledger through the blockchain consensus mechanism to complete traceability and evidence storage.

[0076] This embodiment also provides an efficient event warning management system based on park management, including: The data perception module deploys multiple types of sensors and constructs a hexagonal cellular unit layout to generate a device registry and network topology. It forms a structured data set through data alignment and sliding window verification. Dynamic calculation module: The edge server loads the initial weights and calculates the dynamic weight matrix based on the LSTM network, and generates a high-order feature vector set in combination with the graph neural network; The risk warning module inputs high-order feature vector sets into the spatiotemporal prediction model to generate environmental evolution trend prediction values, integrates network topology and BIM model parameters to calculate dynamic risk indexes and trigger graded warnings; In the strategy optimization module, the command center analyzes warning signals to generate a fusion visualization interface, matches historical cases through federated learning, optimizes resource allocation plans, and outputs the optimal response strategy through Monte Carlo simulation. The dynamic update module collects data in real time during the processing process to generate dynamic adjustment instructions, update network parameters, and reconstruct the knowledge graph through federated learning to complete blockchain evidence storage.

[0077] This embodiment also provides a computer device, which is suitable for the case of an efficient event warning management method based on campus management, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the efficient event warning management method based on campus management proposed in the above embodiment.

[0078] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0079] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the efficient event warning management method based on campus management proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0080] In summary, the present invention solves the resource waste and sensitivity imbalance problems caused by static allocation through: LSTM network contribution analysis, integration of real-time network bandwidth and device power parameters, and dynamic adjustment of sensor weights, thereby improving the weight of infrared sensors in highly thermally sensitive areas and improving the efficiency of abnormal capture; designing differentiated timing correction strategies based on the characteristics of heterogeneous data streams, reducing the time axis alignment error of multidimensional data to the millisecond level, and ensuring the spatiotemporal consistency of the fused data set; coupling the grid vulnerability coefficient through the spatiotemporal prediction model, constructing a risk index calculation engine, and realizing the joint deduction of environmental evolution trends and network topology, thereby improving the accuracy of early warning; the knowledge graph update mechanism driven by the federated learning framework, combined with blockchain evidence storage and Monte Carlo simulation evaluation, optimizes the dynamic adaptability of emergency plans, improves the efficiency of resource scheduling path planning, and shortens the collaborative response time.

[0081] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An efficient event warning management method based on park management, characterized by: include, Deploy multiple types of sensors and build a hexagonal cellular unit layout to generate a device registry and network topology. Then, form a structured data set through data alignment and sliding window validation. The edge server loads the initial weights and calculates the dynamic weight matrix based on the LSTM network, and generates a high-order feature vector set in combination with the graph neural network; The high-order feature vector set is input into the spatiotemporal prediction model to generate the predicted value of environmental evolution trend, and the network topology and BIM model parameters are integrated to calculate the dynamic risk index and trigger graded warnings; The command center analyzes warning signals to generate a fusion visualization interface, matches historical cases through federated learning, optimizes resource allocation plans, and outputs the optimal response strategy through Monte Carlo simulation. Real-time data collection and processing generates dynamic adjustment instructions, updates network parameters, and reconstructs the knowledge graph through federated learning to complete blockchain evidence storage.

2. The efficient event warning management method based on park management according to claim 1, characterized in that: The deployment of multiple types of sensors includes the following steps: Deploy wireless temperature and humidity sensors at grid coordinate points, infrared thermal imaging sensors at building entrances and exits, video surveillance sensors on top of streetlight poles, and gas concentration sensors in underground pipe corridors and distribution rooms. All sensors have built-in Beidou positioning chips to generate device fingerprint data packets; A hexagonal cellular unit layout is constructed based on the geographical location and a self-organizing network configuration table is generated.

3. The efficient event warning management method based on park management according to claim 1 is characterized in that: The forming of the structured data set comprises the following steps: Encapsulate the data collected by the sensor into raw data packets and transmit them to the edge computing node through dual channels; The Kalman filter algorithm is used to align the time axis of asynchronous data streams within the same grid, construct a multidimensional tensor data stream, and verify the data integrity to form a structured data set.

4. The efficient event warning management method based on park management according to claim 1 is characterized in that: The generation of high-order feature vector sets The following steps are included: The LSTM network processes the historical data contribution sequence and outputs a contribution coefficient matrix. This matrix is ​​weighted and summed based on the real-time network bandwidth and the remaining power of the device to generate a dynamically adjusted weight update vector. Performing Hadamard product operation on the dynamically adjusted weight update vector and the initial weight, combined with network topology normalization and weight upper limit constraint, generates an enhanced weight matrix; Multimodal graph neural networks are used to aggregate node features and edge features to generate a set of high-order feature vectors.

5. The efficient event warning management method based on park management according to claim 1 is characterized in that: The triggering of the graded warning includes the following steps: The predicted value of environmental evolution trend is combined with the grid vulnerability coefficient and the time decay factor is superimposed to generate a dynamic risk index; According to the continuous period of the risk index exceeding the critical value, the third-level warning is triggered and a warning message is generated; The early warning message is attached with Beidou coordinates and risk index change curve, and synchronized to mobile terminals and service endpoints through the MQTT protocol.

6. The efficient event warning management method based on park management according to claim 1, characterized in that: The output of the optimal disposal strategy includes the following steps: Analyze BIM model coordinates and render real-time data into a 3D scene to generate a fusion visualization interface; Match historical case feature vectors through federated learning, filter similar cases, and calculate the shortest evacuation path; Construct a resource matrix and a time-path weight mapping table, and use the simplex method to solve the solution that maximizes resource coverage; The loss expectation is evaluated through Monte Carlo simulation and the optimal disposal strategy is output.

7. The efficient event warning management method based on park management according to claim 1 is characterized in that: The reconstruction of the knowledge graph includes the following steps: Parse the operation records and path offsets in the disposal data, generate dynamic adjustment instructions and update the digital twin engine; Desensitize and encrypt incremental data sets, optimize spatiotemporal prediction model parameters, and write them back to edge nodes; Extract feature importance reports to construct historical event feature matrices, and generate knowledge graphs with spatiotemporal attributes through graph neural networks; Write the hash value of the knowledge graph with time and space attributes into the blockchain distributed ledger to complete traceability and evidence storage.

8. An efficient event warning management system based on park management, based on the efficient event warning management method based on park management according to any one of claims 1 to 7, characterized in that: include, The data perception module deploys multiple types of sensors and constructs a hexagonal cellular unit layout to generate a device registry and network topology. It forms a structured data set through data alignment and sliding window verification. Dynamic calculation module: The edge server loads the initial weights and calculates the dynamic weight matrix based on the LSTM network, and generates a high-order feature vector set in combination with the graph neural network; The risk warning module inputs high-order feature vector sets into the spatiotemporal prediction model to generate environmental evolution trend prediction values, integrates network topology and BIM model parameters to calculate dynamic risk indexes and trigger graded warnings; In the strategy optimization module, the command center analyzes warning signals to generate a fusion visualization interface, matches historical cases through federated learning, optimizes resource allocation plans, and outputs the optimal response strategy through Monte Carlo simulation. The dynamic update module collects data in real time during the processing process to generate dynamic adjustment instructions, update network parameters, and reconstruct the knowledge graph through federated learning to complete blockchain evidence storage.

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 steps of the efficient event warning management method based on park management described in any one of claims 1 to 7 are 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 steps of the efficient event warning management method based on park management described in any one of claims 1 to 7 are implemented.

Images