Real scene emergency surveying and mapping data management method and system based on dynamic modeling

Abstract

The invention relates to the field of three-dimensional surveying and mapping, and particularly discloses a live-action emergency surveying and mapping data management method and system based on dynamic modeling, and the method comprises the steps: carrying out the global initial sampling of a target region, building a gated circulation unit neural network model integrated with a space-time attention mechanism based on the sampling data, and taking the model as a baseline dynamic model; carrying out dynamic sampling, calculating a spatial change rate, a characteristic change rate and sudden change intensity, and generating a real-time change report; calculating an update frequency corresponding to the baseline dynamic model according to the real-time change report; and collecting new live-action emergency surveying and mapping data, and performing prediction based on the latest baseline dynamic model to obtain a risk value corresponding to the new live-action emergency surveying and mapping data. According to the method, the weight of a disaster core area is enhanced through a space-time attention mechanism, so that the model response delay is shortened, the medium-dynamic field starts medium-frequency updating, the incremental learning computing power consumption is reduced, and the low-dynamic scene is switched to low-frequency updating and redundant data archiving is combined.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional surveying and mapping, specifically to a method and system for managing real-scene emergency surveying and mapping data based on dynamic modeling. Background Technology

[0002] Because emergency mapping data is real-time, the corresponding dynamic prediction models also need to be updated in real time. Existing technologies often struggle to determine the frequency of updates for these dynamic prediction models. Current dynamic models in real-world emergency mapping data management use a fixed update frequency, which has significant limitations. When real-world data changes drastically, fixed, low-frequency updates cause the model to lag behind and fail to reflect the true extent of the disaster. When real-world data tends to stabilize, fixed high-frequency updates lead to wasted computing power and redundant data storage. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for managing real-scene emergency mapping data based on dynamic modeling, so as to solve the problems mentioned in the background art.

[0004] The present invention provides a method for managing real-scene emergency mapping data based on dynamic modeling, comprising the following steps: A global initial sampling is performed on the target region, and a gated recurrent unit neural network model with spatiotemporal attention mechanism is constructed based on the sampled data as a baseline dynamic model; Perform dynamic sampling, calculate spatial rate of change, characteristic rate of change, and intensity of sudden changes, and generate real-time change reports; The baseline dynamic model is updated according to the real-time change report based on the corresponding update frequency. New real-scene emergency mapping data is collected, and the risk value corresponding to the new real-scene emergency mapping data is predicted based on the latest baseline dynamic model. The baseline dynamic model is then updated based on the new real-scene emergency mapping data. Archive the risk values ​​and sampling records corresponding to the new real-world emergency mapping data, and archive the updated baseline dynamic model.

[0005] Furthermore, the baseline dynamic model consists of an input layer, a gated recurrent unit hidden layer, a spatiotemporal attention module, and an output layer; The input layer receives standardized multi-source sampled data and transforms this data into feature vectors of uniform dimension. The gated loop unit hidden layer is used to capture the continuous pattern of real-world data changes over time, while filtering out irrelevant noise; The spatiotemporal attention module is used to calculate the importance weights of different regional features through the spatial attention component, thereby enhancing the influence of key regional features in disaster core areas and high-change areas, and weakening redundant information in stable areas. The output layer is used to integrate the processed feature information and generate model parameters that include the basic geographic features of the region and the temporal variation patterns, forming a baseline dynamic model.

[0006] Furthermore, to capture the continuous patterns of real-world data changes over time, the temporal dependencies in the data are processed through update and reset gates. The update gate controls the retention ratio of historical features, while the reset gate adjusts the weight of the current input data.

[0007] Furthermore, the spatiotemporal attention module also assigns dynamic weights to data at different sampling times through the time attention component, focusing on data characteristics during periods of drastic change.

[0008] Furthermore, the spatial rate of change, characteristic rate of change, and intensity of abrupt changes are calculated, specifically including: When calculating the spatial change rate, the region is divided into fixed grids. The contours of ground features in the previous and next sampling periods are extracted using an edge detection algorithm. The area of ​​newly added or disappeared ground features in each grid is calculated and divided by the total area of ​​the grids to obtain the change rate of a single grid. The average value of all grids is taken as the regional spatial change rate. When calculating the feature change rate, for building features, the building facade outline is fitted by laser point cloud, and the difference in tilt angle between the previous and next periods is the building tilt angle change rate; for water level features, water level data of the same monitoring point is extracted, the difference between the current water level and the previous period is calculated and divided by the sampling interval time to obtain the water level change rate. When calculating the intensity of abrupt changes, an abrupt threshold is set. When the rate of change of any feature exceeds the corresponding threshold, it is determined to be an abrupt change. The adjacent high-change grids are merged into an abrupt change region by a spatial clustering algorithm. The total area of ​​this region is the quantified value of the intensity of the abrupt change.

[0009] Furthermore, the update frequency of the baseline dynamic model is calculated based on the real-time change report, specifically including: Extract the spatial change rate value, the specific indicators of the characteristic change rate, and the markers for sudden changes from the real-time change report. First, determine if there are any sudden changes marked in the report. If so, directly determine the update frequency of the baseline dynamic model to the frequency corresponding to the high-frequency mode. If there are no sudden changes, check the spatial change rate value. When the spatial change rate is >10%, or the characteristic change rate exceeds the preset threshold, the update frequency is set to 1-5 minutes / time. When the spatial change rate is in the 3%-10% range, and the characteristic change rate does not exceed the threshold, the update frequency is set to 10-15 minutes / time. When the spatial change rate is <3%, check the characteristic change data of three consecutive samples. If they are all stable within the threshold range, the update frequency is set to 30-60 minutes / time. If the current baseline dynamic model is in the low-frequency update mode, and a sudden change marker appears in the new real-time change report, immediately adjust the update frequency to 1-5 minutes / time.

[0010] Furthermore, risk values ​​corresponding to new real-world emergency mapping data are obtained through prediction based on the latest baseline dynamic model, specifically including: The preprocessed new data is input into the latest baseline dynamic model. The model's spatiotemporal attention module assigns higher weights to core regional features and data from periods of drastic change, amplifying the impact of key disaster areas and sudden changes. The gated loop unit processes data by updating and resetting gates, integrating historical change patterns with current data characteristics, and outputs a comprehensive assessment result including the disaster spread range, building damage level, and risk of people being trapped. This results in a risk value corresponding to the new real-scene emergency mapping data, which is determined through quantitative analysis of the intensity of ground feature changes, the rate of feature variation, and the impact range of sudden changes.

[0011] Furthermore, the baseline dynamic model is updated based on new real-world emergency mapping data, specifically including: The preprocessed new real-world emergency mapping data is feature-aligned with the historical data of the baseline dynamic model. The model's gated loop unit integrates the temporal variation patterns in the new data, updates the gates to dynamically adjust the fusion ratio of historical and new features, and resets the gates to optimize the weight allocation of the current input data. The spatiotemporal attention module recalculates the feature importance weights for areas with significant spatial changes and periods of drastic temporal changes in the new data, increases the influence weight of data in high-change areas and key periods in the model, weakens redundant information in stable areas, and adopts incremental updates, adjusting only the parameters in the model that are related to the changes in the new data.

[0012] Furthermore, archive the risk values ​​and sampling records corresponding to the new real-world emergency mapping data, specifically including: The risk values ​​corresponding to the new real-scene emergency mapping data are classified according to risk level, and associated with the corresponding sampling records. Search identifiers are created for risk values ​​and sampling records based on the sampling timestamp and area coordinates. For risk values ​​and sampling records in high-frequency mode, retain complete details; for continuous and stable risk values ​​and sampling records in low-frequency mode, merge redundant information.

[0013] Furthermore, the archived updated baseline dynamic model specifically includes: The updated baseline dynamic model is associated with and stored along with the corresponding update time marker and frequency mode. The model file contains optimized network parameters and accuracy calibration information. For models updated in high-frequency mode, complete parameter details are retained. For models updated in low-frequency mode, versions with minor parameter changes in adjacent periods are merged, and only the model states with significant feature changes are retained. Through a spatiotemporal indexing mechanism, the model is associated with the sampling data and real-time change reports of the corresponding region. At the same time, the triggering factors for parameter adjustments during the model update process are recorded.

[0014] The beneficial effects of this invention are: This invention is based on a dynamic modeling-based real-scene emergency mapping data management method that matches the frequency and intensity of change in real time. High-dynamic scenes automatically switch to high-frequency updates, strengthen the weight of the disaster core area through a spatiotemporal attention mechanism to shorten the model response delay, enable medium-frequency updates in medium-dynamic fields, and adjust parameters only in the changed areas through incremental learning, thereby reducing computing power consumption. Low-dynamic scenes switch to low-frequency updates and merge redundant data for archiving. Attached Figure Description

[0015] Figure 1 The flowchart of the real-scene emergency mapping data management method based on dynamic modeling of the present invention is shown below; Detailed Implementation

[0016] This application discloses a method for managing real-scene emergency mapping data based on dynamic modeling, such as... Figure 1 The method for managing real-scene emergency mapping data based on dynamic modeling includes the following steps: Step 1: Perform global initial sampling on the target region, and construct a gated recurrent unit neural network model that integrates spatiotemporal attention mechanism based on the sampled data as the baseline dynamic model; Step 2: Perform dynamic sampling, calculate the spatial rate of change, characteristic rate of change, and intensity of sudden changes, and generate a real-time change report; Step 3: Calculate the corresponding update frequency of the baseline dynamic model based on the real-time change report, and update the baseline dynamic model according to the update frequency; Step 4: Collect new real-scene emergency mapping data, predict the risk value corresponding to the new real-scene emergency mapping data based on the latest baseline dynamic model, and update the baseline dynamic model based on the new real-scene emergency mapping data. Step 5: Archive the risk values ​​and sampling records corresponding to the new real-world emergency mapping data, and archive the updated baseline dynamic model.

[0017] The real-scene emergency mapping data management method based on dynamic modeling proposed in this application can be implemented by a computer program. The computer program used to implement the method of this application can be written in any combination of one or more programming languages. These computer programs can be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program can be executed entirely on the machine, partially on the machine, or as a standalone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0018] For the initial sampling of the target area, the specific implementation involves covering the entire target area, including the core and peripheral areas where disasters may occur; sampling is carried out using a fixed base frequency, and optical images and laser point cloud data are acquired by using UAVs equipped with optical cameras and lidar equipment. At the same time, ground sensors are deployed to collect water level and displacement monitoring data, and human dynamic data such as personnel distribution and rescue trajectories are collected simultaneously; the collected data is preprocessed to convert it into a standardized format compatible with the national geodetic coordinate system, and outliers caused by rain and fog interference and sensor drift are removed to ensure the integrity and accuracy of the sampled data, providing a reliable data foundation for the subsequent construction of a baseline dynamic model.

[0019] A gated recurrent unit (GRU) neural network model based on sampled data and incorporating a spatiotemporal attention mechanism is used as the baseline dynamic model. Specifically, this baseline dynamic model consists of an input layer, a GRU hidden layer, a spatiotemporal attention module, and an output layer. The input layer receives standardized multi-source sampled data, including ground feature features from optical imagery, elevation information from laser point clouds, and temporal data from sensor monitoring, converting this data into feature vectors of a unified dimension. The GRU hidden layer processes temporal dependencies in the data through update and reset gates. The update gate controls the retention ratio of historical features, and the reset gate adjusts the weights of the current input data, thereby capturing the continuous patterns of real-world data changes over time while filtering out irrelevant noise. In the spatiotemporal attention module, the spatial attention component calculates the importance weights of features in different regions, enhancing the influence of key regional features in disaster core areas and high-change areas, while weakening redundant information in stable areas. The temporal attention component assigns dynamic weights to data from different sampling times, focusing on data features during periods of drastic change, thus improving the model's sensitivity to sudden changes. The output layer integrates the processed feature information to generate model parameters that include the basic geographic features of the region and the temporal variation patterns, forming a baseline dynamic model.

[0020] For dynamic sampling, calculating spatial change rate, characteristic change rate, and intensity of sudden changes, the specific implementation involves the data sampling module adjusting the sampling frequency according to the current update mode (high frequency / medium frequency / low frequency). In the core area, optical images and laser point clouds are collected every 1-5 minutes by drones, while in the edge area, they are collected every 10-60 minutes. Simultaneously, real-time data on water level and building tilt angle are acquired through ground sensors, and human dynamic data are uploaded in real time through mobile terminals. The collected data is preprocessed and converted into a standardized format under the national geodetic coordinate system, and interference from rain and fog and sensor drift values ​​are removed.

[0021] When calculating the spatial change rate, the region is divided into fixed grids. Edge detection algorithms are used to extract the contours of features from previous and subsequent sampling periods. The area of ​​newly added or disappeared features within each grid is calculated, and this area is divided by the total grid area to obtain the single-grid change rate. The average of all grids is taken as the regional spatial change rate. When calculating the feature change rate, for building features, the building facade contour is fitted using laser point clouds, and the difference in tilt angle between previous and subsequent periods is the building tilt angle change rate. For water level features, water level data from the same monitoring point is extracted, and the difference between the current and previous water level is calculated and divided by the sampling interval to obtain the water level change rate. When calculating the intensity of sudden changes, a sudden change threshold is set (e.g., a building tilt angle change exceeding 5° or a water level rise exceeding 1m within 5 minutes). When any feature change rate exceeds the corresponding threshold, it is determined to be a sudden change. A spatial clustering algorithm is used to merge adjacent high-change grids into a sudden change region, and the total area of ​​this region is the quantified value of the sudden change intensity.

[0022] For generating real-time change reports, the implementation specifically includes two parts: basic sampling information and change index analysis. The basic sampling information clearly defines the time range of data collection, the covered area (including core and peripheral areas), and the data source type (such as optical imagery, laser point clouds, sensor monitoring data, and human dynamic data). It also describes the preprocessed state of the data (e.g., converted to a unified coordinate format, and the removal of blurred images due to rain and fog, and outliers caused by sensor drift). In the change index analysis section, the spatial change rate is reflected by describing the percentage of newly added or disappeared land features within the area, and the locations of areas with significant changes are marked. The characteristic change rate is described separately for each land feature type; for example, for building features, it describes changes in their shape and tilt; for water level features, it describes the rise and fall trends of water levels at monitoring points. For the sudden change intensity section, if a sudden event occurs, it describes the event type, the approximate area where it occurred, and the types of land features affected; if there are no non-sudden events, they are clearly marked.

[0023] The specific implementation of calculating the update frequency of the baseline dynamic model based on real-time change reports involves extracting spatial change rate values, specific indicators of characteristic change rates (such as building tilt angle changes and water level changes), and markers indicating the presence of sudden changes from the real-time change reports. First, it is determined whether any sudden changes are marked in the report. If so, the update frequency of the baseline dynamic model is directly set to the frequency corresponding to the high-frequency mode. If no sudden changes are found, the spatial change rate values ​​are checked. When the spatial change rate is >10%, the update frequency is set to 1-5 minutes / time, depending on whether the characteristic change rate exceeds a preset threshold (e.g., building tilt angle >2°, water level >0.2m). If any of these conditions are met, the update frequency is set to 1-5 minutes / time. When the spatial change rate is between 3% and 10%, and none of the characteristic change rates exceed the threshold, the update frequency is set to 10-15 minutes / time. When the spatial change rate is <3%, the characteristic change data from three consecutive samples is checked. If all data remain stable within the threshold range, the update frequency is set to 30-60 minutes / time. If the current baseline dynamic model is in a low-frequency update mode (30-60 minutes / time), and a sudden change marker appears in the new real-time change report, immediately adjust the update frequency to 1-5 minutes / time to ensure high-frequency response to sudden scenarios.

[0024] For collecting new real-world emergency mapping data, the risk value corresponding to the new real-world emergency mapping data is predicted based on the latest baseline dynamic model. Specifically, this involves using a drone equipped with an optical camera and lidar to dynamically sample the target area, simultaneously receiving water level, displacement data, and human dynamic data transmitted from ground sensors. The collected optical images are used to extract ground feature outlines, the lidar point cloud is used for elevation information analysis, and the sensor data undergoes time-series calibration. Abnormal data affected by environmental interference is removed and converted to a standardized format. The preprocessed new data is then input into the latest baseline dynamic model. The model's spatiotemporal attention module assigns higher weights to core regional features and data from periods of drastic change, amplifying the impact of critical disaster areas and sudden changes. The gating loop unit processes the temporal dependencies of the data through updating and resetting gates, integrating historical change patterns with current data characteristics, and outputting a comprehensive assessment result including the disaster spread range, building damage level, and risk of people being trapped. This results in the risk value corresponding to the new real-world emergency mapping data. The risk value is determined through quantitative analysis of the intensity of ground feature changes, the rate of feature variation, and the impact range of sudden changes, directly reflecting the current emergency risk status of the area.

[0025] For updating the baseline dynamic model based on new real-world emergency mapping data, the specific implementation involves: aligning the preprocessed new real-world emergency mapping data with the historical data of the baseline dynamic model; integrating the temporal variation patterns in the new data through the model's gated recurrent units; dynamically adjusting the fusion ratio of historical and new features using the update gates; and resetting the gates to optimize the weight allocation of the current input data, thereby enhancing the capture of the latest trends. The spatiotemporal attention module recalculates the feature importance weights for areas with significant spatial changes and periods of dramatic temporal changes in the new data, increasing the influence weight of data in high-change areas and critical periods in the model while reducing redundant information in stable areas. An incremental update approach is used, adjusting only the parameters in the model related to the changes in the new data without reconstructing the overall model structure.

[0026] For archiving new real-world emergency mapping data and corresponding risk values ​​and sampling records, the specific implementation involves classifying the risk values ​​according to risk level and associating them with the corresponding sampling records. These sampling records include the sampling time range, covered area, data type (e.g., optical imagery, laser point clouds, sensor monitoring data, human dynamic data), and preprocessing status (e.g., whether converted to a standardized format, whether noise filtering has been completed). A spatiotemporal indexing mechanism is used to create retrieval identifiers for risk values ​​and sampling records based on the sampling timestamp and regional coordinates, ensuring quick queries by time range or specific region. For risk values ​​and sampling records under high-frequency patterns, complete details are retained, including the risk characteristics at the moment of sudden change and the original state of the corresponding sampling data. For continuously stable risk values ​​and sampling records under low-frequency patterns, redundant information is merged, records with similar characteristics in adjacent time periods are integrated, and only key content of characteristic change nodes is retained.

[0027] For the archived updated baseline dynamic model, the implementation involves storing the updated model in association with its corresponding update time identifier and frequency mode (high-frequency / medium-frequency / low-frequency). The model file includes optimized network parameters (weight coefficients of gated recurrent units and feature weights of spatiotemporal attention modules) and accuracy calibration information. For models updated in high-frequency mode, complete parameter details are retained, especially records related to model structure adjustments during periods of sudden changes. For models continuously and stably updated in low-frequency mode, versions with minor parameter changes in adjacent periods are merged, retaining only the model states with significant feature changes. A spatiotemporal indexing mechanism is used to link the model with sampled data and real-time change reports for the corresponding region, supporting rapid retrieval by update time period and regional range. Simultaneously, triggering factors for parameter adjustments during model updates (such as spatial rate of change values ​​and feature rate of change indices) are recorded.

[0028] Given that "calculating the baseline dynamic model update frequency based on real-time change reports" cannot effectively quantify the disorder and suddenness of data changes, this application proposes a method for updating the baseline dynamic model frequency through spatiotemporal change entropy adaptive adjustment to overcome the limitations of traditional threshold rules and achieve stronger scenario adaptability. This method includes: The first step is to define the spatial distribution entropy H. s It is used to quantify the degree of disorder in changes in geographical features within a region.

[0029] ; N: The number of grids into which the region is divided (e.g., 100×100 grids). P s,i H represents the proportion of the area of ​​newly added / disappeared features in the i-th grid to the total area of ​​that grid. sA higher value indicates more dispersed spatial changes (such as multiple small-scale collapses), requiring more frequent updates to capture details; a lower value indicates concentrated changes (such as a single large-scale landslide), which can reduce the frequency.

[0030] The second step is to define the time fluctuation entropy H. t It is used to quantify the stability of the rate of change of features.

[0031] ; Δf t : The characteristic rate of change at time t (such as the building tilt angle, the rate of change of water level); μ Δf ,σ Δf : The mean and standard deviation of Δf within the sliding window; T: Time step of the sliding window; H t A high value indicates a large fluctuation in the rate of change (such as a sudden rise or fall in water level), requiring frequent updates; a low value indicates a stable change.

[0032] Define the event burst entropy H e To quantify the unpredictability and scope of impact of emergencies.

[0033] ; M: Number of clusters of sudden events (obtained through DBSCAN spatial clustering identification); Aj: The area affected by the j-th event; A total Total area of ​​the region; I j Event severity weight (e.g., building collapse Ij=1.5, crack Ij=0.8); a high He value indicates multiple points and types of concurrent events, requiring immediate response; a low value indicates isolated or minor events.

[0034] Then calculate the spatiotemporal change entropy H. total And calculate the dynamic adjustment model update frequency, H total =αH s +βH t +γH e ; α, β, γ: weighting coefficients (optimized through training with historical data, e.g., α=0.4, β=0.3, γ=0.3); Calculate the dynamic adjustment model update frequency f update , ;f min ,f max : Minimum / maximum update frequency (e.g., 1 minute / time, 60 minutes / time); H0: Entropy threshold (e.g., for regional history H) totalThe median); k: slope factor (controls the sensitivity to frequency changes, e.g., k=2). This calculation combines information theory with disaster dynamics, continuously mapping entropy values ​​to frequency ranges to avoid frequency jumps caused by hard thresholds.

[0035] As you can understand, this application discloses an electronic device, including: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which are then executed by the at least one processor to enable the at least one processor to perform the above-described method for managing real-scene emergency mapping data based on dynamic modeling.

[0036] This application discloses a real-scene emergency mapping data management system based on dynamic modeling. It adopts a modular architecture, including an initial modeling unit, a dynamic sampling and quantization unit, a frequency decision unit, an incremental modeling unit, and a data archiving unit. Each unit interacts in real time through a standardized communication protocol to achieve adaptive adjustment of the dynamic model update frequency and efficient data management. The initial modeling unit is used to perform initial sampling of the entire target area, collecting optical images, laser point clouds, sensor monitoring data, and human dynamic data. After standardized preprocessing (converting to the national geodetic coordinate system format and removing noisy data), a gated recurrent unit neural network model with spatiotemporal attention mechanism is constructed as the baseline dynamic model. The model includes an input layer, a gated recurrent unit hidden layer, a spatiotemporal attention module, and an output layer, which can capture the spatiotemporal variation patterns of real-world data. The dynamic sampling and quantization unit adjusts the sampling frequency according to the current update mode (the sampling density in the core area is 3 times that in the edge area), continuously collects and preprocesses multi-source real-scene data; calculates the spatiotemporal coupling variability (based on the coupling value of the grid state transition probability matrix entropy and time fluctuation variance), feature gradient entropy (quantification of the disorder of feature change intensity and direction), and sudden diffusion index (characterizing the spatial radiation effect of sudden changes through density clustering and distance decay function), and generates a real-time change report containing the above indicators. The frequency decision unit receives real-time change reports and calculates the dynamic update frequency based on the spatiotemporal change entropy algorithm: by weighted fusion of spatial distribution entropy, temporal fluctuation entropy, and event burst entropy (weight coefficients are optimized through training with historical data), a comprehensive entropy value is generated and mapped to high-frequency (1-5 minutes / time), medium-frequency (10-15 minutes / time), or low-frequency (30-60 minutes / time) mode instructions. When the burst diffusion index exceeds the threshold, the high-frequency mode is forcibly triggered, solving the problem of insufficient scene adaptability of traditional threshold rules. The incremental modeling unit executes incremental updates of the baseline dynamic model according to the instructions of the frequency decision unit: In high-frequency mode, it focuses on generating differential models in high-change areas and uses GPU parallel computing to complete the fusion, completing the update within 1 minute; In medium and low-frequency modes, it optimizes local features of the model (terrain elevation, road conditions, etc.) and completes the update within 10-15 minutes and 30-60 minutes respectively; After each update, the model is calibrated by combining GNSS control points and RTK measurement data to ensure that the error is ≤5cm. At the same time, the new data features are dynamically integrated through the update gate and reset gate of the gated loop unit to enhance the model's ability to capture the latest changes. The data archiving unit stores updated baseline dynamic models, risk prediction values, and sampling records according to frequency patterns, and establishes data associations (model-sampled data-change reports) using a spatiotemporal indexing mechanism. In high-frequency mode, complete snapshots of sudden changes and model parameter details are retained, while in low-frequency mode, continuous and stable data are merged to reduce redundancy. On a weekly basis, the entropy weights and mapping rules of the frequency decision unit are optimized based on the "frequency-change index-model accuracy" association log to improve the long-term adaptability of the system.

[0037] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0038] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for managing real-scene emergency mapping data based on dynamic modeling, characterized in that, Including the following steps: A global initial sampling is performed on the target region, and a gated recurrent unit neural network model with spatiotemporal attention mechanism is constructed based on the sampled data as a baseline dynamic model; Perform dynamic sampling, calculate spatial rate of change, characteristic rate of change, and intensity of sudden changes, and generate real-time change reports; The baseline dynamic model is updated according to the real-time change report based on the corresponding update frequency. New real-scene emergency mapping data is collected, and the risk value corresponding to the new real-scene emergency mapping data is predicted based on the latest baseline dynamic model. The baseline dynamic model is then updated based on the new real-scene emergency mapping data. Archive the risk values ​​and sampling records corresponding to the new real-world emergency mapping data, and archive the updated baseline dynamic model.

2. The method for managing real-scene emergency mapping data based on dynamic modeling according to claim 1, characterized in that, The baseline dynamic model consists of an input layer, a gated recurrent unit hidden layer, a spatiotemporal attention module, and an output layer; The input layer receives standardized multi-source sampled data and transforms this data into feature vectors of uniform dimension. The gated loop unit hidden layer is used to capture the continuous pattern of real-world data changes over time, while filtering out irrelevant noise; The spatiotemporal attention module is used to calculate the importance weights of different regional features through the spatial attention component, thereby enhancing the influence of key regional features in disaster core areas and high-change areas, and weakening redundant information in stable areas. The output layer is used to integrate the processed feature information and generate model parameters that include the basic geographic features of the region and the temporal variation patterns, forming a baseline dynamic model.

3. The method for managing real-scene emergency mapping data based on dynamic modeling according to claim 2, characterized in that, To capture the continuous patterns of real-world data changes over time, the update gate and reset gate are used to handle the temporal dependencies in the data. The update gate controls the retention ratio of historical features, while the reset gate adjusts the weight of the current input data.

4. The method for managing real-scene emergency mapping data based on dynamic modeling according to claim 2, characterized in that, The spatiotemporal attention module also assigns dynamic weights to data at different sampling times through the time attention component, focusing on data characteristics during periods of drastic change.

5. The method for managing real-scene emergency mapping data based on dynamic modeling according to claim 1, characterized in that, The calculation of spatial rate of change, characteristic rate of change, and intensity of abrupt changes includes: When calculating the spatial change rate, the region is divided into fixed grids. The contours of ground features in the previous and next sampling periods are extracted using an edge detection algorithm. The area of ​​newly added or disappeared ground features in each grid is calculated and divided by the total area of ​​the grids to obtain the change rate of a single grid. The average value of all grids is taken as the regional spatial change rate. When calculating the feature change rate, for building features, the building facade outline is fitted by laser point cloud, and the difference in tilt angle between the previous and next periods is the building tilt angle change rate; for water level features, water level data of the same monitoring point is extracted, the difference between the current water level and the previous period is calculated and divided by the sampling interval time to obtain the water level change rate. When calculating the intensity of abrupt changes, an abrupt threshold is set. When the rate of change of any feature exceeds the corresponding threshold, it is determined to be an abrupt change. The adjacent high-change grids are merged into an abrupt change region by a spatial clustering algorithm. The total area of ​​this region is the quantified value of the intensity of the abrupt change.

6. The method for managing real-scene emergency mapping data based on dynamic modeling according to claim 1, characterized in that, The update frequency of the baseline dynamic model is calculated based on the real-time change report, specifically including: Extract the spatial change rate value, the specific indicators of the characteristic change rate, and the markers for sudden changes from the real-time change report. First, determine if there are any sudden changes marked in the report. If so, directly determine the update frequency of the baseline dynamic model to the frequency corresponding to the high-frequency mode. If there are no sudden changes, check the spatial change rate value. When the spatial change rate is >10%, or the characteristic change rate exceeds the preset threshold, the update frequency is set to 1-5 minutes / time. When the spatial change rate is in the 3%-10% range, and the characteristic change rate does not exceed the threshold, the update frequency is set to 10-15 minutes / time. When the spatial change rate is <3%, check the characteristic change data of three consecutive samples. If they are all stable within the threshold range, the update frequency is set to 30-60 minutes / time. If the current baseline dynamic model is in the low-frequency update mode, and a sudden change marker appears in the new real-time change report, immediately adjust the update frequency to 1-5 minutes / time.

7. The method for managing real-scene emergency mapping data based on dynamic modeling according to claim 1, characterized in that, Risk values ​​corresponding to new real-world emergency mapping data are obtained by predicting based on the latest baseline dynamic model, specifically including: The preprocessed new data is input into the latest baseline dynamic model. The model's spatiotemporal attention module assigns higher weights to core regional features and data from periods of drastic change, amplifying the impact of key disaster areas and sudden changes. The gated loop unit processes data by updating and resetting gates, integrating historical change patterns with current data characteristics, and outputs a comprehensive assessment result including the disaster spread range, building damage level, and risk of people being trapped. This results in a risk value corresponding to the new real-scene emergency mapping data, which is determined through quantitative analysis of the intensity of ground feature changes, the rate of feature variation, and the impact range of sudden changes.

8. The method for managing real-scene emergency mapping data based on dynamic modeling according to claim 1, characterized in that, The baseline dynamic model is updated based on new real-world emergency mapping data, specifically including: The preprocessed new real-world emergency mapping data is feature-aligned with the historical data of the baseline dynamic model. The model's gated loop unit integrates the temporal variation patterns in the new data, updates the gates to dynamically adjust the fusion ratio of historical and new features, and resets the gates to optimize the weight allocation of the current input data. The spatiotemporal attention module recalculates the feature importance weights for areas with significant spatial changes and periods of drastic temporal changes in the new data, increases the influence weight of data in high-change areas and key periods in the model, weakens redundant information in stable areas, and adopts incremental updates, adjusting only the parameters in the model that are related to the changes in the new data.

9. The method for managing real-scene emergency mapping data based on dynamic modeling according to claim 1, characterized in that, Archive the risk values ​​and sampling records corresponding to the new real-scene emergency mapping data, specifically including: The risk values ​​corresponding to the new real-scene emergency mapping data are classified according to risk level, and associated with the corresponding sampling records. Search identifiers are created for risk values ​​and sampling records based on the sampling timestamp and area coordinates. For risk values ​​and sampling records in high-frequency mode, retain complete details; for continuous and stable risk values ​​and sampling records in low-frequency mode, merge redundant information.

10. The method for managing real-scene emergency mapping data based on dynamic modeling according to claim 1, characterized in that, The archived updated baseline dynamic model includes: The updated baseline dynamic model is associated with and stored along with the corresponding update time marker and frequency mode. The model file contains optimized network parameters and accuracy calibration information. For models updated in high-frequency mode, complete parameter details are retained. For models updated in low-frequency mode, versions with minor parameter changes in adjacent periods are merged, and only the model states with significant feature changes are retained. Through a spatiotemporal indexing mechanism, the model is associated with the sampling data and real-time change reports of the corresponding region. At the same time, the triggering factors for parameter adjustments during the model update process are recorded.

Images