Disaster rescue resource dynamic scheduling method and system fusing three-dimensional situation reconstruction

Abstract

This invention discloses a method and system for dynamic scheduling of disaster relief resources based on 3D situational awareness reconstruction, belonging to the field of resource scheduling technology. The method includes: acquiring a multi-source data network from the disaster site and establishing a spatiotemporal data grid; performing semantic interaction between grids based on the spatiotemporal data grid to establish a spatiotemporal change link; analyzing and evolving the 3D disaster situation based on the spatiotemporal change link to identify core nodes in the situational evolution; and searching for rescue resource scheduling strategies based on the evolutionary reconstruction characteristics of the core nodes in the situational evolution to determine a dynamic scheduling strategy. This application solves the technical problems in existing technologies where disaster relief scheduling relies on static information, perception and decision-making are disconnected, scheduling schemes have poor dynamic adaptability, and response accuracy is insufficient.

Description

Technical Field

[0001] This invention relates to the field of resource scheduling technology, specifically to a method and system for dynamic scheduling of disaster relief resources based on three-dimensional situation reconstruction. Background Technology

[0002] With the increasing frequency of extreme weather and geological disasters, the efficiency of disaster emergency response and the ability to accurately allocate resources have become the core focus of emergency management system construction. Smart emergency dispatch technology that integrates multi-source perception and intelligent decision-making has become an important direction for industry development.

[0003] Existing disaster resource dispatching technologies rely heavily on static statistical data and manual assessment, which suffers from problems such as limited on-site perception dimensions, lagging updates on disaster situations, and insufficient dynamic adaptability of dispatching plans. These technologies struggle to meet the precise rescue needs in scenarios where disaster situations evolve rapidly, thus hindering the improvement of overall rescue efficiency. Summary of the Invention

[0004] This application provides a method and system for dynamic scheduling of disaster relief resources that integrates three-dimensional situational reconstruction, aiming to solve the technical problems in the existing technology of disaster relief scheduling relying on static information, disconnect between perception and decision-making, poor dynamic adaptability of scheduling schemes, and insufficient response accuracy.

[0005] In view of the above problems, this application provides a method and system for dynamic scheduling of disaster relief resources that integrates three-dimensional situational reconstruction.

[0006] The first aspect disclosed in this application provides a method for dynamic scheduling of disaster relief resources that integrates three-dimensional situational reconstruction. This method includes: acquiring a multi-source data network from the disaster site; preprocessing and aligning the multi-source data to establish a data spatiotemporal grid; performing semantic interaction between grids based on the data spatiotemporal grid to establish a spatiotemporal time-series change link; performing three-dimensional disaster situational analysis and evolution based on the spatiotemporal time-series change link to identify core nodes in the situational evolution; and searching for rescue resource scheduling strategies based on the evolutionary reconstruction characteristics of the core nodes in the situational evolution to determine a dynamic scheduling strategy.

[0007] Another aspect of this application discloses a dynamic scheduling system for disaster relief resources that integrates three-dimensional situational reconstruction. This system includes: a data spatiotemporal grid establishment module, used to acquire a multi-source data network from the disaster site, preprocess and align the multi-source data, and establish a data spatiotemporal grid; a spatiotemporal temporal change link establishment module, used to perform semantic interaction between grids based on the data spatiotemporal grid to establish a spatiotemporal temporal change link; a core node identification module, used to analyze and evolve the three-dimensional situation of the disaster based on the spatiotemporal temporal change link, and identify core nodes in the situational evolution; and a dynamic scheduling strategy determination module, used to search for rescue resource scheduling strategies based on the evolutionary reconstruction characteristics of the core nodes in the situational evolution, and determine a dynamic scheduling strategy.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: A closed-loop technical solution was adopted, which involves aligning multi-source sensing data with spatiotemporal grids, extracting semantics to construct spatiotemporal and temporal change links, generating a disaster situation evolution model through 3D reconstruction and identifying core nodes, and conducting iterative search for multi-objective scheduling with time constraints based on situation features. This solution solves the technical problems in existing technologies, such as disaster relief scheduling relying on static information, the disconnect between perception and decision-making, and insufficient dynamic adaptability of scheduling schemes. It achieves the technical effect of connecting the entire chain from disaster perception to scheduling decision-making, dynamically adapting to the evolution of the disaster situation, and significantly improving the accuracy of disaster relief resource scheduling and the efficiency of emergency response.

[0009] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0010] Figure 1 A flowchart illustrating the dynamic scheduling method for disaster relief resources that integrates three-dimensional situational reconstruction is provided for the embodiments of this application.

[0011] Figure 2 This application provides a schematic diagram of the structure of a disaster relief resource dynamic scheduling system that integrates three-dimensional situational reconstruction.

[0012] Figure labeling: Data spatiotemporal grid establishment module 11, spatiotemporal time series change link establishment module 12, core node identification module 13, dynamic scheduling strategy determination module 14. Detailed Implementation

[0013] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0014] The overall concept of the technical solution provided in this application is as follows: This application provides a method and system for dynamic scheduling of disaster relief resources that integrates 3D situational awareness reconstruction. Focusing on the dynamic scheduling of disaster relief resources, it constructs a spatiotemporal data grid by spatiotemporally aligning multi-source sensing data, extracting semantic information and establishing spatiotemporal change links, generating a disaster situational evolution model through 3D reconstruction and identifying core nodes, and finally conducting scheduling optimization with time constraints based on situational characteristics to achieve a closed-loop process from perception to decision-making.

[0015] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0016] Example 1, as Figure 1 As shown in the embodiments of this application, a method for dynamic scheduling of disaster relief resources integrating three-dimensional situational reconstruction is provided. The method includes: Step S100: Obtain the multi-source data network of the disaster site, preprocess and align the multi-source data, and establish a spatiotemporal grid for the data.

[0017] Furthermore, in step S100, acquiring a multi-source data acquisition network at the disaster site includes: establishing a multi-source monitoring network, including two or more of the following: UAV aerial photography, airborne lidar, ground 3D laser scanning, and satellite remote sensing; connecting the monitoring equipment of the multi-source monitoring network to obtain multi-source monitoring data; and establishing a multi-source data acquisition network covering the disaster site according to the data source distribution location and acquisition coverage of the monitoring equipment.

[0018] Specifically, a multi-source monitoring network is a distributed acquisition system composed of various mapping and sensing devices using different technological principles and on different platforms. It can acquire information on ground features and disaster conditions at disaster sites from multiple spatial scales and data dimensions. Drone aerial photography uses small drones as platforms, equipped with visible light or multispectral imaging equipment, to flexibly capture high-resolution surface image data of disaster sites at low to medium altitudes. Airborne lidar, mounted on drones, fixed-wing aircraft, and other flight platforms, is a laser ranging device that rapidly acquires high-precision 3D point cloud data of ground features by emitting laser pulses to the ground and receiving echo signals. It can penetrate some vegetation to reconstruct the true terrain and geometric shapes of ground features. Ground-based 3D laser scanning involves setting up 3D laser scanners at ground points to perform close-range panoramic scanning of the surrounding disaster area, acquiring high-density, high-precision 3D point cloud mapping data of local areas. This method is often used for detailed observation of key damaged areas. Satellite remote sensing relies on optical, radar, and other remote sensors carried by on-orbit satellites to conduct large-scale observations of the Earth's surface from space, acquiring wide-area disaster site images and spectral information. This allows for rapid understanding of the overall disaster distribution pattern. The data source distribution location refers to the spatial coordinates of each monitoring device during its data acquisition mission, including the installation locations of ground equipment, the flight path points of the flight platform, and the nadir position of the satellite. This serves as the baseline for spatial positioning of the data.

[0019] Specifically, considering the spatial scale, disaster characteristics, and monitoring priorities of the disaster site, two or more monitoring methods are selected from four categories: UAV aerial photography, airborne lidar, ground-based 3D laser scanning, and satellite remote sensing. A multi-scale collaborative multi-source monitoring network is established. All monitoring devices within the network are connected through on-site wireless communication networks, data relay nodes, or satellite communication links. The heterogeneous multi-source monitoring data, such as images, point clouds, and spectra, output by each device are acquired in real time or near real time. The spatial reference of the data is determined based on the data source distribution location of each device. The coverage area of ​​all data is spatially stitched, deduplicated, and filled in based on the acquisition coverage range of a single device. This eliminates coverage gaps and overlapping redundancies between different devices, ultimately establishing a multi-source acquisition data network that continuously covers the entire disaster site in space and has a clear and traceable data hierarchy.

[0020] This step achieves full-domain, multi-dimensional perception of the disaster site through complementary networking of heterogeneous monitoring devices, effectively solving the problems of coverage blind spots, single information dimension, and poor scale adaptability of single data sources, and providing comprehensive and reliable raw data support for subsequent data spatiotemporal alignment and disaster situation analysis.

[0021] Furthermore, in step S100, the multi-source collected data is preprocessed and aligned to establish a data spatiotemporal grid, including: performing spatiotemporal benchmark unification and alignment processing on the multi-source collected data according to the data collection time stamps of each multi-source monitoring device and the spatial location distribution of each device at the disaster site; and establishing the data spatiotemporal grid by performing spatiotemporal grid segmentation and multi-source data mapping based on the spatiotemporally aligned multi-source collected data, using a regular grid as the spatial segmentation unit and a unified time interval as the time segmentation unit.

[0022] Specifically, the data acquisition time stamp is a timestamp inherent to each piece of monitoring data, accurately recording the acquisition time and serving as the core basis for data time-dimensional sorting and registration. Spatial location distribution refers to the geographic spatial coordinates of each monitoring device during data acquisition and the geographic coverage area coordinates of a single data point, serving as the benchmark for data spatial positioning and regional matching. A regular grid refers to a standard spatial unit obtained by uniformly dividing the disaster site's planar space according to fixed geographic side lengths. Each unit corresponds to a unique spatial code, possessing a unified spatial scale and clear geographic boundaries. A time segmentation unit refers to a standard time window obtained by uniformly dividing the time axis according to a fixed duration. Each window corresponds to a unique time code and is used for standardized segmentation of time-series data.

[0023] Specifically, the data acquisition time stamps of all multi-source monitoring data and the spatial location distribution parameters of each monitoring device at the disaster site are extracted. The time system of all data is converted into a unified standard time, and the spatial coordinates are converted into a unified geographic coordinate system, thus completing the spatiotemporal benchmark unification of multi-source data. On this basis, time registration and spatial registration are carried out for multi-source data covering the same geographic area to correct the spatiotemporal misalignment caused by different devices due to acquisition start-up time difference and positioning error, thus completing the alignment processing of multi-source acquired data. Subsequently, combined with the spatial scale of the disaster site and the accuracy requirements of subsequent analysis, the spatial side length of the regular grid and the duration of the unified time interval are set, and the spatial range of the disaster site and the monitoring time axis are uniformly divided synchronously to generate several standardized spatiotemporal grid units. Finally, all multi-source acquired data, such as images and point clouds, that have completed spatiotemporal alignment are matched and assigned according to their corresponding spatial coordinate range and acquisition time window, and mapped to the corresponding spatiotemporal grid units, so that each grid unit integrates all multi-source data within the corresponding spatiotemporal range, and finally a data spatiotemporal grid covering the entire disaster site is established.

[0024] Preferably, the specific implementation method of spatiotemporal grid segmentation and multi-source data mapping is as follows: The spatial dimension adopts a regular square grid division mechanism; firstly, the coordinates of the disaster site are transformed to a unified planar projection coordinate system; and then the reference coordinates of the lower left corner of the outer rectangle of the site space are determined. If the spatial side length of a single grid is set to d, then the row and column numbers of the spatial grid to which any spatial data point (X, Y) belongs can be calculated using the formula: Row number , column number ,in The function is a floor function; the time dimension is based on the monitoring start time T0, and a fixed time segmentation interval is set. The time window number to which any data acquisition time t belongs is Ultimately, each spatiotemporal grid cell is uniquely coded and identified by a triplet (i, j, k). After grid segmentation, multi-source data mapping is performed. For discrete point data such as point clouds, the spatial row and column numbers and time window numbers are calculated point by point to assign data points to the corresponding spatiotemporal grids. For areal data such as images and raster remote sensing data, the grid row and column range covered by the data boundary is first calculated, then the spatial grid is matched pixel by pixel, and the corresponding time window is matched according to the data acquisition timestamp to complete the grid assignment of the entire data. For example, a UTM coordinate system is used at an earthquake disaster site, with the reference coordinates... The spatial grid has a side length d = 50m, and the time interval is... =5min, the coordinates of a certain airborne lidar point cloud data point are (354200m, 3456100m), and the data collection time is 12.5 minutes after the disaster. Substituting into the formula, we can get , , This point is mapped to the spatiotemporal grid cell numbered (2, 4, 2), thereby completing the standardized grid mapping and encapsulation of all multi-source heterogeneous data.

[0025] This step transforms heterogeneous and fragmented multi-source data into standardized four-dimensional data units through spatiotemporal benchmark unification and gridded encapsulation. This not only eliminates the spatiotemporal misalignment problem of multi-source data, but also provides a unified and efficient computing platform for subsequent batch semantic extraction and temporal evolution analysis, greatly improving the efficiency and accuracy of subsequent data processing.

[0026] Step S200: Perform semantic interaction between the grids based on the data spatiotemporal grid to establish a spatiotemporal temporal change link.

[0027] Furthermore, step S200 includes: extracting semantic features from multi-source data within each grid cell of the data spatiotemporal grid to obtain semantic information corresponding to each grid cell, wherein the semantic information includes land cover category, land cover status, and disaster impact attributes; associating the semantic information of the same spatial grid cell at different times according to the time sequence to establish a semantic temporal change sequence of each grid cell; and establishing a spatiotemporal change link of the disaster site as a whole based on the spatial adjacency relationship and temporal change sequence of each grid cell.

[0028] Specifically, land cover category refers to the classification results of land cover objects, representing the type attributes of land cover objects within the grid. It is the basic classification dimension for semantic analysis and typically includes five core land cover types strongly related to disaster relief: buildings, roads, water bodies, vegetation, and vehicles. Land cover status is used to characterize the real-time morphology and functional status of land cover under the influence of disasters. It is a core dynamic attribute reflecting the degree of disaster damage, such as building damage level and road capacity. Spatial adjacency refers to the adjacent topological relationships of each spatial grid unit in geographic space, usually including four-neighbor or eight-neighbor domains. It is the topological basis for analyzing the spatial transmission and regional linkage of disasters. Spatiotemporal change link refers to a global semantic association network that integrates the evolutionary laws of the time dimension and the adjacency relationship of the spatial dimension. It can simultaneously characterize the temporal evolution trend and spatial propagation correlation of disasters and is the core structured data carrier for subsequent disaster situation analysis.

[0029] Specifically, for each independent grid cell in the spatiotemporal data grid, semantic features are extracted from the multi-source heterogeneous data such as images and point clouds integrated within the cell. Land cover category information is obtained through pixel-by-pixel semantic classification of images, and land cover status information is obtained through three-dimensional geometric feature analysis of point clouds. These two types of basic semantic information are then fused and calculated in conjunction with disaster relief business rules to generate corresponding disaster impact attributes, completing the full-dimensional semantic information output for a single grid cell. Based on this, semantic association is carried out along the time dimension, processing all semantic information belonging to the same spatial grid cell but belonging to different time windows according to their timestamp order. The process involves row matching and concatenation to generate semantic temporal change sequences for each individual grid, thus binding the evolution of a single-region disaster in the temporal dimension. Subsequently, based on the spatial adjacency relationships of each grid unit, semantic interaction verification is performed on spatially adjacent grid units to match the continuity of land feature boundaries and the spatial consistency of disaster spread between adjacent grids. At the same time, the semantic temporal change sequences of all grids are networked together according to spatial topological relationships, so that the temporal evolution data of adjacent grids form an organic whole with spatial linkage. Finally, a spatiotemporal change link covering the entire disaster site and carrying the temporal evolution law and spatial propagation relationship is established.

[0030] Preferably, the semantic information association across time periods within the same spatial grid unit is performed according to the process of index matching, temporal regularization, dimensional association, and sequence integration. First, using the unique row and column code (i, j) of the spatial grid as the retrieval index, the full semantic data of land cover categories, land cover states, and disaster impact attributes corresponding to the grid under all time windows are extracted from the spatiotemporal grid of the global data. These are then arranged in ascending order according to the time window number k to form an initial semantic temporal set. Next, for time windows in the initial set with missing data, linear interpolation of semantic features of adjacent time periods combined with semantic weighted verification of the surrounding eight neighboring grids is used to fill in the missing data, ensuring that the temporal granularity of the time series is uniform and there are no data breaks. Subsequently, semantic association binding is carried out dimensionally: the classification labels of land cover categories are compared time-by-time, and changes in land cover types are recorded. The time nodes and change types are used to form a time-series label chain for land cover categories. For land cover status, qualitative labels such as damage level and accessibility are first mapped to continuous quantitative values. For example, the building damage level is mapped to a quantitative value of 1 to 5 from intact to completely collapsed. Then, these values ​​are concatenated in time sequence to form a quantitative time-series curve of land cover status, and the magnitude and rate of status change at adjacent times are calculated simultaneously. For disaster impact attributes, the spatial range and severity level of flooded, collapsed, and trapped points are matched moment by moment, and key time nodes of attribute addition, expansion, reduction, and decline are marked to form a disaster attribute evolution subsequence. Finally, the three types of subsequences are aligned and integrated according to a unified time axis, and semantic change feature labels are added to each time node to generate a complete semantic time-series change sequence for the spatial grid unit, completing the full-dimensional association of semantic information across time.

[0031] This step transforms scattered, multi-source raw data into a spatiotemporally linked semantic system with business implications through semantic extraction and spatiotemporal correlation. It not only clearly depicts the temporal evolution of disasters in a single region, but also establishes the correlation logic of disaster spatial propagation, providing structured and computable core data support for subsequent analysis of the three-dimensional evolution of disaster situations.

[0032] Furthermore, semantic features are extracted from the multi-source data within each grid cell of the spatiotemporal data grid to obtain semantic information corresponding to each grid cell. This includes: performing pixel-by-pixel semantic classification on the image data within each grid cell to identify land cover categories, including buildings, roads, water bodies, vegetation, and vehicles; performing three-dimensional geometric feature analysis on the point cloud data within each grid cell to determine the land cover status, including the structural damage level of buildings and the passability of roads; and combining the land cover categories and land cover status to generate disaster impact attributes corresponding to each grid cell, including flooded area markings, collapsed area markings, and trapped personnel location markings.

[0033] Specifically, the road passability status is a classification of passability based on obstacles such as road surface water accumulation, collapsed debris, and road surface cracks and collapses, which directly determines the feasibility and efficiency of rescue vehicle passage.

[0034] Specifically, using a single spatiotemporal grid cell as the basic processing unit, image data within the cell is extracted. A semantic segmentation algorithm is employed for pixel-by-pixel semantic classification, assigning each pixel in the image a corresponding category label from five categories: buildings, roads, water bodies, vegetation, and vehicles. Simultaneously, the pixel percentage and spatial distribution range of each type of feature within the grid are statistically analyzed to complete the identification and output of the feature category for that grid. Simultaneously, 3D point cloud data within the grid is extracted. After preprocessing steps including point cloud denoising, ground filtering, and single-point cloud segmentation of features, 3D geometric feature analysis is performed. For building point clouds, structural deformation and the proportion of collapsed volume are calculated by comparing them with a pre-disaster baseline 3D model to determine the structural integrity of the corresponding building. For road point clouds, the damage level is determined by extracting indicators such as road surface elevation undulation, obstacle point cloud ratio, and road surface fracture degree to assess the passability of the corresponding road and output the ground feature status information of the grid. Finally, within the same grid unit framework, the ground feature category and ground feature status results are integrated, and three types of disaster impact attributes are generated by combining disaster relief business rules: land areas with water body categories and elevations below the inundation threshold are marked as inundated areas; building-covered areas with structural damage levels of severe or above are marked as collapsed areas; and the locations of trapped personnel are marked by combining the personnel feature detection results or on-site reported points within the collapsed areas. Finally, complete multi-dimensional semantic information of the grid unit is generated.

[0035] Preferably, when performing pixel-by-pixel semantic classification, the orthophoto within a single spatiotemporal grid cell is used as input. First, radiometric normalization, geometric fine correction, and uniform resolution resampling preprocessing are performed on the image to eliminate illumination differences and geometric distortions caused by different acquisition devices. Then, DeepLabv3, finely tuned with post-disaster scene samples, is combined with a semantic segmentation network to perform pixel-level inference. The semantic segmentation network uses ResNet-50 as the backbone feature extraction network and captures multi-scale image features through a hollow spatial pyramid pooling module, taking into account both small-scale targets such as buildings and vehicles and large-scale areas such as water bodies and vegetation, including texture and contour semantic information. The decoder progressively upsamples and restores the feature map to the original image resolution. The Softmax classification layer outputs the probability of each pixel belonging to five types of land features: buildings, roads, water bodies, vegetation, and vehicles. The category with the highest probability is taken as the final semantic label for that pixel. After inference, a fully connected conditional random field is introduced to optimize the boundary smoothing and local consistency of the initial classification results, suppressing the salt-and-pepper noise and land feature boundary fragmentation problems that are prone to occur in pixel-level classification. At the same time, the elevation threshold verification of the point cloud data in the same grid is combined to correct the land feature misclassification caused by elevation anomalies. Finally, the pixel-by-pixel land feature classification results with accurate boundaries and accurate categories are output.

[0036] The semantic segmentation network is constructed hierarchically from three parts: a backbone feature extraction network, a dilated spatial pyramid pooling module, and a decoder. The backbone uses a ResNet-50 network with dilated convolutions, replacing the ordinary convolutions in the last two residual stages with dilated convolutions. The stride is set to 1, and the dilation rates are 2 and 4 respectively, expanding the receptive field while preserving higher-resolution spatial details, simultaneously outputting low-level detail feature maps and high-level semantic feature maps. The dilated spatial pyramid pooling module is connected, with five branches running in parallel: 1×1 convolutions, 3×3 dilated convolutions with dilation rates of 6 / 12 / 18, and global average pooling, capturing post-disaster data at multiple scales. The scene features large bodies of water, scattered buildings, and small vehicles with significant size differences. The output features of each branch are stitched together and then fused using a 1×1 convolution to achieve channel dimensionality reduction and feature fusion. The decoder upsamples the fused high-level features by 4 times and stitches them together with the low-level detail features of the same size output by the backbone network. Then, the feature boundaries are refined by two sets of 3×3 convolutions. Finally, the images are upsampled by 4 times to restore the original image resolution. The pixel-level classification probability of five types of land features—buildings, roads, water bodies, vegetation, and vehicles—is output through a 5-channel Softmax classification head.

[0037] The network training employs a two-stage training strategy: general pre-training and disaster scenario fine-tuning. The first stage uses publicly available remote sensing semantic segmentation datasets such as ISPRSPotsdam and LoveDA as training samples. Backbone network parameters are initialized based on ImageNet pre-trained weights, training the network to learn the classification features and boundary recognition capabilities of general land features. The second stage constructs a dedicated post-disaster scenario dataset, collecting and labeling drone aerial photography and satellite remote sensing image samples from typical disaster scenarios such as earthquakes and floods. The labeled categories strictly correspond to the five land feature categories at the inference end. Data augmentation strategies such as random cropping, horizontal and vertical flipping, illumination perturbation, Gaussian noise, and blurring are introduced during training to expand sample diversity and improve network generalization ability. The loss function uses a weighted combination of cross-entropy loss and Dice loss, with the total loss formula being: ,in For pixel-level cross-entropy loss, , Predicting probability for pixels, Pixel-level true labels are used to alleviate the imbalance problem of niche categories such as vehicles and collapsed buildings in post-disaster scenes; the optimizer used is AdamW, with an initial learning rate set to 1×10⁻⁶. -4 By combining a cosine annealing learning rate decay strategy with an early stopping mechanism, training is terminated when the average intersection-union ratio of the validation set shows no significant improvement for 10 consecutive rounds, ultimately resulting in a high-precision semantic segmentation network adapted to the complex environment of disaster sites.

[0038] Preferably, when performing 3D geometric feature analysis, the original point cloud data within a single spatiotemporal grid cell is used as input. First, statistical filtering is used to remove outliers and noise points from the point cloud. Then, a cloth-simulation filtering algorithm is used to separate the ground point set from the non-ground point set. Next, combined with the land cover category labels output from the semantic classification of the preceding image, Euclidean clustering is used to segment and obtain the corresponding building point cloud clusters and road area point clouds. For the building point cloud clusters, the corresponding building point clouds of the pre-disaster baseline 3D model are first matched, and the average elevation of the pre-disaster building roof surface is calculated. Average elevation of the top surface of the cloud corresponding to the disaster The relative collapse height difference was obtained. Simultaneously, extract the unit normal vector of the building facade point cloud. Let the unit normal vector of the k-th facade point be... The vertical unit vector is Calculate the deviation angle between the single-point normal vector and the vertical direction. ; and thus the overall average normal vector deviation of the facade is obtained. N represents the total number of point clouds on the facade, which is ultimately combined with the difference in collapse height. Average normal vector deviation Percentage of collapsed volume: ; , The three-dimensional convex hull volume of building point clouds before and after the disaster, along with three quantitative indicators, are used to classify the degree of building structural damage into five levels: intact, slightly damaged, moderately damaged, severely damaged, and completely collapsed. For road area point clouds, high-order sequences of effective road surface points are extracted. Calculate the standard deviation of road surface elevation: ; The average road surface elevation is used as a quantitative indicator of elevation undulation. Simultaneously, the number of obstacle point clouds above the road surface reference elevation is counted. Obtain the percentage of obstacle point cloud Combining the two indicators of elevation undulation and the proportion of obstacles, the passability of roads is divided into four categories: unobstructed, slightly obstructed, severely obstructed, and completely blocked.

[0039] This step achieves a hierarchical transformation from raw perception data to operational semantics by complementing and fusing multi-source data through image semantic classification and point cloud geometric analysis. It balances the comprehensiveness of ground feature classification with the accuracy of disaster status assessment, providing reliable semantic data support for subsequent temporal evolution analysis and disaster three-dimensional situation reconstruction.

[0040] Step S300: Perform three-dimensional disaster situation analysis and evolution based on the spatiotemporal change link to identify the core nodes of situation evolution.

[0041] Furthermore, step S300 includes: generating a three-dimensional situation evolution model of the disaster site at continuous moments through three-dimensional spatial interpolation and surface reconstruction based on the spatiotemporal change link; extracting the change characteristics of the disaster impact attributes of each grid unit in the time dimension from the three-dimensional situation evolution model; and marking grid units or regions whose change amplitude exceeds a preset threshold as core nodes of situation evolution.

[0042] Specifically, 3D spatial interpolation is a processing technique that uses known attribute values ​​of discrete grid cells as a basis to calculate the corresponding attribute values ​​of any point in space through mathematical interpolation algorithms. This transforms discrete grid data into a continuous spatial field, filling data gaps between grid cells and forming a continuous spatial distribution result across the entire domain. Surface reconstruction is a technical process that uses discrete elevation points and grid attribute data to construct a continuous 3D surface model through methods such as triangulation and surface fitting. It can three-dimensionally recreate the topographic features and spatial distribution of disaster sites. The 3D situation evolution model is a dynamic 3D disaster model system covering continuous time series. It consists of multiple 3D scenes of the site at various times and the distribution of disaster attributes, providing a direct and continuous view of the entire process of disaster development, spread, and decline over time. Disaster impact attribute change characteristics are indicators used to quantify the evolution of disaster impact attributes in each spatial grid over time. These indicators mainly include the magnitude, rate, and trend of attribute changes, and are the core criteria for assessing the severity of disaster development. The core node of the situation evolution refers to the grid unit or contiguous area where the disaster attributes change most drastically and the situation evolution is most active during the development of the disaster. It is the key hub of the disaster evolution and the core area for the allocation of rescue resources and key responses.

[0043] Specifically, using the established spatiotemporal change chain as input, the spatial coordinates, elevation parameters, and quantified values ​​of disaster impact attributes of all grid cells under each time window are extracted. A hybrid three-dimensional spatial interpolation algorithm combining inverse distance weighting and Kriging interpolation is used to extrapolate the disaster attributes of discrete grids to continuous spatial points across the entire disaster site, filling the data gaps between grids. Then, through irregular triangular mesh surface fitting and three-dimensional surface reconstruction, and superimposed with three-dimensional geometric data of ground features, a complete three-dimensional situation scene of the disaster site at a single moment is constructed. According to a unified fine time step, temporal interpolation frame-filling processing is performed on the three-dimensional situation scene at all discrete moments to generate the time dimension. A continuous and smooth three-dimensional situation evolution model of the disaster site enables three-dimensional dynamic simulation of the entire disaster development process. Then, using the original spatial grid unit as the statistical unit, the time-series quantification value of the corresponding disaster impact attribute within the set analysis period is extracted from the three-dimensional situation evolution model grid by grid. The time dimension change characteristics such as the change amplitude and change rate of the attribute of each grid unit are obtained by differential calculation between adjacent time points. Finally, the attribute change amplitude of each grid is compared with the preset threshold one by one. The single grid unit or the multi-grid area with spatially adjacent contiguous areas whose change amplitude exceeds the preset threshold is officially marked as the core node of situation evolution, and its spatial location, disaster attribute type and change intensity are recorded simultaneously.

[0044] The determination of the preset threshold is based on the industry classification standards in the field of disaster prevention and control, the statistical patterns of historical disasters of the same type, and the emergency response level of the current disaster. The process of benchmark anchoring, statistical initial value, business calibration, and dynamic optimization is adopted to classify and set different disaster impact attributes. First, targeting three core disaster impact attributes—changes in inundation depth, leaps in building damage levels, and changes in road accessibility—we anchored the corresponding national and industry standard threshold values ​​for flood disaster classification and building earthquake damage grading as benchmark references for threshold setting. Second, we collected spatiotemporal grid sample data from historical disaster scenarios of the same type, statistically analyzed the temporal variation distribution of disaster impact attributes in each grid unit, and took the 85th percentile of the variation sequence as the initial statistical threshold to distinguish between normal fluctuations and significant evolutions of the disaster from a probabilistic perspective. Subsequently, we conducted operational calibration based on the emergency response level, total rescue resources, and on-site control requirements of this disaster. Under high-level emergency responses, we appropriately lowered the threshold to improve situational awareness sensitivity and expand the coverage of high-risk areas, while under low-level responses, we appropriately raised the threshold to focus on the core areas where the disaster was most severe. At the same time, we established a threshold feedback correction mechanism to iteratively fine-tune the threshold based on the actual disaster data verified on-site during the disaster development process, ultimately forming a differentiated preset threshold system that adapts to the current disaster scenario and balances statistical objectivity with the practicality of rescue operations.

[0045] Preferably, the three-dimensional spatial interpolation uses the spatial coordinates of the center of each spatiotemporal grid unit and the corresponding quantified values ​​of disaster impact attributes as input samples, and adopts a hybrid interpolation strategy that adapts to different attributes to implement it step by step. First, the entire grid sample is preprocessed. Outliers with abnormal attribute values ​​are removed using the 3σ criterion. Continuous physical quantities such as elevation and inundation depth are tested for normality to meet the requirements of Kriging interpolation. Corresponding interpolation algorithms are selected for different attribute characteristics: ordinary Kriging interpolation is used for continuous quantities with strong spatial autocorrelation, such as ground elevation and inundation depth. The attenuation law of attributes with distance is quantified by fitting a Gaussian semivariogram, and the optimal unbiased estimate of the interpolation point is solved to generate a global continuous attribute field. Inverse distance weighted interpolation is used for graded quantified attributes such as building damage level and road traffic status. The distance attenuation power exponent is set to 2, and the interpolation point value is calculated by assigning weights inversely proportional to distance with the grid center as the sample point. Land feature category boundaries are introduced as hard constraints throughout the interpolation process to prohibit interpolation calculations from crossing the boundaries of land features such as water bodies, shorelines, and building foundations, avoiding cross-interference between different land feature attributes. The final output is a continuous spatial attribute raster with a resolution higher than the original grid. The interpolation accuracy is verified by leave-one-out cross-validation, and local samples are added for iterative optimization in areas where the error exceeds the standard.

[0046] The 3D surface reconstruction is based on the interpolated continuous spatial attribute field and the preprocessed multi-source point cloud data. The reconstruction scheme with semantic hierarchical constraints is used to construct an integrated 3D situational surface step by step. First, a terrain base surface is constructed using a continuous elevation field obtained through interpolation as the elevation benchmark. A constrained Delaunay triangulation algorithm is used for mesh construction, embedding the boundaries of features such as road edges, water shorelines, and building outlines as forced constraint lines into the triangulation process to ensure that triangular faces do not cross the boundaries of feature types, generating a terrain triangulation network that closely matches the distribution of real features. Second, a three-dimensional feature surface is constructed. For non-ground three-dimensional targets such as buildings and collapsed deposits, high-density point clouds of the corresponding regions are used as input. The normal vectors of the point clouds are calculated through neighborhood covariance analysis and the normal orientation is unified. The Poisson surface reconstruction algorithm is used to solve the implicit surface, fitting a closed and smooth three-dimensional surface. Then, the bottom boundaries of the features are matched with the terrain surface for elevation matching and gap repair to achieve seamless splicing of features and terrain. Disaster impact attributes such as inundation depth and damage level are mapped to the corresponding triangular units of the three-dimensional surface by assigning vertex attributes, so that the three-dimensional surface simultaneously carries both geometric and disaster semantic information, ultimately generating a complete three-dimensional situation model of the disaster site at a single moment.

[0047] This step transforms discrete temporal semantic data into a continuous and dynamic three-dimensional situational awareness system through three-dimensional interpolation and surface reconstruction, accurately identifying the core nodes where the disaster situation evolves most dramatically. This not only enables an intuitive three-dimensional simulation of the disaster development process but also precisely anchors the highest priority response targets for subsequent rescue resource allocation.

[0048] Step S400: Based on the evolution and reconstruction characteristics of the core nodes of the situation evolution, search for rescue resource scheduling strategies and determine dynamic scheduling strategies.

[0049] Furthermore, step S400 includes: extracting decision parameters required for dispatching rescue resources from the core nodes of the situation evolution, the decision parameters including the location coordinates of each demand point, the type and quantity of demanded materials, and the passability and travel time of each road segment; based on the decision parameters, constructing a dispatch optimization space with the objectives of minimizing the total arrival time of rescue resources and maximizing the utilization rate of rescue resources; performing an iterative search in the dispatch optimization space to obtain the dispatch strategy with the largest target value, which is used as the dynamic dispatch strategy, wherein the dispatch optimization space includes a search time constraint, and when the time constraint is reached, the dispatch strategy with the largest target value is selected as the dynamic dispatch strategy within the time constraint window.

[0050] Specifically, evolutionary reconstruction features refer to the set of dynamic disaster features extracted from the three-dimensional situational evolution model of the core nodes of the situational evolution. These features include information such as the rate of disaster expansion, the trend of damage changes, and fluctuations in demand intensity, serving as a core bridge connecting situational awareness and dispatch decision-making. Decision parameters are the basic input parameters supporting the operation of the rescue dispatch model, divided into demand-side and path-side parameters. The demand-side includes the location of demand points, material types, and demand quantities, while the path-side includes the road segment traffic status and travel time, forming the underlying data foundation for dispatch optimization. Demand points refer to the core disaster-stricken locations at the disaster site where there is a need for personnel rescue and material replenishment. These correspond to the spatial location of the core nodes of the situational evolution and represent the delivery targets of rescue resources. The dispatch optimization space is the solution set consisting of all feasible dispatch schemes that satisfy the constraints. It includes clear optimization objectives, constraint rules, and scheme evaluation criteria, and forms the computational domain for dispatch strategy search. Search time constraints refer to the maximum solution time limit set to ensure the timeliness of emergency decision-making, avoiding decision lag due to excessive pursuit of the optimal solution. This is a key time constraint in disaster emergency scenarios.

[0051] Specifically, using the identified core nodes of the evolving situation as the core data source, all decision parameters required for the dispatch of rescue resources are extracted. This involves locating the spatial coordinates of each core node as the demand point location, combining the node's disaster impact attribute level, disaster coverage area, and affected population size to deduce the type and quantity of corresponding required materials. Simultaneously, the system matches the terrain status assessment results of the road network associated with the core nodes to clarify the passability level of each road segment, and converts this into single-segment travel time parameters based on road grade and vehicle speed limits. Then, based on all decision parameters, a multi-objective dispatch optimization space is constructed, setting the minimization of the total arrival time of rescue resources and the maximization of rescue resource utilization as dual optimization objectives. This also incorporates constraints on the total amount of rescue materials and the scale of rescue transport capacity. Boundary conditions such as road capacity constraints and timeliness requirements for demand points are used to clarify the range of feasible solutions and the rules for calculating the comprehensive target value. Then, an intelligent optimization algorithm is used to conduct iterative search within this scheduling optimization space. Each iteration generates a set of feasible scheduling strategies, calculates their comprehensive target value, and generates a better next-generation solution through selection, crossover, and mutation operations. The target value of the solution is continuously optimized through each iteration, while a search time constraint is set throughout the process. If the preset time constraint limit is reached during the iteration process, the search is immediately terminated, and the scheduling solution with the highest comprehensive target value is directly selected within the time window of the completed iteration. If the algorithm converges within the time constraint, the optimal solution after convergence is output, and finally, a dynamic scheduling strategy for rescue resources adapted to the current disaster situation is determined.

[0052] Preferably, the scheduling optimization space is essentially a constrained multi-objective combinatorial optimization feasible region for dynamic scheduling scenarios in disaster relief. It consists of three core structural levels: a set of decision variables, a dual-objective evaluation system, and multi-dimensional constraints. The set of decision variables is the basic operational object of the iterative search, containing three types of core variables: firstly, resource allocation variables... 1 represents the quantity of type m supplies delivered by the i-th rescue vehicle; 2 represents the path decision variables. The first variable is a 0-1 Boolean variable. A value of 1 indicates that the i-th vehicle follows the route from the j-th demand point to the k-th demand point, while a value of 0 indicates that the vehicle does not pass through that route. The second variable is the arrival time variable. , representing the specific time when the i-th vehicle arrives at the j-th demand point. The dual-objective evaluation system provides a quantitative standard for judging the merits of the solutions, corresponding to two optimization directions: minimizing the total arrival time of rescue resources and maximizing the utilization rate of rescue resources. The former is calculated based on the weighted total time for the arrival of materials at all demand points, while the latter is calculated based on the comprehensive score of vehicle loading rate and category matching degree. Multi-dimensional constraints are used to define the legal boundaries of feasible solutions, covering five categories: supply and demand balance constraints, vehicle capacity constraints, road segment traffic constraints, rescue timeliness constraints, and variable value constraints. All scheduling schemes that satisfy all constraints constitute the effective feasible region of the optimization space.

[0053] The method for constructing the scheduling optimization space is as follows: First, the business parameters are mapped and converted into model parameters. Parameters such as demand point coordinates, material demand, and road segment traffic status extracted from the core nodes of the situation evolution are transformed into basic model inputs. Specifically, a weighted road network adjacency matrix is ​​constructed based on demand point coordinates and road segment travel time. The traffic weight of road segments determined to be completely blocked is set to infinity to directly eliminate invalid paths. Simultaneously, the material demand vectors of each demand point, the material supply vectors of each reserve depot, and the vehicle rated transport capacity parameters are organized. Second, a dual-objective quantitative evaluation function is constructed. The total arrival time and resource utilization rate, two objectives with different dimensions, are subjected to 0-1 normalization processing, and then combined with the current disaster relief priority settings. The first step involves weighting coefficients to obtain a comprehensive objective value that can be directly used for scheme ranking. The second step embeds full-dimensional business constraints, transforming on-site rescue rules into mathematical constraint expressions. These constraints include supply-demand balance constraints requiring the total delivery volume of a single type of material to not exceed the total inventory in the reserve and not be less than the total demand at the demand point; capacity constraints requiring the total weight and volume of materials loaded on a single vehicle to not exceed the vehicle's rated limits; and timeliness constraints requiring the arrival time of materials at each demand point to not exceed the golden response time limit for rescue. The third step generates an initial feasible solution set, using a nearest neighbor heuristic algorithm to generate a batch of initial scheduling schemes that satisfy all constraints, completing the boundary delineation and initialization of the feasible domain, ultimately forming a scheduling optimization space that can directly support iterative search operations. The dual-objective quantitative evaluation function, minimizing the total arrival time of rescue and maximizing the utilization rate of rescue resources, is obtained by normalizing and weighting the two single-objective functions. The final output is a comprehensive objective value that can be directly used for scheme ranking; a larger value indicates better overall performance of the scheduling strategy. The specific composition is as follows: The overall arrival time target for relief resources measures the timeliness of the relief response, prioritizing the speed of material delivery to severely affected areas. Its mathematical expression is: ; Where n is the total number of disaster site demand points; The disaster weight for the j-th demand point is calculated by normalizing the change range of the disaster impact attribute of that node, and the value range is (0, 1]. The more severe the disaster evolution and the more people trapped, the higher the weight, which is used to reflect the difference in rescue priority. The effective arrival time of core supplies for the j-th demand point is the arrival time of the first batch of critical supplies among all rescue vehicles serving that point.

[0054] The target for the utilization rate of rescue resources measures the efficiency of rescue capacity utilization, aiming to avoid empty vehicles and resource waste. Its mathematical expression is: ; Where m is the total number of rescue vehicles participating in this dispatch; , These are the actual total loaded weight and actual total loaded volume of the i-th vehicle, respectively. , These are the maximum rated load capacity and maximum rated volume of the i-th vehicle, respectively. , The weights for load capacity utilization and volume utilization are typically taken as follows: =0.6、 =0.4, balancing weight load capacity and space utilization efficiency.

[0055] Since the two objectives have opposite optimization directions and inconsistent dimensions, they need to be normalized to the positive score interval of [0, 1] using the extreme value method first, and then linearly weighted to obtain the final comprehensive objective value: Time-based objectives are normalized and positively converted; shorter timeframes result in higher scores. ; in , This represents the extreme value of the total arrival time for all feasible solutions in the current iteration batch.

[0056] Utilization rate is normalized; the higher the utilization rate, the higher the score. ; in , This represents the extreme value of resource utilization for all feasible solutions in the current iteration batch.

[0057] Dual-objective quantification evaluation function: ; in, , The weighting coefficients are for dual objectives, satisfying... + =1, which can be dynamically adjusted according to the emergency response level: Under a high-level emergency response, it is usually set to 1. =0.7、 =0.3, prioritizing the timeliness of rescue efforts; the post-disaster recovery phase can be adjusted to =0.5、 =0.5, taking into account resource conservation. The iterative search process continuously seeks the maximum value of F, corresponding to the scheduling strategy with optimal overall performance.

[0058] For example, in a flood disaster relief scenario, two core nodes of situation evolution were identified, and the corresponding extracted decision parameters were: demand point A (coordinates E119.2°, N30.5°) needs 150 life jackets and 200 servings of drinking water; demand point B (coordinates E119.3°, N30.4°) needs 8 inflatable boats and 500 servings of food. At the same time, it was determined that the rural road X01 connecting the points was slightly obstructed due to slight flooding, with a passage time of 30 minutes; the provincial road S201 was completely blocked due to deep flooding and could not be passed; and the county road X05, which was at a higher elevation, remained unobstructed, with a passage time of 18 minutes. When constructing the optimization space, two emergency material reserve depots within the county are taken as the starting point. The constraints are set as follows: the total arrival time of a single batch of rescue is no more than 60 minutes and the load of a single assault boat is no more than 1 ton. Particle swarm optimization algorithm is used to carry out iterative search, with a preset search time constraint of 20 seconds. If the algorithm still does not converge after 120 iterations but the running time has reached the 20-second time constraint, the 98th scheme with the shortest total arrival time and the highest material loading rate is directly selected from the 120 generated schemes as the dynamic scheduling strategy, and the delivery route and deployment order of 4 transport vehicles and 6 assault boats are determined.

[0059] This step deeply couples the dynamic evolution characteristics of the three-dimensional disaster situation with a multi-objective optimization scheduling model. By adapting the search time constraint to meet the strong timeliness requirements of emergency scenarios, it achieves both accurate matching of rescue resources and disaster needs, and takes into account the speed of rescue response and the efficiency of resource utilization. It effectively solves the problems of traditional static scheduling schemes lagging behind the development of the disaster and insufficient adaptability, and can provide scientific and efficient decision support for the dynamic allocation of resources at the disaster site.

[0060] In summary, the disaster relief resource dynamic scheduling method integrating three-dimensional situational reconstruction provided in this application has the following technical effects: 1. A complete technical closed loop has been constructed, which includes multi-source data acquisition and alignment, spatiotemporal semantic association, three-dimensional situation evolution, and dynamic resource scheduling. By reconstructing the three-dimensional situation, the barrier between front-end perception and back-end decision-making is broken down. The scheduling plan can be dynamically updated as the disaster develops, effectively solving the problems of traditional rescue scheduling relying on static information and delayed response, and significantly improving the accuracy and dynamic adaptability of disaster relief resource scheduling.

[0061] 2. By generating a three-dimensional situation evolution model of continuous time through three-dimensional spatial interpolation and surface reconstruction, the entire process of disaster development and spread can be intuitively deduced. Then, based on the magnitude of changes in disaster attributes, the core nodes of situation evolution can be identified, and the key areas with the most drastic disaster evolution and the highest priority of rescue can be accurately located, so as to accurately anchor the core response targets for subsequent rescue resource scheduling.

[0062] 3. Based on real-time situational parameters, a dual-objective scheduling optimization space is constructed. Combining iterative search and time constraint mechanisms, the optimal scheduling strategy is output, taking into account both the timeliness of rescue response and the efficiency of resource utilization. Under the premise of ensuring the timeliness of decision-making, it can achieve precise matching of resources and disaster needs, and provide scientific and efficient decision support for the dynamic allocation of resources at disaster sites.

[0063] Example 2, based on the same inventive concept as the disaster relief resource dynamic scheduling method integrating three-dimensional situational reconstruction in the aforementioned examples, such as... Figure 2 As shown in the embodiment of this application, a disaster relief resource dynamic scheduling system integrating three-dimensional situational reconstruction is provided. The system includes: The data spatiotemporal grid establishment module 11 is used to acquire a multi-source data network from the disaster site, preprocess and align the multi-source data, and establish a data spatiotemporal grid. The spatiotemporal change link establishment module 12 is used to perform semantic interaction between grids based on the data spatiotemporal grid to establish a spatiotemporal change link. The core node identification module 13 is used to perform three-dimensional disaster situation analysis and evolution based on the spatiotemporal change link to identify core nodes in the situation evolution. The dynamic scheduling strategy determination module 14 is used to search for rescue resource scheduling strategies based on the evolution and reconstruction characteristics of the core nodes in the situation evolution and determine the dynamic scheduling strategy.

[0064] Furthermore, the data spatiotemporal grid establishment module 11 is also used to perform the following steps: establish a multi-source monitoring network, including two or more of the following: UAV aerial photography, airborne lidar, ground three-dimensional laser scanning, and satellite remote sensing; connect the monitoring equipment of the multi-source monitoring network to obtain multi-source monitoring data, and establish a multi-source data acquisition network covering the disaster site according to the data source distribution location and acquisition coverage of the monitoring equipment.

[0065] Furthermore, the data spatiotemporal grid establishment module 11 is also used to perform the following steps: according to the data acquisition time stamps of each multi-source monitoring device and the spatial location distribution of each device at the disaster site, perform spatiotemporal benchmark unification and alignment processing on the multi-source acquisition data; based on the spatiotemporally aligned multi-source acquisition data, perform spatiotemporal grid segmentation and multi-source data mapping with a regular grid as the spatial segmentation unit and a unified time interval as the time segmentation unit to establish the data spatiotemporal grid.

[0066] Furthermore, the spatiotemporal change link establishment module 12 is also used to perform the following steps: extracting semantic features from multi-source data in each grid cell of the data spatiotemporal grid to obtain semantic information corresponding to each grid cell, wherein the semantic information includes land cover category, land cover status and disaster impact attributes; associating the semantic information of the same spatial grid cell at different times according to the time sequence to establish the semantic temporal change sequence of each grid cell; and establishing the overall spatiotemporal change link of the disaster site based on the spatial adjacency relationship and temporal change sequence of each grid cell.

[0067] Furthermore, the spatiotemporal change link establishment module 12 is also used to perform the following steps: perform pixel-by-pixel semantic classification on the image data in each grid cell to identify the land cover categories, which include buildings, roads, water bodies, vegetation, and vehicles; perform three-dimensional geometric feature analysis on the point cloud data in each grid cell to determine the land cover status, which includes the structural damage level of buildings and the passability of roads; and generate disaster impact attributes corresponding to each grid cell by combining the land cover categories and land cover status, which include flooded area markings, collapsed area markings, and trapped personnel location markings.

[0068] Furthermore, the core node identification module 13 is also used to perform the following steps: based on the spatiotemporal change link, a three-dimensional situation evolution model of the disaster site at continuous moments is generated through three-dimensional spatial interpolation and surface reconstruction; the change characteristics of the disaster impact attributes of each grid unit in the time dimension are extracted from the three-dimensional situation evolution model; and grid units or regions whose change amplitude exceeds a preset threshold are marked as core nodes of situation evolution.

[0069] Furthermore, the dynamic scheduling strategy determination module 14 is also used to perform the following steps: extracting decision parameters required for the scheduling of rescue resources from the core nodes of the situation evolution, the decision parameters including the location coordinates of each demand point, the type and quantity of demanded materials, and the passability status and travel time of each road segment; based on the decision parameters, constructing a scheduling optimization space with the objectives of minimizing the total arrival time of rescue resources and maximizing the utilization rate of rescue resources; performing an iterative search in the scheduling optimization space to obtain the scheduling strategy with the largest target value as the dynamic scheduling strategy, wherein the scheduling optimization space includes a search time constraint, and when the time constraint is reached, the scheduling strategy with the largest target value is selected as the dynamic scheduling strategy within the time constraint window period.

[0070] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for dynamic scheduling of disaster relief resources integrating three-dimensional situational reconstruction, characterized in that, include: Acquire multi-source data networks from disaster sites, preprocess and align the multi-source data, and establish a spatiotemporal data grid; Based on the data spatiotemporal grid, semantic interaction between each grid is performed to establish a spatiotemporal temporal change link; Based on the aforementioned spatiotemporal change links, a three-dimensional disaster situation analysis and evolution is performed to identify the core nodes of the situation evolution; Based on the evolutionary and reconstructive characteristics of the core nodes of the situation evolution, a search for rescue resource scheduling strategies is conducted to determine a dynamic scheduling strategy.

2. The method for dynamic scheduling of disaster relief resources based on three-dimensional situation reconstruction according to claim 1, characterized in that, A multi-source data acquisition network for disaster sites, including: Establish a multi-source monitoring network, including two or more of the following: UAV aerial photography, airborne lidar, ground-based three-dimensional laser scanning, and satellite remote sensing; The monitoring equipment connected to the multi-source monitoring network obtains multi-source monitoring data, and establishes a multi-source data acquisition network covering the disaster site according to the data source distribution location and acquisition coverage of the monitoring equipment.

3. The method for dynamic scheduling of disaster relief resources based on three-dimensional situation reconstruction according to claim 2, characterized in that, Preprocessing and aligning multi-source acquired data to establish a spatiotemporal data grid, including: Based on the data acquisition time stamps of each multi-source monitoring device and the spatial distribution of each device at the disaster site, the spatiotemporal reference of the multi-source acquired data is unified and aligned. Based on the spatiotemporally aligned multi-source acquisition data, spatiotemporal grid segmentation and multi-source data mapping are performed using a regular grid as the spatial segmentation unit and a uniform time interval as the time segmentation unit to establish the data spatiotemporal grid.

4. The method for dynamic scheduling of disaster relief resources based on three-dimensional situation reconstruction according to claim 1, characterized in that, Based on the aforementioned spatiotemporal data grid, semantic interaction is performed between each grid to establish a spatiotemporal temporal change link, including: Semantic features are extracted from the multi-source data in each grid cell of the spatiotemporal data grid to obtain the semantic information corresponding to each grid cell. The semantic information includes land cover category, land cover status and disaster impact attributes. Based on the time series, the semantic information of the same spatial grid unit at different times is associated to establish the semantic temporal change sequence of each grid unit; Based on the spatial adjacency relationships and temporal change sequences of each grid unit, a spatiotemporal change link for the entire disaster site is established.

5. The method for dynamic scheduling of disaster relief resources based on three-dimensional situation reconstruction according to claim 4, characterized in that, Semantic features are extracted from multi-source data within each grid cell of the spatiotemporal data grid to obtain semantic information corresponding to each grid cell, including: The image data within each grid cell is subjected to pixel-by-pixel semantic classification to identify land cover categories, which include buildings, roads, water bodies, vegetation, and vehicles. Three-dimensional geometric feature analysis is performed on the point cloud data within each grid cell to determine the status of ground features, including the structural damage level of buildings and the passability of roads. Based on the aforementioned land feature categories and land feature states, disaster impact attributes corresponding to each grid unit are generated. These disaster impact attributes include markings of inundated areas, collapsed areas, and the locations of trapped personnel.

6. The method for dynamic scheduling of disaster relief resources based on three-dimensional situation reconstruction according to claim 1, characterized in that, Based on the aforementioned spatiotemporal change links, a three-dimensional disaster situation analysis and evolution is performed to identify core nodes in the situation evolution, including: Based on the aforementioned spatiotemporal change links, a three-dimensional situation evolution model of the disaster site at continuous moments is generated through three-dimensional spatial interpolation and surface reconstruction. Extract the time-dimensional variation characteristics of the disaster impact attributes of each grid cell from the three-dimensional situation evolution model; Grid cells or regions whose changes exceed a preset threshold are marked as core nodes of situational evolution.

7. The method for dynamic scheduling of disaster relief resources based on three-dimensional situation reconstruction according to claim 1, characterized in that, Based on the evolutionary reconstruction characteristics of the core nodes of the situational evolution, a rescue resource scheduling strategy search is performed to determine a dynamic scheduling strategy, including: The decision parameters required for dispatching rescue resources are extracted from the core nodes of the situation evolution. The decision parameters include the location coordinates of each demand point, the type and quantity of demanded materials, and the passability and passage time of each road segment. Based on the aforementioned decision parameters, a scheduling optimization space is constructed with the objectives of minimizing the total arrival time of rescue resources and maximizing the utilization rate of rescue resources. An iterative search is performed in the scheduling optimization space to obtain the scheduling strategy with the largest target value, which is used as the dynamic scheduling strategy. The scheduling optimization space includes a search time constraint. When the time constraint is reached, the scheduling strategy with the largest target value is selected as the dynamic scheduling strategy within the time constraint window.

8. A disaster relief resource dynamic scheduling system integrating three-dimensional situational reconstruction, characterized in that, The system is used to execute the disaster relief resource dynamic scheduling method based on fusion three-dimensional situation reconstruction as described in any one of claims 1 to 7, the system comprising: The data spatiotemporal grid establishment module is used to acquire multi-source data networks from disaster sites, preprocess and align the multi-source data, and establish a data spatiotemporal grid. The spatiotemporal change link establishment module is used to establish spatiotemporal change links by performing semantic interaction between each grid according to the data spatiotemporal grid. The core node identification module is used to perform three-dimensional disaster situation analysis and evolution based on the spatiotemporal change links, and to identify the core nodes of the situation evolution. The dynamic scheduling strategy determination module is used to search for rescue resource scheduling strategies and determine dynamic scheduling strategies based on the evolution and reconstruction characteristics of the core nodes of the situation evolution.

Images