Emergency resource intelligent scheduling optimization method based on spatiotemporal big data

Abstract

The application discloses an emergency resource intelligent scheduling optimization method based on space-time big data, and relates to the technical field of data analysis.The emergency resource intelligent scheduling optimization method based on space-time big data comprises disaster space-time data analysis, construction of an emergency resource scheduling framework, output of a single-step emergency resource scheduling strategy and execution of all resource scheduling decision steps.The application forms time sequence linkage between the emergency resource scheduling process and the disaster evolution process by constructing a continuous resource scheduling decision framework and dynamically updating scheduling boundary conditions in different scheduling decision steps in combination with real-time disaster data, and realizes collaborative scheduling calculation under multiple rescue points, multiple resource categories and multiple constraint conditions by calculating the dynamically updated scheduling boundary conditions in combination with a group optimization algorithm.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to an intelligent scheduling and optimization method for emergency resources based on spatiotemporal big data. Background Technology

[0002] Most existing emergency resource allocation technologies are based on static disaster information or single-step optimization models, lacking continuous analysis of the spatiotemporal dynamic evolution of disasters and making them ill-suited to complex scenarios such as changes in road conditions, disaster spread, and secondary disasters. Furthermore, existing solutions typically fail to adequately consider the spatial correlations and resource competition among different disaster-stricken areas, easily leading to uneven resource allocation, delayed responses at key rescue points, and decreased overall allocation efficiency. In addition, traditional allocation methods lack continuous decision-making and dynamic boundary update mechanisms, making it difficult to achieve global collaborative optimization in complex disaster environments.

[0003] Therefore, this invention introduces spatiotemporal feature maps, an asymmetric spatial distance metric model, and a Shapley value game mechanism to achieve disaster evolution prediction and dynamic collaborative scheduling of emergency resources, thereby optimizing the real-time performance and intelligence level of emergency resource scheduling. Summary of the Invention

[0004] This invention aims to provide an intelligent scheduling and optimization method for emergency resources based on spatiotemporal big data. It constructs a dynamic intelligent scheduling scheme for emergency resources oriented towards continuous decision-making processes, realizing global optimization and real-time adaptive scheduling of emergency resources under complex disaster environments.

[0005] The emergency resource intelligent scheduling and optimization method based on spatiotemporal big data includes the following steps: Step S1: Disaster spatiotemporal data analysis; real-time acquisition of disaster spatiotemporal data of the disaster-stricken area; wherein, the disaster spatiotemporal data includes disaster geographic information data, road condition data of the disaster-stricken area, and disaster evolution spatiotemporal data; based on the disaster spatiotemporal data analysis, obtain the dynamic demand vector of disaster emergency resources for each rescue point in the disaster-stricken area. Step S2: Construct an emergency resource scheduling framework; the emergency resource scheduling framework contains several consecutive resource scheduling decision steps; for the first resource scheduling decision step, analyze the dynamic demand vector of disaster emergency resources to obtain the spatiotemporal urgency characteristics of each rescue point in the disaster area; perform initial scheduling calculation with minimizing the resource scheduling response time as the boundary condition for emergency resource scheduling to obtain the initial single-step emergency resource scheduling strategy. Step S3: Output the single-step emergency resource scheduling strategy; in the subsequent resource scheduling decision steps, update the corresponding updated emergency resource scheduling boundary conditions based on the corresponding spatiotemporal urgency characteristics of the disaster in each rescue point and the real-time acquired dynamic demand vector of disaster emergency resources; perform scheduling calculations based on the corresponding updated emergency resource scheduling boundary conditions to obtain the corresponding single-step emergency resource scheduling strategy. Step S4: Execute all resource scheduling decision steps until the dynamic demand vector of disaster emergency resources in the disaster-stricken area meets the preset disaster resource safety demand range.

[0006] As a preferred technical solution of the present invention, the specific steps for analyzing disaster-affected spatiotemporal data include: Based on disaster-affected geographic information data, an adaptive grid partitioning algorithm is used to map the disaster-affected area into several disaster-affected spatiotemporal grid units; the cumulative disaster intensity of each disaster-affected spatiotemporal grid unit is calculated by combining disaster evolution spatiotemporal data; and a disaster spatiotemporal feature map is constructed based on the cumulative disaster intensity. Based on road condition data of the disaster-stricken area, an asymmetric spatial distance measurement model with time cost as the core is established; a spatiotemporal clustering algorithm is used in combination with the asymmetric spatial distance measurement model to cluster several spatiotemporal grid units of the disaster-stricken area to obtain several rescue points. The spatiotemporal feature map of the disaster area is extrapolated using a pre-set spatiotemporal prediction model to obtain the disaster evolution trend prediction results for each rescue point. The disaster evolution trend prediction results include the disaster evolution trend characteristics and potential secondary disaster risk points. Based on the disaster evolution trend prediction results, feature vectorization processing is performed to obtain the dynamic demand vector of disaster emergency resources for each rescue point in the disaster area.

[0007] As a preferred technical solution of the present invention, the specific steps for analyzing the dynamic demand vector of disaster emergency resources to obtain the spatiotemporal urgency characteristics of each rescue point within the disaster area include: For any rescue point, the rate of change of the dynamic demand vector of disaster emergency resources in the time dimension is extracted to obtain the resource demand growth gradient characteristics; combined with the disaster evolution trend characteristics and resource demand growth gradient characteristics between adjacent rescue points, the spatial diffusion intensity of the disaster is calculated. The model uses an asymmetric spatial distance metric to analyze the golden time window for each rescue point within the first resource scheduling decision step; it calculates the resource transportation time redundancy of each rescue point based on road condition data of the disaster area and the golden time window for rescue; and it calculates resources based on the spatial diffusion intensity of the disaster and the resource transportation time redundancy to obtain the spatiotemporal urgency characteristics of the disaster.

[0008] As a preferred embodiment of the present invention, the specific steps for performing initial scheduling calculations using minimizing resource scheduling response time as the boundary condition for emergency resource scheduling include: Obtain the location distribution of all available emergency resource reserve points within the radiation range of the disaster area and the real-time inventory of heterogeneous resources in each available emergency resource reserve point; classify the real-time inventory of heterogeneous resources to obtain the first type of guarantee resources and the second type of guarantee resources; construct the time-series state matrix of resource supply side; The real-time transportation time characteristics between each available emergency resource reserve point and each rescue point are calculated using an asymmetric spatial distance metric model. An initial scheduling objective function is constructed with minimizing the resource scheduling response time as the boundary condition for emergency resource scheduling. An initial single-step emergency resource scheduling strategy is obtained by using a swarm optimization algorithm in combination with the time-series state matrix of the resource supply side and the real-time transportation time characteristics.

[0009] As a preferred technical solution of the present invention, the specific steps for updating based on the spatiotemporal urgency characteristics of the disaster at each rescue point and the real-time acquired dynamic demand vector of disaster emergency resources include: Within the current resource scheduling decision step, calculate the remaining heterogeneous resource inventory of each available emergency resource reserve point to obtain the remaining emergency resource constraint; extract the emergency resource consumption of each rescue point in the previous resource scheduling decision step to obtain the remaining emergency resource quantity of each rescue point; Each rescue point is treated as an emergency resource game unit, and an emergency resource scheduling characteristic function is constructed with the goal of minimizing the total global disaster loss. The Shapley value calculation algorithm is used to traverse several subsets of emergency resource game units. Based on the emergency resource scheduling characteristic function, the marginal contribution value of the emergency resource game unit is obtained. Based on the marginal contribution value of resources, a rescue game weight vector is generated for each rescue point; the spatiotemporal urgency characteristics of the disaster are weighted and corrected based on the rescue game weight vector for each rescue point to obtain the corrected spatiotemporal urgency characteristics of the disaster. Based on the modified spatiotemporal urgency characteristics of disaster and the dynamic demand vector of disaster emergency resources, the boundary conditions for emergency resource scheduling are updated in multiple dimensions to obtain the corresponding updated boundary conditions for emergency resource scheduling. Among them, the multi-dimensional update includes the update of resource allocation upper limit, the update of time window constraint, and the update of path reachability constraint.

[0010] As a preferred technical solution of the present invention, the specific steps of performing scheduling calculations based on the corresponding updated emergency resource scheduling boundary conditions to obtain the corresponding single-step emergency resource scheduling strategy include: using a swarm optimization algorithm in combination with the updated emergency resource scheduling boundary conditions to perform scheduling calculations to obtain the corresponding single-step emergency resource scheduling strategy.

[0011] As a preferred technical solution of the present invention, the preset spatiotemporal time series prediction model is constructed based on a graph convolutional network model.

[0012] The present invention has the following advantages: 1. This invention constructs a continuous resource scheduling decision framework and dynamically updates the scheduling boundary conditions in different scheduling decision steps by combining real-time disaster data, enabling the emergency resource scheduling process to form a temporal linkage with the disaster evolution process; by performing spatiotemporal raster modeling of the disaster-stricken area and constructing a disaster spatiotemporal feature map, it realizes a unified expression of the disaster distribution status, evolution relationship and spatial correlation characteristics within the disaster-stricken area; by combining a swarm optimization algorithm to calculate the dynamically updated scheduling boundary conditions, it realizes collaborative scheduling calculation under multiple rescue points, multiple resource categories and multiple constraints.

[0013] 2. This invention establishes an asymmetric spatial distance metric model, incorporating differences in road traffic direction, traffic disruption status, and real-time transportation time into the distance calculation process, enabling resource transportation paths to have dynamic time-varying analysis capabilities; by combining spatiotemporal clustering algorithms to dynamically cluster disaster-affected spatiotemporal grid units, it achieves adaptive generation and regionalized organization of rescue points; and by utilizing a spatiotemporal time-series prediction model based on graph convolutional networks to extrapolate the disaster evolution trend, it achieves joint prediction of disaster evolution trend characteristics and potential secondary disaster risk points. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the intelligent scheduling and optimization method for emergency resources based on spatiotemporal big data used in an embodiment of the present invention. Detailed Implementation

[0015] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0016] Example 1: An intelligent scheduling and optimization method for emergency resources based on spatiotemporal big data (see [link]). Figure 1 As shown, it includes the following steps: Step S1: Disaster spatiotemporal data analysis; real-time acquisition of disaster spatiotemporal data of the disaster-stricken area; wherein, the disaster spatiotemporal data includes disaster geographic information data, road condition data of the disaster-stricken area, and disaster evolution spatiotemporal data; based on the disaster spatiotemporal data analysis, obtain the dynamic demand vector of disaster emergency resources for each rescue point in the disaster-stricken area. In this embodiment, multi-source sensing devices deployed within the disaster-stricken area are first used to acquire real-time spatiotemporal data of the disaster. These devices include satellite remote sensing terminals, drone inspection equipment, road monitoring cameras, vehicle-mounted positioning terminals, and regional environmental sensors. The disaster-affected geographic information data is used to characterize the spatial topography distribution within the disaster-stricken area, specifically including road structure information, building distribution information, terrain elevation information, water area distribution information, and administrative boundary information. The disaster-affected area road condition data is used to characterize the real-time road traffic status, specifically including dynamic traffic information such as road congestion levels, road interruption status, water depth, landslide areas, and average vehicle speed. The disaster evolution spatiotemporal data is used to characterize the spread of the disaster over time, specifically including the trajectory of changes in the affected area, the trend of changes in disaster intensity, the status of population gathering and migration, and information on the triggering of secondary disasters.

[0017] Data from different data sources undergoes unified time alignment, mapping data collected at different times to a unified time axis based on the timestamps of each data acquisition device. Spatial coordinate unification is then performed, transforming data from different coordinate systems to a unified geographic coordinate system to eliminate spatial discrepancies between multi-source data. Data with missing values, outliers, and noise interference is preprocessed. For short-term missing data, nearest-time interpolation is used to complete the data, while a sliding window smoothing algorithm is used to correct abnormal fluctuations. This algorithm calculates the average trend of data across consecutive time periods centered on the current moment, reducing the impact of instantaneous anomalies on subsequent analysis results. After data cleaning, disaster-affected geographic information data, road condition data for disaster-affected areas, and disaster evolution spatiotemporal data are fused according to time and spatial location to form a unified disaster spatiotemporal data set, providing fundamental data support for subsequent dynamic analysis and emergency resource allocation in disaster-affected areas.

[0018] Based on the analysis of the spatiotemporal data of the disaster, a dynamic demand vector of emergency resources for each rescue point in the disaster area is obtained. The specific steps for analyzing disaster-related spatiotemporal data include: Based on disaster-affected geographic information data, an adaptive grid partitioning algorithm is used to map the disaster-affected area into several disaster-affected spatiotemporal grid units; the cumulative disaster intensity of each disaster-affected spatiotemporal grid unit is calculated by combining disaster evolution spatiotemporal data; and a disaster spatiotemporal feature map is constructed based on the cumulative disaster intensity. In this embodiment, the disaster-stricken area is first spatially discretized based on the disaster-affected geographic information data. An adaptive grid partitioning algorithm is then used to map the entire disaster-affected area into several disaster-affected spatiotemporal grid units. A disaster-affected spatiotemporal grid unit refers to a basic analysis unit formed according to the spatial location and temporal change relationship. Each grid unit corresponds to the disaster status information of a specific area and a corresponding time period. The adaptive grid partitioning algorithm does not use a fixed-size grid, but dynamically adjusts the grid size according to the complexity of the disaster situation in the region. The specific process is as follows: First, a regional complexity index is generated based on building density, road distribution density, and the frequency of disaster changes. Then, using the regional complexity index as the partitioning basis, a smaller grid is used for fine-grained partitioning in areas with drastic disaster changes, and a larger grid is used for coarse-grained partitioning in areas with slower disaster changes. Finally, grids of different scales are mapped to the same spatiotemporal coordinate system.

[0019] After the grid division is completed, the cumulative disaster intensity of each disaster spatiotemporal grid unit is calculated by combining the disaster evolution spatiotemporal data. The cumulative disaster intensity is used to characterize the comprehensive disaster degree of the corresponding area in a continuous time period. The calculation process includes: extracting parameters such as disaster intensity change value, disaster duration, disaster spread speed and personnel impact range, normalizing each parameter and assigning corresponding weights, and then forming the cumulative disaster intensity value corresponding to the grid unit through a weighted accumulation method. A disaster spatiotemporal feature map is constructed based on the spatial adjacency and temporal evolution relationships between each grid cell. The disaster spatiotemporal feature map is a network structure used to describe the disaster propagation relationship and evolutionary association relationship between different regions. The nodes in the map correspond to each disaster spatiotemporal grid cell, and the connecting edges between nodes are used to represent the disaster spread path and the spread association strength. The weight of the connecting edges is calculated based on the disaster propagation speed, geographical adjacency, and traffic connectivity.

[0020] It should be noted that the disaster intensity change value, the duration of the disaster, the speed of the disaster spread, and the scope of human impact are core spatiotemporal characteristic parameters for quantitatively describing the evolution of the disaster. The disaster intensity change value is used to characterize the increase or decrease in the degree of disaster impact per unit time. It is usually calculated from multi-source monitoring data, such as the change of disaster index retrieved from remote sensing images, and the differential results of continuous monitoring data such as water level, temperature, smoke concentration, or vibration intensity collected by sensors. The duration of disaster refers to the length of time that a certain spatial area experiences from the occurrence of a disaster to the dissipation of its impact or the attainment of a stable state. It is generally obtained by identifying the start and end points of disaster event timestamp records. Data sources include emergency event reporting systems, historical disaster databases, and real-time event streams. The disaster spread rate describes how quickly the disaster's impact area expands over time. Its calculation depends on the changes in the disaster's spatial boundaries at different time points, and is usually obtained through spatial comparative analysis of the disaster area boundaries at consecutive time points, such as by using raster cell state changes or vector boundary expansion rates. The impact range on people represents the spatial distribution and scale of the population affected by the disaster. It is generally estimated by combining population geographic information data, mobile communication signaling data, positioning terminal trajectory data, and emergency registration information, and is obtained by spatially overlaying the disaster-covered area with the population distribution raster. The above parameters are usually derived from a multi-source heterogeneous data fusion system, including satellite remote sensing systems, UAV patrol data, ground sensor networks, traffic and road condition monitoring systems, and emergency management information platforms. After unified time alignment and spatial standardization, they are used for subsequent disaster modeling and resource demand analysis.

[0021] Based on road condition data of the disaster-stricken area, an asymmetric spatial distance measurement model with time cost as the core is established; a spatiotemporal clustering algorithm is used in combination with the asymmetric spatial distance measurement model to cluster several spatiotemporal grid units of the disaster-stricken area to obtain several rescue points. After completing the construction of the spatiotemporal feature map of the disaster area, an asymmetric spatial distance measurement model with time cost as the core is established based on the road condition data of the disaster area. Asymmetric spatial distance refers to the fact that the transportation time from region A to region B is not exactly the same as the transportation time from region B to region A. The difference comes from factors such as one-way road blockage, differences in the direction of traffic congestion, and dynamic changes in the scope of disaster impact.

[0022] In the specific construction process, road condition parameters such as road traffic speed, road blockage level, road slope change, and water depth are first extracted. Then, the corresponding time travel cost is calculated according to different road conditions, where the higher the degree of road blockage, the greater the corresponding time cost. After that, a directional transportation path relationship is established based on road directionality, thereby forming an asymmetric spatial distance matrix with directional differences. After completing the distance modeling, a spatiotemporal clustering algorithm combined with an asymmetric spatial distance metric model is used to perform cluster analysis on multiple disaster-affected spatiotemporal grid units.

[0023] Spatiotemporal clustering algorithms refer to regional aggregation methods that simultaneously consider spatial proximity and temporal evolution. The specific steps include: First, using the cumulative disaster intensity, disaster evolution speed, and time travel cost of each grid cell as clustering features; then, calculating the spatiotemporal correlation between any two grid cells; when the correlation exceeds a preset threshold, dividing the corresponding grid cells into the same cluster region; subsequently, performing boundary correction and isolated region merging on the formed cluster regions, ultimately forming several rescue points, each corresponding to a set of regions with similar disaster conditions and transportation characteristics.

[0024] The spatiotemporal feature map of the disaster area is extrapolated using a pre-set spatiotemporal prediction model to obtain the disaster evolution trend prediction results for each rescue point. The disaster evolution trend prediction results include disaster evolution trend features and potential secondary disaster risk points. Based on the disaster evolution trend prediction results, feature vectorization processing is performed to obtain the dynamic demand vector of disaster emergency resources for each rescue point in the disaster area. After obtaining information from various rescue points, a pre-set spatiotemporal prediction model is used to analyze the spatiotemporal characteristics of the disaster area. The spatiotemporal prediction model is based on a graph convolutional network model, which is a data processing model that can simultaneously analyze spatial relationships and temporal changes. Its core idea is to learn the spread pattern of disaster in different regions by propagating information between nodes through a graph structure.

[0025] In the specific prediction process, the spatiotemporal feature map of the disaster is first input into the graph convolutional network model, and the cumulative disaster intensity, traffic status and historical disaster change sequence of each grid cell are used as input features; the disaster impact relationship between each node is calculated through the information propagation of adjacent nodes; the future trend of disaster change is analyzed by using the time-series recursion method in the time dimension, so as to output the disaster evolution trend prediction results corresponding to each rescue point.

[0026] The disaster evolution trend prediction results include disaster evolution trend characteristics and potential secondary disaster risk points. The disaster evolution trend characteristics characterize the future spread direction, spread speed, and disaster intensity trends of the disaster, while potential secondary disaster risk points characterize risk areas that may experience landslides, road collapses, or flooding. The disaster evolution trend prediction results undergo feature vectorization, converting different types of disaster indicators into a unified set of numerical features. Specifically, this includes extracting feature parameters such as the degree of demand for medical supplies, food, personnel relocation, road repair, and potential risk spread levels. These feature parameters are then standardized and combined in a preset order to form a multi-dimensional feature vector. Finally, a dynamic demand vector for emergency resources corresponding to each rescue point within the disaster area is obtained, which is used for subsequent intelligent emergency resource scheduling and analysis.

[0027] In this embodiment, the preset spatiotemporal prediction model is constructed based on graph convolutional networks and time series recursion mechanisms. The spatiotemporal feature map of the disaster is transformed into graph structure data, where nodes represent spatiotemporal grid cells or rescue points of the disaster, and node features include cumulative disaster intensity, historical disaster change sequence, road traffic status and personnel impact indicators. Edges represent the spatial adjacency relationship and disaster propagation relationship between different nodes, and the edge weights are jointly determined by asymmetric spatial distance and disaster diffusion intensity.

[0028] The algorithm employs graph convolution operations to aggregate and compute node neighborhood information. Its core process involves extracting the features of each node's neighboring nodes and performing a weighted summation. Simultaneously, it learns spatial correlation feature representations through parameterized transformation functions, thereby obtaining a node implicit representation that integrates spatial dependencies. In the time dimension, a temporal recursive structure is introduced to model the node state at different time sections. By recursively updating the node implicit representation at the current moment with the prediction result at the previous moment, the algorithm captures the dynamic changes in the disaster situation over time.

[0029] During model training, historical disaster development trajectories are used as supervisory signals. Parameter optimization is performed by minimizing the error function between the predicted results and the actual disaster state, enabling the model to learn the spatiotemporal evolution patterns of disaster spread, intensification, and attenuation. Finally, by forward extrapolating the spatiotemporal graph structure input at the current moment, the model outputs the disaster evolution trend characteristics and potential secondary disaster risk distribution results for multiple future time steps, which are used for subsequent emergency resource demand vector generation and scheduling decisions.

[0030] Step S2: Construct an emergency resource scheduling framework; the emergency resource scheduling framework contains several consecutive resource scheduling decision steps; for the first resource scheduling decision step, analyze the dynamic demand vector of disaster emergency resources to obtain the spatiotemporal urgency characteristics of each rescue point in the disaster area; perform initial scheduling calculation with minimizing the resource scheduling response time as the boundary condition for emergency resource scheduling to obtain the initial single-step emergency resource scheduling strategy. Constructing an emergency resource scheduling framework refers to establishing a multi-stage resource allocation decision-making system oriented towards time-series evolution, used to optimize the scheduling of relief resources in stages during the continuous changes of the disaster situation. The framework consists of several consecutive resource scheduling decision steps, where each resource scheduling decision step represents an independent scheduling calculation unit executed within a fixed time interval based on the current disaster situation and resource status. The decision steps form a sequential relationship through a state transfer mechanism, thereby realizing the continuous evolution of the scheduling process.

[0031] The specific steps for analyzing the dynamic demand vector of disaster emergency resources to obtain the spatiotemporal urgency characteristics of each rescue point within the disaster-stricken area include: For any rescue point, the rate of change of the dynamic demand vector of disaster emergency resources in the time dimension is extracted to obtain the resource demand growth gradient characteristics; combined with the disaster evolution trend characteristics and resource demand growth gradient characteristics between adjacent rescue points, the spatial diffusion intensity of the disaster is calculated. Extracting the rate of change of the dynamic demand vector of disaster emergency resources for any rescue point in the time dimension refers to performing differential or trend fitting processing on the resource demand values ​​of the rescue point at multiple consecutive time sampling points to characterize the speed at which resource demand changes over time. The dynamic demand vector for disaster relief resources is used to represent the combination of demand intensity for various types of resources, such as medical supplies, food, drinking water, rescue equipment, and personnel transfer, at the rescue point. The specific processing procedure is as follows: First, the demand vector is aligned with the time series according to a uniform time interval; then, the difference between the demand vectors at adjacent moments is extracted using a sliding time window, and the slope of its changing trend is calculated using the least squares fitting method; finally, this slope is used as the gradient feature of resource demand growth to characterize the growth rate and direction of demand.

[0032] After obtaining the characteristics of the resource demand growth gradient, a joint analysis is performed on the disaster evolution trend characteristics between the current rescue point and its neighboring rescue points. The disaster evolution trend characteristics refer to the information on the direction, intensity, and scope of disaster spread obtained through the spatiotemporal prediction model. Specifically, when calculating the spatial diffusion intensity of the disaster, the spatial neighborhood set of the current rescue point is first determined, i.e., neighboring rescue points are selected based on geographical distance and accessibility. Then, the correlation between the disaster evolution trend characteristics of each neighboring rescue point and the resource demand growth gradient characteristics of the current rescue point is calculated. The consistency between the two can be measured by weighted correlation coefficient or vector cosine similarity. Finally, the correlation results of all neighboring rescue points are weighted and accumulated, and corrected by spatial distance attenuation factor, so as to obtain the spatial diffusion intensity of the disaster, which is used to describe the spatial propagation ability and impact expansion trend of the disaster.

[0033] The model uses an asymmetric spatial distance metric to analyze the golden time window for each rescue point within the first resource scheduling decision step; it calculates the resource transportation time redundancy of each rescue point based on road condition data of the disaster area and the golden time window for rescue; and it calculates resources based on the spatial diffusion intensity of the disaster and the resource transportation time redundancy to obtain the spatiotemporal urgency characteristics of the disaster.

[0034] When calculating the golden time window for rescue, an asymmetric spatial distance metric model is used to analyze each rescue point. This model is used to characterize the asymmetric time cost characteristics between resource supply points and rescue points due to road directionality, congestion differences, and blockage situations. The specific steps are as follows: First, the one-way travel time cost of each path is calculated based on road condition data to form a directed time distance matrix. Then, combined with the speed of disaster development and the degree of impact on people, the maximum delay time interval for each rescue point is determined under the premise of no serious risk expansion. This time interval is the golden time window for rescue, which is essentially the time boundary between the current state of the disaster and the uncontrollable state.

[0035] After obtaining the golden time window for rescue, the resource transportation time redundancy is calculated by combining road condition data of the disaster area. The road condition data includes information such as road traffic speed, congestion level, blockage status, and dynamic traffic capacity. The specific calculation process is as follows: First, the actual resource transportation time is calculated based on the current road condition and an asymmetric spatial distance model. Then, the transportation time is compared with the upper limit of the golden time window for rescue, and the difference or proportional relationship between the two is calculated. Finally, the resource transportation time redundancy is obtained through normalization, which is used to characterize whether there is a time margin or time urgency for resource arrival under the current traffic conditions.

[0036] After considering both the spatial diffusion intensity of the disaster and the time redundancy of resource transportation, a spatiotemporal urgency feature for the disaster is constructed. The calculation process is as follows: First, the spatial diffusion intensity of the disaster and the time redundancy of resource transportation are standardized to place them in a unified dimension space. Then, a weighted fusion mechanism is introduced, and weight coefficients are set according to the degree of impact of disaster diffusion and the degree of transportation time constraints. The two indicators are then weighted and summed or nonlinearly combined. Finally, the spatiotemporal urgency feature for the disaster is output, which is used to comprehensively characterize the resource urgency and disaster development risk level of the rescue point in the current spatiotemporal state, and serves as an important input parameter for subsequent emergency resource scheduling decisions.

[0037] The initial single-step emergency resource scheduling strategy is obtained by performing initial scheduling calculations with the minimum resource scheduling response time as the boundary condition for emergency resource scheduling. The specific steps for initial scheduling calculations using minimizing resource scheduling response time as the boundary condition for emergency resource scheduling include: Obtain the location distribution of all available emergency resource reserve points within the radiation range of the disaster area and the real-time inventory of heterogeneous resources in each available emergency resource reserve point; classify the real-time inventory of heterogeneous resources to obtain the first type of guarantee resources and the second type of guarantee resources; construct the time-series state matrix of resource supply side; In this embodiment, the location distribution of all available emergency resource reserve points within the disaster-stricken area's radiation range and their real-time heterogeneous resource inventory information are first obtained. The radiation range refers to the area covered by dispatchable resources, expanded based on the disaster center area and considering road accessibility and transportation network topology. Available emergency resource reserve points include emergency warehouses, medical rescue stations, material transfer stations, and temporary allocation nodes. The real-time heterogeneous resource inventory represents the current stock status of different types of emergency resources, including medical, life support, engineering rescue, and communication support resources. Subsequently, the real-time heterogeneous resource inventory is classified into first-level and second-level support resources. First-level support resources are used to meet life-saving and emergency medical needs and have priority dispatch attributes, while second-level support resources are used to support post-disaster recovery and auxiliary support needs. The classification process is achieved by setting resource importance weight thresholds, that is, classifying and mapping inventory resources according to their impact on life safety and emergency response level.

[0038] After classification, a time-series state matrix for the resource supply side is constructed. This matrix is ​​used to describe the changes in resource inventory at different time points of each resource reserve point. The construction method is as follows: the inventory data of each reserve point is arranged according to the time series with the resource type as the row dimension and the time sampling point as the column dimension. The resource consumption rate and replenishment rate are introduced as dynamic correction factors, thereby forming a resource state expression structure with time-series evolution characteristics.

[0039] The real-time transportation time characteristics between each available emergency resource reserve point and each rescue point are calculated using an asymmetric spatial distance metric model. An initial scheduling objective function is constructed with minimizing the resource scheduling response time as the boundary condition for emergency resource scheduling. An initial single-step emergency resource scheduling strategy is obtained by using a swarm optimization algorithm in combination with the time-series state matrix of the resource supply side and the real-time transportation time characteristics.

[0040] After obtaining the time-series state matrix of the resource supply side, an asymmetric spatial distance metric model is used to calculate the real-time transportation time characteristics between each available emergency resource reserve point and each rescue point. This asymmetric spatial distance metric model transforms traditional spatial distance into a directed distance representation centered on time cost by considering the impact of road directionality, congestion differences, and traffic restrictions on transportation time. The specific calculation process is as follows: First, a directed graph structure is constructed based on road condition data, where nodes represent geographical locations and edges represent passable roads. Second, the one-way travel time cost is calculated for each edge, which is determined by road length, real-time travel speed, congestion coefficient, and blockage status. Finally, a shortest path search algorithm is used to calculate the minimum time cost path between each resource reserve point and each rescue point in the directed graph, thus obtaining the real-time transportation time characteristic matrix.

[0041] When constructing the initial scheduling objective function, modeling is performed with minimizing the resource scheduling response time as the boundary condition. The resource scheduling response time is defined as the total time elapsed from the dispatch of resources from the resource reserve point to the completion of resource delivery at the rescue point. The specific steps are as follows: First, the resource demand of each rescue point is used as the constraint condition, and the inventory capacity of each resource reserve point is used as the supply constraint condition. Then, the transportation time characteristics are used as the core time variable, and the resource allocation decision variable is introduced into the objective function. Finally, a mathematical model is constructed with the minimum total response time as the optimization objective. This model simultaneously satisfies the resource supply and demand balance constraint, the reserve capacity constraint, and the transportation path reachability constraint, thus forming the initial scheduling objective function for subsequent optimization calculations.

[0042] When using swarm optimization algorithms to solve the initial single-step emergency resource scheduling strategy, an iterative search mechanism based on swarm intelligence is employed for computation. The swarm optimization algorithm can be an improved form of particle swarm optimization or ant colony optimization; the specific algorithm chosen is determined by those skilled in the art based on the actual situation. The core idea is to approximate the global optimum through the collaborative search of multiple candidate solutions. The specific process is as follows: First, multiple scheduling candidate solutions are initialized, each representing a resource allocation scheme. Second, the fitness value of each candidate solution is calculated based on the time-series state matrix of the resource supply side and the real-time transportation time characteristics. This fitness value is composed of the degree of resource demand satisfaction and the response time. Then, the search direction and position of the candidate solutions are updated based on the fitness value, causing them to iteratively move towards a better solution region. Finally, after multiple rounds of iteration and convergence, the optimal candidate solution is output. This optimal solution is the initial single-step emergency resource scheduling strategy, used to guide resource allocation and transportation path selection in the first resource scheduling decision step.

[0043] Step S3: Output the single-step emergency resource scheduling strategy; in the subsequent resource scheduling decision steps, update the corresponding updated emergency resource scheduling boundary conditions based on the corresponding spatiotemporal urgency characteristics of the disaster in each rescue point and the real-time acquired dynamic demand vector of disaster emergency resources; perform scheduling calculations based on the corresponding updated emergency resource scheduling boundary conditions to obtain the corresponding single-step emergency resource scheduling strategy. The specific steps for updating the disaster response based on the spatiotemporal urgency characteristics of each rescue point and the real-time dynamic demand vector of disaster emergency resources include: Within the current resource scheduling decision step, calculate the remaining heterogeneous resource inventory of each available emergency resource reserve point to obtain the remaining emergency resource constraint; extract the emergency resource consumption of each rescue point in the previous resource scheduling decision step to obtain the remaining emergency resource quantity of each rescue point; First, within the current resource scheduling decision step, calculate the remaining heterogeneous resource inventory of each available emergency resource reserve point. The remaining heterogeneous resource inventory represents the amount of resources that each reserve point can still allocate after the previous round of resource scheduling is completed. It includes multiple types of resources such as medical resources, life support resources, and engineering emergency rescue resources.

[0044] The specific calculation process is as follows: Based on the initial inventory, subtract the actual dispatched quantity and the amount of resources in transit that have not yet been replenished from the previous decision step, while also considering the resource replenishment input, to obtain a dynamically updated inventory status. On this basis, a remaining emergency resource constraint is formed to limit the maximum boundary conditions for resource allocation in the current decision step. Simultaneously, the emergency resource consumption of each rescue point in the previous resource dispatch decision step is extracted and compared with the current remaining resource status to obtain the remaining emergency resources of each rescue point. This remaining amount reflects the survival capacity of each rescue point after resource consumption at the current stage of disaster development. It is calculated by accumulating the difference between the historical consumption sequence and the current demand trend to obtain the characteristics of resource gap changes.

[0045] Each rescue point is treated as an emergency resource game unit, and an emergency resource scheduling characteristic function is constructed with the goal of minimizing the total global disaster loss. The Shapley value calculation algorithm is used to traverse several subsets of emergency resource game units. Based on the emergency resource scheduling characteristic function, the marginal contribution value of the emergency resource game unit is obtained. After updating the resource status, each rescue point is modeled as an emergency resource game unit. This unit refers to the smallest decision-making entity participating in resource competition and collaborative allocation under limited resource conditions. Each rescue point is considered an independent strategy participant in this model. Subsequently, an emergency resource scheduling feature function is constructed with the objective of minimizing the total global disaster loss. This feature function describes the overall disaster loss change after resource allocation for any combination of rescue points. Its construction method is as follows: First, a loss function is defined, which comprehensively considers the amount of unmet resource demand, rescue delay time, and the intensity of disaster spread. Then, the corresponding loss change value is calculated for different combinations of rescue points, thus forming a set mapping relationship, i.e., different resource allocation subsets correspond to different global loss results.

[0046] When calculating the marginal contribution value of resources, the Shapley value calculation algorithm is used to perform traversal analysis on the emergency resource game units. The Shapley value is used to measure the marginal contribution of a single rescue point in different cooperative combinations. The specific calculation steps are as follows: First, all possible subset combinations of rescue points are generated, and the order in which each rescue point is added to different subsets is determined. Then, the difference in global loss before and after adding the rescue point in each subset is calculated. After that, the difference results of all subsets are weighted and averaged, and the weights are determined by the subset size and the combination probability. Finally, the marginal contribution value of resources corresponding to each rescue point is obtained, which is used to characterize its importance and influence in the overall resource scheduling system.

[0047] Based on the marginal contribution value of resources, a rescue game weight vector is generated for each rescue point; the spatiotemporal urgency characteristics of the disaster are weighted and corrected based on the rescue game weight vector for each rescue point to obtain the corrected spatiotemporal urgency characteristics of the disaster. After obtaining the marginal contribution value of resources, a rescue game weight vector for each rescue point is generated based on this contribution value. The weight vector is used to quantify the priority of different rescue points in resource allocation decisions. The specific generation method is as follows: the marginal contribution value of each rescue point is normalized and then weighted and fused with the spatiotemporal urgency characteristics of the disaster, so that the contribution value not only reflects the degree of game influence, but also reflects the emergency state of the disaster. Finally, a multi-dimensional weight vector structure is formed for subsequent feature correction. Subsequently, the rescue game weight vector is used to weight and correct the spatiotemporal urgency characteristics of each rescue point. The correction process is to multiply the original urgency feature vector with the weight vector dimension by dimension and perform nonlinear mapping to obtain the corrected spatiotemporal urgency characteristics of the disaster, so that it simultaneously reflects the urgency of the disaster and the relationship between resource competition.

[0048] Based on the modified spatiotemporal urgency characteristics of disaster and the dynamic demand vector of disaster emergency resources, the boundary conditions for emergency resource scheduling are updated in multiple dimensions to obtain the corresponding updated boundary conditions for emergency resource scheduling. Among them, the multi-dimensional update includes the update of resource allocation upper limit, the update of time window constraint, and the update of path reachability constraint.

[0049] After feature correction, the original emergency resource scheduling boundary conditions are updated in multiple dimensions based on the corrected spatiotemporal urgency characteristics of the disaster and the dynamic demand vector of emergency resources. These multi-dimensional updates include resource allocation cap updates, time window constraint updates, and path reachability constraint updates. Specifically, the resource allocation cap update adjusts the maximum available resource amount per instance based on the current remaining emergency resource constraints to avoid exceeding reserve capacity. The time window constraint update recalculates the acceptable delay time range for each rescue point based on the corrected urgency characteristics to reflect changes in the disaster situation. The path reachability constraint update combines the latest road condition information and asymmetric spatial distance changes to eliminate or reweight previously inaccessible or high-risk paths. Through this multi-dimensional update process, updated emergency resource scheduling boundary conditions are formed to adapt to changes in the current disaster situation, providing a dynamic constraint basis for subsequent resource scheduling decision-making steps.

[0050] The specific steps for obtaining the corresponding single-step emergency resource scheduling strategy are as follows: scheduling calculations are performed based on the updated emergency resource scheduling boundary conditions, including: using a swarm optimization algorithm in conjunction with the updated emergency resource scheduling boundary conditions to obtain the corresponding single-step emergency resource scheduling strategy.

[0051] In this embodiment, when performing single-step emergency resource scheduling calculations based on the updated emergency resource scheduling boundary conditions, the boundary conditions are first subjected to structured analysis and constraint mapping. The updated emergency resource scheduling boundary conditions include three types of information: resource allocation upper limit constraints, time window constraints, and path reachability constraints. Specifically, the resource allocation upper limit constraints are transformed into the maximum quantity limit that various heterogeneous resources can be scheduled within a single decision step; the time window constraints are transformed into the time boundary that each rescue point must complete resource delivery within a limited time interval; and the path reachability constraints are transformed into a set of feasible transportation paths and their corresponding travel cost weights. The above constraints are uniformly mapped into a constraint space in the scheduling optimization problem, so that resource allocation decisions must be performed within this constraint space for feasibility search, thereby forming the feasible solution domain of the current decision step.

[0052] After constructing the constraint space, a swarm optimization algorithm is introduced for scheduling computation. This algorithm simulates the collaborative search behavior of a swarm to approximate the global optimum. Its core idea is to find the optimal resource allocation strategy through iterative evolution of multiple candidate scheduling schemes. The specific process is as follows: First, multiple initial scheduling schemes are randomly generated within the feasible solution domain. Each scheme represents a combination of resource allocation and transportation from resource reserve points to various rescue points. Then, a fitness evaluation is performed on each scheme. The fitness function consists of the degree of resource demand satisfaction, total transportation time, and constraint violation penalty, where transportation time is calculated using an asymmetric spatial distance model. Next, candidate scheduling schemes are ranked according to their fitness values, and the scheme parameters are updated using an information sharing mechanism to make better schemes converge towards a better solution. Simultaneously, poorer schemes are perturbed and updated to enhance search diversity. During the iteration process, the fitness evaluation and scheme update steps are repeated continuously, gradually narrowing the search range and approaching the optimal solution.

[0053] After the algorithm converges, the scheduling scheme with the optimal fitness function value is selected as the current resource allocation result and defined as a single-step emergency resource scheduling strategy. This single-step emergency resource scheduling strategy specifically represents the resource allocation quantity, resource type, and corresponding transportation path combination relationship from each emergency resource reserve point to each rescue point within the current resource scheduling decision step, and the strategy satisfies all updated scheduling boundary condition constraints. At the same time, the output of this strategy will be used as the initial state input for the next resource scheduling decision step to achieve dynamic connection and iterative optimization of the subsequent continuous scheduling process.

[0054] Step S4: Execute all resource scheduling decision steps until the dynamic demand vector of disaster emergency resources in the disaster-stricken area meets the preset disaster resource safety demand range.

[0055] In this embodiment, step S4, executing all resource scheduling decision steps, refers to a full-process control mechanism that calculates, executes, and updates the status of each resource scheduling decision step in sequence according to a pre-built continuous resource scheduling framework. Each decision step corresponds to a complete closed-loop processing flow of demand analysis, resource allocation, transportation execution, and status feedback.

[0056] In the specific execution process, the input state of the current decision step is first read, including the dynamic demand vector of disaster emergency resources, the remaining resource status of each rescue point, and the updated scheduling boundary conditions. Based on this, the corresponding single-step emergency resource scheduling strategy is called to generate the result, so that resources are allocated and scheduled from the reserve point to each rescue point. Then, the execution result is fed back and calculated, that is, the remaining amount of resources at each rescue point is updated according to the actual resource arrival and consumption, and the time-series state matrix of the resource supply side is simultaneously corrected, so that the system state evolves dynamically with the scheduling process.

[0057] After completing a single resource scheduling decision step, the dynamic demand vector for disaster emergency resources is updated in real time. The update process is driven by both resource consumption feedback and disaster evolution prediction results. The specific method is as follows: First, the resource gap changes at each rescue point are calculated based on the actual resource allocation in the current decision step. Second, the future demand is forward-corrected by combining the disaster change trend output by the spatiotemporal time series prediction model. Finally, the actual consumption changes and predicted changes are fused and calculated to form a new dynamic demand vector for disaster emergency resources, which can reflect the latest disaster status and changes in resource supply and demand.

[0058] During the iterative execution of all resource scheduling decision steps, a convergence determination mechanism is introduced to determine whether the scheduling process has terminated. This mechanism is based on the disaster-affected resource safety demand range, which refers to the minimum threshold set of resources required to ensure basic life rescue and operational maintenance within the disaster-affected area. This includes medical resources, safe drinking water, basic food, and critical engineering support resources. When the dynamic demand vector of disaster-affected emergency resources at all rescue points decreases to within this preset safety threshold range, it indicates that the supply and demand of resources in the disaster area has reached a stable state, and the impact of the disaster has been controlled within an acceptable range. At this point, the scheduling process is considered complete. If this condition is not met, the process continues to the next resource scheduling decision step, repeating the scheduling calculation and state update process until the entire system converges to the preset safety demand range, thus completing the entire emergency resource scheduling process.

[0059] In this embodiment, the preset disaster-affected resource safety demand range refers to the resource demand threshold range used in disaster emergency management to determine whether a disaster-affected area has reached a basic safety guarantee state. Its setting process is based on a comprehensive assessment of historical disaster relief data, regional population structure characteristics, and minimum guarantee standards for different disaster types. Specifically, the setting method is as follows: First, data on the resource consumption completion status of each rescue point in historical disaster events of the same type are collected, including indicators such as medical treatment completion rate, personnel transfer completion rate, and basic material coverage rate. Second, combined with the population density of the disaster-affected area, the proportion of vulnerable populations, and the infrastructure recovery capacity, the minimum guarantee demand for different types of resources is hierarchically modeled to form basic safety thresholds for medical, living security, and engineering support resources. Then, statistical learning methods are used to cluster the resource consumption termination points in historical data, extracting the resource remaining distribution characteristics when the post-disaster system tends to stabilize, and using this as an empirical reference benchmark. Finally, the empirical benchmark is fused and corrected with the matching coefficient of the current disaster type to form a dynamically adjustable resource safety demand range, which can reflect historical rescue experience and adapt to the differences in current disaster scenarios, thus serving as an important constraint condition for judging whether the emergency resource dispatch process has converged.

[0060] In this embodiment, all calculations and simulations are implemented based on a unified computational framework combining spatiotemporal graph modeling and continuous optimization. Spatial relationships are expressed using a graph structure model, temporal evolution uses a temporal recursive model, and resource scheduling is solved using a swarm intelligence optimization algorithm, thus forming a closed-loop computational system of spatiotemporal modeling, state prediction, constraint update, and optimization decision-making. Specifically, the spatiotemporal grid units and rescue points after the disaster-stricken area is divided are constructed as graph structure nodes, and road accessibility, disaster spread, and resource allocation relationships are constructed as graph structure edges. Asymmetric time costs are introduced in the form of edge weights to form a spatiotemporal correlation graph. Subsequently, the node attributes are vectorized, and the cumulative disaster intensity, resource demand, road condition, and population impact are uniformly encoded into multi-dimensional feature vectors. In the time dimension, a recursive update mechanism is adopted, using the output state of each resource scheduling decision step as the input of the next step to realize the dynamic evolution simulation of disaster and resource status.

[0061] At the model solution level, a swarm intelligence optimization algorithm is introduced as the core computational method to achieve global search by constructing a set of candidate scheduling solutions. The specific steps are as follows: First, an initial scheduling solution is randomly generated within the feasible region that satisfies resource constraints and path constraints. Each solution represents a combination of a resource allocation matrix and a transportation path. Second, a unified fitness function is defined, which consists of the deviation in meeting resource requirements, transportation time cost, and a penalty term for constraint violation. Then, the fitness of each candidate solution is evaluated and the solution is sorted and updated. The solution space distribution is gradually optimized through iteration. Finally, the scheduling scheme that satisfies the optimal objective function is output as a single-step emergency resource scheduling strategy.

[0062] To address the issue of inconsistent data units, a unified dimensional standardization mechanism is introduced before the data enters the model for computation. The specific steps are as follows: First, all input data are categorized into spatial data (e.g., distance, area), temporal data (e.g., transportation time, duration), and resource data (e.g., quantity of materials, inventory). Then, different standardization methods are applied to each category. Spatial and temporal data are processed using range normalization to map the original values ​​to a unified interval, eliminating dimensional differences. Resource data is standardized using a proportional coefficient-based method to convert different types of resources into equivalent resource units. For data with non-linear relationships, logarithmic transformation or exponential scaling is used for smoothing to avoid the impact of extreme values ​​on model stability. Finally, all standardized data are mapped to a dimensionless feature space, enabling joint computation within the same optimization model.

[0063] After standardization, the overall calculation process is executed step by step according to the time sequence: First, the spatiotemporal data of the disaster are input and the graph structure is constructed; second, spatiotemporal features are extracted and the disaster evolution is predicted; then, the resource demand vector and constraints are updated according to the prediction results; next, the swarm optimization algorithm is called to solve the resource allocation in each resource scheduling decision step; finally, the system variables are updated through the state feedback mechanism and the loop enters the next decision step until the preset termination condition is met, thus completing the complete emergency resource scheduling simulation and calculation process.

[0064] In this embodiment, the selection of various preset values, weighted weights, prediction models, and swarm optimization algorithms are all based on a comprehensive setting principle that combines historical data statistical characteristics with disaster type adaptation and online adaptive correction. The preset values ​​are determined by first extracting key statistical distribution characteristics from historical disaster emergency response data, such as resource consumption completion thresholds, rescue response time intervals, and disaster spread speed ranges. Then, layered corrections are made based on the evolutionary characteristics of different disaster types (such as floods, earthquakes, and fires), ultimately forming an empirical parameter range that can serve as initialization constraints. During actual operation, these parameters are dynamically fine-tuned based on real-time feedback to avoid the fixed thresholds becoming disconnected from the actual disaster situation.

[0065] The weight parameters in the weighted algorithm are set by combining indicator importance analysis with data-driven calibration. First, based on the disaster emergency goals, various indicators are functionally decomposed, such as resource demand urgency, spatial diffusion intensity, and transportation time cost, which correspond to different decision-making impact dimensions. Then, the sensitivity of each indicator to the final scheduling result is calculated by backtracking through historical cases, and the sensitivity is used as the basis for initial weight allocation. During the operation, an error feedback mechanism is combined to recursively correct the weights, so that the weights are automatically adjusted as the disaster situation evolves, thereby maintaining the model's adaptability to different scenarios.

[0066] In selecting the preset model, the spatiotemporal prediction model prioritizes model systems with graph structure representation capabilities and time-dependent modeling capabilities. By matching and analyzing the spatial correlation and temporal continuity characteristics of disaster propagation, a modeling method combining graph structure learning and temporal recursion is selected. The model complexity is configured in stages according to the data scale and computing resource conditions. When the data volume is large, a deep graph feature extraction structure is used, and when the real-time requirements are high, a lightweight recursive structure is used. Online training or incremental update mechanisms are used to ensure the model's adaptability to the latest disaster data.

[0067] The selection of swarm optimization algorithms is determined based on the constraint complexity and search space characteristics of the scheduling problem. When the resource scheduling problem exhibits high-dimensionality, multiple constraints, and nonlinear characteristics, swarm intelligence algorithm systems with global search capabilities are preferred, such as the improved particle swarm optimization method based on the swarm cooperative search mechanism or the ant colony optimization method based on the pheromone update mechanism. Furthermore, the convergence ability of the algorithm in complex constraint spaces is improved by introducing constraint repair strategies and dynamic inertia adjustment mechanisms. At the same time, according to real-time computing requirements, the algorithm parameters can be adaptively adjusted in different decision steps to achieve a balance between global search capability and computational efficiency, thereby meeting the real-time requirements of emergency resource scheduling.

[0068] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. An intelligent scheduling and optimization method for emergency resources based on spatiotemporal big data, characterized in that, Includes the following steps: Step S1: Disaster spatiotemporal data analysis; real-time acquisition of disaster spatiotemporal data of the disaster-stricken area; wherein, the disaster spatiotemporal data includes disaster geographic information data, road condition data of the disaster-stricken area, and disaster evolution spatiotemporal data; based on the disaster spatiotemporal data analysis, obtain the dynamic demand vector of disaster emergency resources for each rescue point in the disaster-stricken area. Step S2: Construct an emergency resource scheduling framework; the emergency resource scheduling framework contains several consecutive resource scheduling decision steps; for the first resource scheduling decision step, analyze the dynamic demand vector of disaster emergency resources to obtain the spatiotemporal urgency characteristics of each rescue point in the disaster area; perform initial scheduling calculation with minimizing the resource scheduling response time as the boundary condition for emergency resource scheduling to obtain the initial single-step emergency resource scheduling strategy. Step S3: Output the single-step emergency resource scheduling strategy; in the subsequent resource scheduling decision steps, update the corresponding updated emergency resource scheduling boundary conditions based on the corresponding spatiotemporal urgency characteristics of the disaster in each rescue point and the real-time acquired dynamic demand vector of disaster emergency resources; perform scheduling calculations based on the corresponding updated emergency resource scheduling boundary conditions to obtain the corresponding single-step emergency resource scheduling strategy. Step S4: Execute all resource scheduling decision steps until the dynamic demand vector of disaster emergency resources in the disaster-stricken area meets the preset disaster resource safety demand range.

2. The emergency resource intelligent scheduling and optimization method based on spatiotemporal big data according to claim 1, characterized in that, The specific steps for analyzing disaster-related spatiotemporal data include: Based on disaster-affected geographic information data, an adaptive grid partitioning algorithm is used to map the disaster-affected area into several disaster-affected spatiotemporal grid units; the cumulative disaster intensity of each disaster-affected spatiotemporal grid unit is calculated by combining disaster evolution spatiotemporal data; and a disaster spatiotemporal feature map is constructed based on the cumulative disaster intensity. Based on road condition data of the disaster-stricken area, an asymmetric spatial distance measurement model with time cost as the core is established; a spatiotemporal clustering algorithm is used in combination with the asymmetric spatial distance measurement model to cluster several spatiotemporal grid units of the disaster-stricken area to obtain several rescue points. The spatiotemporal feature map of the disaster area is extrapolated using a pre-set spatiotemporal prediction model to obtain the disaster evolution trend prediction results for each rescue point. The disaster evolution trend prediction results include the disaster evolution trend characteristics and potential secondary disaster risk points. Based on the disaster evolution trend prediction results, feature vectorization processing is performed to obtain the dynamic demand vector of disaster emergency resources for each rescue point in the disaster area.

3. The emergency resource intelligent scheduling and optimization method based on spatiotemporal big data according to claim 2, characterized in that, The specific steps for analyzing the dynamic demand vector of disaster emergency resources to obtain the spatiotemporal urgency characteristics of each rescue point within the disaster-stricken area include: For any rescue point, the rate of change of the dynamic demand vector of disaster emergency resources in the time dimension is extracted to obtain the resource demand growth gradient characteristics; combined with the disaster evolution trend characteristics and resource demand growth gradient characteristics between adjacent rescue points, the spatial diffusion intensity of the disaster is calculated. The model uses an asymmetric spatial distance metric to analyze the golden time window for each rescue point within the first resource scheduling decision step; it calculates the resource transportation time redundancy of each rescue point based on road condition data of the disaster area and the golden time window for rescue; and it calculates resources based on the spatial diffusion intensity of the disaster and the resource transportation time redundancy to obtain the spatiotemporal urgency characteristics of the disaster.

4. The emergency resource intelligent scheduling and optimization method based on spatiotemporal big data according to claim 3, characterized in that, The specific steps for initial scheduling calculations using minimizing resource scheduling response time as the boundary condition for emergency resource scheduling include: Obtain the location distribution of all available emergency resource reserve points within the radiation range of the disaster area and the real-time inventory of heterogeneous resources in each available emergency resource reserve point; classify the real-time inventory of heterogeneous resources to obtain the first type of guarantee resources and the second type of guarantee resources; construct the time-series state matrix of resource supply side; The real-time transportation time characteristics between each available emergency resource reserve point and each rescue point are calculated using an asymmetric spatial distance metric model. An initial scheduling objective function is constructed with minimizing the resource scheduling response time as the boundary condition for emergency resource scheduling. An initial single-step emergency resource scheduling strategy is obtained by using a swarm optimization algorithm in combination with the time-series state matrix of the resource supply side and the real-time transportation time characteristics.

5. The emergency resource intelligent scheduling and optimization method based on spatiotemporal big data according to claim 4, characterized in that, The specific steps for updating the disaster response based on the spatiotemporal urgency characteristics of each rescue point and the real-time dynamic demand vector of disaster emergency resources include: Within the current resource scheduling decision step, calculate the remaining heterogeneous resource inventory of each available emergency resource reserve point to obtain the remaining emergency resource constraint; extract the emergency resource consumption of each rescue point in the previous resource scheduling decision step to obtain the remaining emergency resource quantity of each rescue point; Each rescue point is treated as an emergency resource game unit, and an emergency resource scheduling characteristic function is constructed with the goal of minimizing the total global disaster loss. The Shapley value calculation algorithm is used to traverse several subsets of emergency resource game units. Based on the emergency resource scheduling characteristic function, the marginal contribution value of the emergency resource game unit is obtained. Based on the marginal contribution value of resources, a rescue game weight vector is generated for each rescue point; the spatiotemporal urgency characteristics of the disaster are weighted and corrected based on the rescue game weight vector for each rescue point to obtain the corrected spatiotemporal urgency characteristics of the disaster. Based on the modified spatiotemporal urgency characteristics of disaster and the dynamic demand vector of disaster emergency resources, the boundary conditions for emergency resource scheduling are updated in multiple dimensions to obtain the corresponding updated boundary conditions for emergency resource scheduling. Among them, the multi-dimensional update includes the update of resource allocation upper limit, the update of time window constraint, and the update of path reachability constraint.

6. The emergency resource intelligent scheduling and optimization method based on spatiotemporal big data according to claim 5, characterized in that, The specific steps for obtaining the corresponding single-step emergency resource scheduling strategy are as follows: scheduling calculations are performed based on the updated emergency resource scheduling boundary conditions, including: using a swarm optimization algorithm in conjunction with the updated emergency resource scheduling boundary conditions to obtain the corresponding single-step emergency resource scheduling strategy.

7. The emergency resource intelligent scheduling and optimization method based on spatiotemporal big data according to claim 6, characterized in that, The pre-defined spatiotemporal time series prediction model is constructed based on a graph convolutional network model.

Images