A fire risk assessment method for edge cities and related equipment

Abstract

The application discloses an edge city-oriented fire risk assessment method and related equipment, which comprises the following steps: based on the dynamic graph model of the target city, the static risk assessment data and the traffic situation data of the target city at the query time, the node feature sequence of each grid node in the dynamic graph model is obtained; based on the preset recursive deduction model, the node feature sequence of each grid node is subjected to multi-frequency decomposition, and the fire space-time risk of each grid node is analyzed according to the multi-frequency decomposition result of each grid node, so as to obtain the comprehensive risk representation of each grid node; based on the comprehensive risk representation of each grid node, the dynamic fire risk assessment and the fire accessibility assessment of each grid node are carried out, and the coupling analysis of the assessment results is carried out, so as to determine the fire risk assessment result of the target city. The method can more comprehensively and accurately realize the fire risk assessment of the edge city, and can be widely applied to the technical field of city fire risk assessment.

Description

Technical Field

[0001] This application relates to the field of urban fire risk assessment technology, and in particular to a fire risk assessment method and related equipment for peripheral urban areas. Background Technology

[0002] In urban spatial structure research, the "urban fringe" refers to the transitional zone from the periphery of the urban built-up area to the rural hinterland, exhibiting dynamic mixed characteristics of urban and rural land use. Its scope starts from the periphery of the urban core area and extends to the urban administrative boundary or natural geographical barrier.

[0003] Against the backdrop of accelerated urban expansion and integrated urban-rural development, the status of peripheral urban areas in development planning is gradually rising. They are shifting from being backup sites for urban construction to developing autonomously, with expanded industrial layouts and significantly increased land development intensity in some areas, thus elevating their position in overall urban planning. However, this rise in status and increased development intensity is accompanied by a sharp increase in fire risk, while the corresponding fire protection infrastructure capacity has not kept pace. Shortcomings such as low fire station density, excessive service radius, and fragile road network structure are prominent, leading to an increasingly prominent structural contradiction of "high risk - low protection." Fire risk assessment is a crucial component of fire protection development planning in peripheral urban areas. While fire risk assessments in urban areas often rely solely on static road network analysis, this approach is insufficient in capturing the real-time evolution of traffic flow and road network connectivity in peripheral urban areas, and it cannot decouple the long-term accumulation and short-term outbreak of risks. Its limitations are evident when dealing with the complex and rapidly changing structures of peripheral urban areas. Therefore, the accuracy of existing technologies for fire risk assessment in peripheral urban areas needs improvement.

[0004] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0005] The main objective of this application is to propose a fire risk assessment method and related equipment for peripheral urban areas. This method can combine the static risk characteristics of the urban area with dynamic traffic situation data to construct a deductive model input, and then use the frequency components of different frequency levels obtained by multi-frequency decomposition of the model to obtain the comprehensive risk characterization of each node in the dynamic map of the urban area. Finally, it combines the dynamic risk assessment of the node with the fire accessibility assessment results for coupling analysis, thereby achieving a more comprehensive and accurate fire risk assessment of peripheral urban areas.

[0006] To achieve the above objectives, one aspect of this application proposes a fire risk assessment method for peripheral urban areas, the method comprising:

[0007] A dynamic graph model of the target urban area is constructed based on the road network of the target urban area; the dynamic graph model includes several grid nodes and interaction edges between grid nodes; the types of grid nodes include demand points and fire stations; The static risk assessment data of the target urban area and its traffic situation data at the query time are obtained. The static risk assessment data includes data on disaster-causing factors, exposure factors, and disaster mitigation factors related to fire risk. The traffic situation data is obtained based on a preset window. The preset window records historical traffic interaction events between grid nodes within a preset time capacity before the query time based on a time elimination mechanism. The node feature sequence of each grid node is obtained by calculating based on the dynamic graph model of the target urban area, the static risk assessment data, and the traffic situation data at the query time. Based on a pre-defined recursive deduction model, the node feature sequence of each grid node is decomposed using multiple frequencies. The spatiotemporal fire risk of each grid node is analyzed based on the multiple frequency decomposition results to obtain a comprehensive risk characterization of each grid node. The multiple frequency decomposition results include frequency components at several different frequency levels. Dynamic fire risk assessment and fire accessibility assessment are performed on each grid node based on its comprehensive risk characterization. A coupling analysis is then conducted based on the results of the dynamic fire risk assessment and fire accessibility assessment of each grid node. Finally, the fire risk assessment result of the target urban area is determined based on the coupling analysis results of each grid node.

[0008] In some embodiments, the node feature sequence of the target grid node among the plurality of grid nodes is obtained in the following manner: The static risk assessment data of the target urban area is standardized, and the standardized data results are mapped to several grid nodes. The static feature vector of each grid node is obtained based on the mapping results. Based on the dynamic graph model of the target urban area and its traffic situation data at the query time, the neighbor node sequence and interaction edge sequence of each grid node are determined; the neighbor node sequence includes several neighbor nodes that have historical traffic interaction events with the grid node; the interaction edge sequence indicates the interaction edges between the grid node and its neighbor nodes; the attributes of the interaction edges are determined by the traffic path planning between the nodes. Construct several node pairs corresponding to the target grid node; the node pairs indicate the correspondence between the target grid node and the fire stations in the several grid nodes; analyze the neighboring node sequence of the target grid node and its corresponding fire station in each node pair to obtain a multidimensional statistical vector of each node pair; encode and fuse the multidimensional statistical vectors of the several node pairs to obtain the structural feature sequence of the target grid node; The original feature sequence of the target grid node is constructed based on the static feature vectors of the neighboring nodes in the neighboring node sequence and the attributes of the interactive edges in the interactive edge sequence. Based on the structural feature sequence of the target grid node and the original feature sequence, feature fusion processing is performed to obtain the node feature sequence of the target grid node.

[0009] In some embodiments, the analysis of the neighboring node sequence of the target grid node and its corresponding fire station in each node pair to obtain a multidimensional statistical vector for each node pair includes: The first feature is obtained by counting the total number of neighboring nodes in the neighboring node sequence of the target grid node; The second feature is obtained by statistically analyzing the frequency of occurrence of each neighboring node in the neighboring node sequence of the target grid node in the neighboring node sequence of the corresponding fire station. The third feature is obtained by counting the number of intersection nodes between the neighboring node sequences of the target grid node and the neighboring node sequences of the corresponding fire station. The frequency of occurrence of the corresponding fire station in the neighboring node sequence of the target grid node is counted to obtain the fourth feature; the frequency of occurrence of the target grid node in the neighboring node sequence of the corresponding fire station is counted to obtain the fifth feature; the product of the fourth feature and the fifth feature is calculated to obtain the sixth feature. A multidimensional statistical vector of the node pair is constructed based on the first feature, the second feature, the third feature, the fourth feature, the fifth feature, and the sixth feature.

[0010] In some embodiments, the comprehensive risk characterization of the target grid node among the plurality of grid nodes is obtained in the following manner: Calculate the time difference between the time of the historical traffic interaction event between the target grid node and its neighboring nodes and the query time, and map the calculation result through sinusoidal position coding to obtain the time interval sequence of the target grid node; By performing multi-frequency decomposition based on the node feature sequence and time interval sequence of the target grid node through a recursive residual network, several layers of frequency components indicating different levels of frequency risk and time feature components corresponding to each layer of frequency components are obtained. The discretization step size corresponding to the frequency component of each layer is generated based on the time feature component corresponding to the frequency component of each layer. Based on the preset equation, the frequency components of each layer and their corresponding discretization step size, the state evolution modeling of the frequency components of each layer is performed to obtain the state evolution modeling result of the frequency components of each layer; A bottom-up strategy is used to fuse the state evolution modeling results of several layers of frequency components to obtain a comprehensive risk characterization of the target grid node.

[0011] In some embodiments, the step of performing state evolution modeling on each layer of frequency components based on a preset equation, each layer of frequency components, and their corresponding discretization step size to obtain the state evolution modeling result of each layer of frequency components includes: The preset state transition matrix is ​​adjusted based on the discretization step size corresponding to each frequency component in each layer, and the state transition matrix corresponding to each frequency component in each layer is determined according to the adjustment result. The discretization input projection matrix corresponding to each frequency component in each layer is calculated based on the structural feature sequence of the target grid node, the preset state transition matrix, and the discretization step size corresponding to each frequency component in each layer. Based on the preset equations, the frequency components of each layer and their corresponding state transition matrices and discretized input projection matrices, the hidden state vectors corresponding to the frequency components of each layer are calculated. An output projection matrix is ​​constructed based on the structural feature sequence of the target grid nodes; the state evolution modeling result of each frequency component is obtained by calculating based on the output projection matrix and the hidden state vector corresponding to each frequency component.

[0012] In some embodiments, the fire accessibility assessment process for a target grid node among the plurality of grid nodes is implemented through the following steps: Construct several joint node pairs corresponding to the target grid node; the joint node pairs indicate the correspondence between the grid node and the corresponding fire stations in the several grid nodes; Based on the comprehensive risk characterization of the target grid nodes and their corresponding fire stations in each joint node pair, a joint risk characterization corresponding to each joint node pair is constructed. Each joint node is input into a preset fire accessibility prediction model for calculation, and the fire accessibility probability between the target grid node and each corresponding fire station is determined based on the calculation results. The fire accessibility assessment result of the target grid node is obtained by analyzing the fire accessibility probability between the target grid node and each corresponding fire station.

[0013] In some embodiments, the construction of a joint risk representation for each joint node pair based on the comprehensive risk representation of the target grid node and its corresponding fire station in each joint node pair includes: The comprehensive risk representation of the target grid nodes and their corresponding fire stations in the joint node pair is compressed based on the weighted aggregation mechanism to obtain the node embedding vector of the target grid nodes and their corresponding fire stations in the joint node pair. By concatenating the node embedding vectors of the target grid node and its corresponding fire station in the joint node pair, the joint risk representation corresponding to the joint node pair is obtained.

[0014] In some embodiments, the dynamic fire risk assessment result of the grid node includes a dynamic fire risk index obtained by mapping the comprehensive risk characterization of the grid node based on a preset multilayer sensor; the fire accessibility assessment result indicates the fire accessibility probability between the grid node and each corresponding fire station in the dynamic graph model; the coupling analysis based on the dynamic fire risk assessment and fire accessibility assessment results of each grid node includes: The fire accessibility probability between each corresponding fire station in the dynamic model and the grid node is analyzed to determine the fire accessibility probability between the corresponding fire station and the grid node that meets the preset requirements. The dynamic fire risk index of the grid node is judged according to the preset fire risk judgment threshold to obtain the first judgment result; The fire accessibility probability between the corresponding fire station that meets the preset requirements and the grid node is judged according to the preset accessibility judgment threshold to obtain a second judgment result; The coupling analysis results of the grid nodes are obtained by calculating based on the preset coupling discrimination function, the first discrimination result, and the second discrimination result.

[0015] To achieve the above objectives, another aspect of this application proposes a fire risk assessment system for peripheral urban areas, the system comprising: A dynamic graph construction module is used to construct a dynamic graph model of the target urban area based on the road network of the target urban area; the dynamic graph model includes several grid nodes and interaction edges between grid nodes; the types of grid nodes include demand points and fire stations; The data acquisition module is used to acquire static risk assessment data of the target urban area and its traffic situation data at the query time; the static risk assessment data includes data on disaster-causing factors, exposure factors, and disaster mitigation factors related to fire risk; the traffic situation data is acquired based on a preset window; the preset window records historical traffic interaction events between grid nodes within a preset time capacity before the query time based on a time elimination mechanism; and the node feature sequence of each grid node is obtained by calculation based on the dynamic graph model of the target urban area, the static risk assessment data, and the traffic situation data at the query time. The risk assessment module is used to perform multi-frequency decomposition on the node feature sequence of each grid node based on a preset recursive deduction model, and analyze the fire spatiotemporal risk of each grid node based on the multi-frequency decomposition results to obtain a comprehensive risk characterization of each grid node. The multi-frequency decomposition results include frequency components at several different frequency levels. Based on the comprehensive risk characterization of each grid node, dynamic fire risk assessment and fire accessibility assessment are performed on each grid node. Coupled analysis is performed based on the results of the dynamic fire risk assessment and fire accessibility assessment of each grid node, and the fire risk assessment result of the target urban area is determined based on the coupled analysis results of each grid node.

[0016] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0017] The embodiments of this application include at least the following beneficial effects: This application provides a method, system, and electronic device for fire risk assessment in peripheral urban areas. The scheme first constructs a dynamic graph model of the target urban area based on the road network of the target urban area. The model includes several grid nodes and interaction edges between grid nodes, serving as the basis for urban risk assessment. Then, it acquires static risk assessment data related to fire risk in the target urban area and its traffic situation data at the query time. Based on the dynamic graph model and the acquired data, it constructs a node feature sequence for each grid node, serving as input to a recursive deduction model. In this application's method, the traffic situation data is recorded through a preset window, based on a time elimination mechanism, recording historical traffic interaction events occurring within a preset time capacity before the query time. This data recording method allows for real-time updates. Traditional GIS or AHP models are mostly based on "snapshot" static data, while this application's method introduces real-time updated traffic situation data, improving the ability to capture spatiotemporal entanglement. That is, it simultaneously considers static road network data, risk characteristic data, and dynamic traffic situation data, thereby effectively improving the comprehensiveness and accuracy of urban fire risk assessment. Furthermore, the recursive deduction model... The method performs multi-frequency decomposition on the input data and analyzes the frequency components at different frequency levels to determine the comprehensive risk characterization of grid nodes. By analyzing the frequency components at different frequency levels, the model can analyze the evolution of risks at different frequency levels, thus more accurately determining the comprehensive risk characterization of grid nodes. Finally, dynamic fire risk assessment and fire accessibility assessment of the nodes are performed based on the comprehensive risk characterization of the grid nodes. Coupled analysis is then performed based on the two assessment results to obtain the fire risk assessment results of the target urban area. By coupling the dynamic fire risk assessment results and fire accessibility assessment results of the nodes, the method can analyze the fire risk of grid nodes based on the correspondence between different assessment results, thereby improving the effectiveness of the fire risk assessment results of the grid nodes. In addition, compared with the central urban area, the fire protection and transportation infrastructure construction level of peripheral urban areas may be lower. Therefore, introducing real-time traffic interaction data into the fire risk assessment process of peripheral urban areas and considering the correspondence between the assessment results of dynamic fire risk and fire accessibility can effectively improve the effectiveness of the fire risk assessment results of peripheral urban areas. Attached Figure Description

[0018] Figure 1 This is a flowchart of a fire risk assessment method for peripheral urban areas provided in an embodiment of this application; Figure 2 This is a schematic diagram illustrating the result of identifying the edge urban areas of a city, as provided in an embodiment of this application. Figure 3A This is a schematic diagram illustrating the result of identifying the density of points of interest in a city, as provided in an embodiment of this application. Figure 3B This is a schematic diagram showing the result of dividing a city into kernel density contour lines, as provided in an embodiment of this application. Figure 3C This is a schematic diagram illustrating the results of identifying the outskirts of a city, as provided in an embodiment of this application. Figure 4 This is a schematic diagram of the distribution of urban fire stations provided in an embodiment of this application; Figure 5 This is a flowchart of the steps of the urban fire risk assessment and accessibility prediction method based on dynamic graph learning provided in the embodiments of this application; Figure 6 This is a process architecture diagram of the urban fire risk assessment and accessibility prediction method based on dynamic graph learning provided in the embodiments of this application; Figure 7 This is a schematic diagram of the structure of the continuous-time hierarchical state-space model provided in the embodiments of this application; Figure 8 This is a schematic diagram of the structure of the continuous-time state-space model provided in the embodiments of this application; Figure 9 This is a dynamic fire risk heat map of a certain urban area provided in an embodiment of this application; Figure 10A This is a schematic diagram of the kernel density of a charging station in a certain urban area provided in an embodiment of this application; Figure 10B This is a schematic diagram of the nuclear density of a chemical plant in a certain urban area provided in an embodiment of this application; Figure 10C This is a schematic diagram of the kernel density of a gas station in a certain urban area provided in an embodiment of this application; Figure 11 This is an overlay image of the distribution of fire stations and dynamic fire risk heat map of a certain urban area provided in the embodiments of this application; Figure 12 This is a schematic diagram of the isochronous distribution of fire stations in a certain urban area provided in an embodiment of this application; Figure 13A This is a 5-minute fire accessibility distribution map of a certain urban area provided in an embodiment of this application; Figure 13B This is a 10-minute fire accessibility distribution map of a certain urban area provided in an embodiment of this application; Figure 13C This is a 15-minute fire accessibility distribution map of a certain urban area provided in an embodiment of this application; Figure 14A This is a schematic diagram illustrating the results of a coupled analysis of fire risk and 5-minute fire accessibility in a certain urban area, provided by an embodiment of this application. Figure 14B This is a schematic diagram illustrating the results of a coupled analysis of fire risk and 10-minute fire accessibility in a certain urban area, provided by an embodiment of this application. Figure 14C This is a schematic diagram illustrating the results of a coupled analysis of fire risk and 15-minute fire accessibility in a certain urban area, provided by an embodiment of this application. Figure 15 This is a structural block diagram of a fire risk assessment system for outlying urban areas provided in an embodiment of this application; Figure 16 This is a structural block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0021] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0022] 1) State-space model (SSM): A mathematical framework for describing dynamic systems (such as time series, signal or control systems). It models time series data containing latent variables through two equations: the state equation (which describes the evolution of the internal state of the system over time) and the observation equation (which describes the relationship between the observable output and the internal state). It is widely used in filtering, prediction, control and other fields.

[0023] 2) Dynamic Graph Learning: This is an extension of graph neural networks in the temporal dimension. It aims to capture the patterns of node attributes and graph topology evolution over time, thereby enabling the modeling and prediction of complex interactive relationships in dynamic systems such as social networks and transportation networks.

[0024] 3) Geographic Information Model (GIS Model): A computer-based tool that integrates, manages, analyzes, and visualizes geospatial data to reveal spatial distribution patterns and support spatial decision-making and problem-solving.

[0025] 4) AHP model (Analogous Hierarchical Analysis): A multi-criteria decision-making model that hierarchically and structurally presents complex decision problems and judges the relative importance of each option through pairwise comparisons and calculations.

[0026] 5) POI (Point of Interest): A basic data unit in a geographic information system that represents a specific location in the real world (such as a shop, school, or station) and contains its precise coordinates, name, category, and other attribute information. It is the core carrier for spatial querying, analysis, and application.

[0027] 6) Vector concatenation: A basic and important tensor operation that connects two or more vectors (or feature tensors) end-to-end along a specified dimension (usually the feature dimension) to form a new vector with a higher dimension that contains all the input information, in order to achieve feature fusion or expansion.

[0028] 7) Hierarchical State Space Model (DyGHydra): A deep learning architecture designed specifically for dynamic graphs. It combines a node-level continuous-time state space model with hierarchical, multi-scale graph representation learning through a gated routing mechanism, thereby simultaneously and efficiently modeling the evolving node attributes and topology in dynamic graphs.

[0029] 8) Kernel Density Estimation: A nonparametric statistical method that estimates a continuous probability density function from a finite number of discrete sample points by placing a smooth kernel function (such as a Gaussian kernel) at each data point and superimposing all kernel functions, thus visually revealing the overall distribution of the data.

[0030] 9) n-hop neighbors: In graph theory and network analysis, "n-hop neighbors" specifically refers to the set of all other nodes that can be reached from a certain starting node through exactly or no more than n edges (i.e., "hops"). It is a basic topological concept for measuring the local influence of nodes, conducting network propagation and cluster analysis.

[0031] 10) Recursive Residual Network: It is an efficient extension of the classic Residual Network (ResNet). Its core idea is to recursively reuse the same residual block in multiple deep layers of the network (weight sharing), so that a very deep network can be built with fewer parameters, and effectively promote feature reuse and gradient flow, thereby improving parameter efficiency.

[0032] 11) Sine position coding: It is a fixed coding scheme based on sine / cosine functions in the Transformer architecture, which aims to provide the model with absolute position information of the sequence and enable it to capture the relative relationship between positions.

[0033] 12) Linear ordinary differential equations: Differential equations in which the unknown function and its derivatives all appear in the form of first powers. Their solutions have superposition properties and can usually be expressed as the sum of homogeneous general solutions and non-homogeneous particular solutions. They are the core mathematical tools for describing the deterministic evolution process of many physical and engineering systems.

[0034] 13) Multilayer Perceptron (MLP): It is the most basic and core feedforward neural network structure in deep learning. It consists of at least three layers (input layer, hidden layer, and output layer) of fully connected neurons and learns the complex mapping relationship between input and output through non-linear activation functions.

[0035] 14) Binary Cross Entropy Loss: A core metric used to evaluate the difference between the predicted probability of a binary classification model and the true label. It quantifies the prediction error by calculating the cross entropy of the two distributions and drives the update of model parameters.

[0036] In urban spatial structure research, the "urban fringe" refers to the transitional zone from the periphery of the urban built-up area to the rural hinterland, exhibiting a dynamic mixture of urban and rural land use characteristics. Its scope begins at the periphery of the urban core area and extends to the urban administrative boundary or natural geographical barriers. In this application, the "fringe urban area" is defined as the hinterland outside the built-up area relative to the central urban area, distinct from the "urban fringe," encompassing various land use types such as residential areas, industrial areas, and agricultural areas, as well as land use states including those under development, awaiting development, and prohibited from development.

[0037] Against the backdrop of accelerated urban expansion and integrated urban-rural development, the status of peripheral urban areas in some cities has undergone significant changes: from being reserves for urban construction to developing autonomously, industrial layout has expanded, land development intensity in some areas has increased significantly, and their status in the overall urban plan has risen accordingly, with infrastructure layout and construction gradually improving. However, the fire risk in peripheral urban areas has shown a significant upward trend. Influenced by industrial layout (chemical and logistics parks) and population influx, coupled with the large number of rural areas and high vegetation coverage, the risks of open fires and forest fire spread are superimposed, making fires difficult to control once they occur. At the same time, fire-fighting facilities in peripheral urban areas have obvious shortcomings: low density of fire stations and excessive service radius are prominent problems, making it difficult for fire trucks to meet response times. More importantly, the road network density in peripheral urban areas is relatively low, and the redundancy and resilience of the transportation network are insufficient. Once key road sections are dynamically disturbed by sudden traffic flows or accidents, they are prone to forming regional "rescue islands," resulting in actual rescue times far exceeding the theoretical values ​​calculated based on static road networks.

[0038] In view of this, this application proposes a fire risk assessment method and related equipment for peripheral urban areas. This method can combine the static risk characteristics of the urban area with dynamic traffic situation data to construct a deductive model input, and then use the frequency components of different frequency levels obtained by multi-frequency decomposition of the model to obtain the comprehensive risk characterization of each node in the dynamic map of the urban area. Finally, it combines the dynamic risk assessment of the node with the fire accessibility assessment results for coupled analysis, thereby achieving a more comprehensive and accurate fire risk assessment of peripheral urban areas.

[0039] Figure 1 This is an optional flowchart of a fire risk assessment method for peripheral urban areas provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S100 to S300.

[0040] Step S100: Construct a dynamic graph model of the target urban area based on the road network of the target urban area; the dynamic graph model includes several grid nodes and interaction edges between grid nodes; the types of grid nodes include demand points and fire stations.

[0041] A dynamic map model of the target urban area is constructed as the data foundation for urban fire risk assessment.

[0042] Step S200: Obtain static risk assessment data of the target urban area and its traffic situation data at the query time; the static risk assessment data includes data on disaster-causing factors, exposure factors, and disaster mitigation factors related to fire risk; the traffic situation data is obtained based on a preset window; the preset window records historical traffic interaction events that occurred between grid nodes within a preset time capacity before the query time based on a time elimination mechanism; the node feature sequence of each grid node is obtained by calculating based on the dynamic graph model of the target urban area, the static risk assessment data, and the traffic situation data at the query time.

[0043] Obtain static risk assessment data related to fire risk in the target urban area and its traffic situation data at the query time. Construct the node feature sequence of each grid node based on the dynamic graph model and the acquired data, and use it as input for the recursive inference model.

[0044] Step S300: Based on the preset recursive deduction model, multi-frequency decomposition is performed on the node feature sequence of each grid node, and the fire spatiotemporal risk of each grid node is analyzed according to the multi-frequency decomposition results to obtain the comprehensive risk characterization of each grid node; the multi-frequency decomposition results include frequency components of several different frequency levels; dynamic fire risk assessment and fire accessibility assessment of each grid node are performed based on the comprehensive risk characterization of each grid node, and coupling analysis is performed based on the results of dynamic fire risk assessment and fire accessibility assessment of each grid node, and the fire risk assessment result of the target urban area is determined according to the coupling analysis results of each grid node.

[0045] The recursive model performs multi-frequency decomposition on the input data and analyzes the frequency components at several different frequency levels to determine the comprehensive risk characterization of the grid nodes. Then, based on the comprehensive risk characterization of the grid nodes, it performs dynamic fire risk assessment and fire accessibility assessment of the nodes. Finally, it performs a coupled analysis based on the two assessment results to obtain the fire risk assessment results for the target urban area. In some embodiments, the identification process of the peripheral urban areas of a city using the method of this application is implemented in the following manner: Based on POI data of a city from map software, a kernel density estimation algorithm is used to calculate the clustering degree of spatial elements in the city, obtaining spatial distribution characteristics reflecting urban functions. Secondly, the relationship between density value d and distance S from the city center (corresponding to the search radius) is analyzed using the density-distance analysis method. When a specific density threshold is reached (i.e., the boundary of the outer urban area), the POI density value will show a significant decreasing trend. The area enclosed by this threshold is taken as the spatial range of the city's outer urban area. Figure 2 The image shown is a schematic diagram illustrating the results of identifying the outer urban areas of a city according to an embodiment of this application; wherein, Figure 2Figures a, b, c, and d in the figure are schematic diagrams of the interest point density identification results with search radii of 1km, 3km, 4km, and 5km, respectively (where the color from purple to red indicates the density from small to large). Figures 3A to 3C This is another schematic diagram showing the results of identifying the peripheral urban areas of a certain city; Figure 3A This is a schematic diagram of the results of interest point density identification in a certain city (where the color from purple to red indicates density from small to large). Figure 3B This is a schematic diagram showing the results of kernel density contour line division for a certain city; Figure 3C This is a schematic diagram showing the results of identifying the outskirts of a city.

[0046] In some embodiments, step S100, which involves constructing a dynamic graphical model of the target urban area based on the road network of the target urban area, includes: To construct a dynamic graph model, the geographic space within the identified edge urban area is divided into regular grid cells, and each grid cell is mapped to a spatial node V in the graph model. The physical road network connections between nodes are mapped to interactive edges E, thus completing the transformation from geographic space to graph topology. The spatial scale of the grid cells is set based on the effective operating radius of fire trucks and the minimum turning radius of narrow alleys in the edge urban area (e.g., 100m×100m to 500m×500m) to ensure that the spatial resolution can accurately reflect the traffic restrictions of the physical road network.

[0047] In some instances, step S200 involves obtaining static risk assessment data through the following steps: (1) Disaster-causing and exposure factor data: used to characterize the risk baseline of a node itself. Among them, disaster-causing factors ( This mainly refers to point hazards that may cause fires, including key fire safety units and fire hazard point (POI) data such as chemical plants, gas stations, and charging stations obtained from the open platforms of map software. This type of data often has the characteristics of high frequency and sudden occurrence; exposure factor ( This refers to objects that may bear fire damage, including population density data obtained from world population datasets and building outline and density data obtained from public street map websites. This type of data usually serves as a low-frequency accumulation risk base.

[0048] (2) Disaster mitigation factor data: used to characterize the infrastructure capacity of nodes to withstand disasters. This mainly includes the distribution locations of various fire stations (special service stations, level 1 stations, level 2 stations, micro-stations, etc.) obtained from map software, as well as road network density and road grade data reflecting regional rescue accessibility. For example... Figure 4The diagram shown is a schematic diagram of the distribution of urban fire stations provided in the embodiments of this application. In the diagram, circles represent mini fire stations, stars represent small fire stations, triangles represent secondary fire stations, hexagons represent primary fire stations, and rhombuses represent special service fire stations.

[0049] In some embodiments, step S200 involves a preset window recording historical traffic interaction events between grid nodes within a preset time capacity prior to the query time, based on a time elimination mechanism. This process includes: This application employs an Evolutionary Structure Context Window (ESCW) to dynamically maintain the road network's topology. Considering the real-time changes in traffic flow and fire risk, a static adjacency matrix cannot reflect the current accessibility status. The ESCW is designed as a dynamic memory window with a capacity of W, where W can be set to correspond to a physical time of 2 hours (i.e., covering a typical cycle of urban traffic congestion from occurrence to dissipation). It caches recent traffic interaction events in the road network in real time and uses a time-based eviction mechanism to maintain a fixed window capacity. When a new query occurs, the ESCW can not only sample the neighborhood list of each grid node from the dynamic graph but also quickly index the historical neighbor sequence of nodes through the cached topology, providing a complete local topological view for subsequent structural feature extraction. Here, let... For window capacity, with new traffic flow data or risk events... The input automatically removes the oldest history records, thus maintaining a timely awareness of the current road network topology in the edge urban areas.

[0050] In some embodiments, in step S200, the node feature sequence of the target grid node among several grid nodes is obtained in the following way: Step S210: Standardize the static risk assessment data of the target urban area, map the standardized data results to several grid nodes, and obtain the static feature vector of each grid node based on the mapping results.

[0051] Based on GIS spatial mapping technology, disaster-causing factors, exposure factors, and disaster mitigation factors across the entire region are distributed to each grid node: for point data (such as chemical plants and fire station POIs), the number or kernel density value falling into each grid is counted; for areal data (such as population distribution), the average density within each grid is calculated. Through the above processing, macro-level urban data is transformed into micro-level node feature vectors, thereby supporting the model to perform independent static risk assessments for each grid node.

[0052] Step S220: Based on the dynamic graph model of the target urban area and its traffic situation data at the query time, determine the neighbor node sequence and interaction edge sequence of each grid node; the neighbor node sequence includes several neighbor nodes that have historical traffic interaction events with the grid node; the interaction edge sequence indicates the interaction edges between the grid node and the neighbor nodes; the attributes of the interaction edges are determined by the traffic path planning between the nodes.

[0053] The neighboring node sequence and interaction edge sequence of each grid node are sampled by ESCW as the data basis for the subsequent construction of the node feature sequence of the grid node.

[0054] Step S230: Construct several node pairs corresponding to the target grid node; the node pairs indicate the correspondence between the target grid node and the fire stations in the several grid nodes; analyze the neighboring node sequence of the target grid node and its corresponding fire station in each node pair to obtain the multidimensional statistical vector of each node pair; encode and fuse the multidimensional statistical vectors of several node pairs to obtain the structural feature sequence of the target grid node.

[0055] Node pairs are constructed based on the correspondence between target grid nodes and their corresponding fire stations, and analysis is performed on these node pairs. Based on the neighboring node sequences of the target grid nodes and their corresponding fire stations, the interaction between the target grid nodes and their corresponding fire stations, as well as potential risk transmission paths, are analyzed. This yields a multi-dimensional statistical vector between the target grid nodes and each corresponding fire station, and a structural feature sequence. .

[0056] Step S240: Construct the original feature sequence of the target grid node based on the static feature vectors of the neighboring nodes in the neighboring node sequence and the attributes of the interactive edges in the interactive edge sequence. .

[0057] Encode the static features of the neighboring node sequence and the attribute data of the interactive edge sequence of the target grid node by encoding them respectively, construct the node and edge feature codes of the target grid node, and concatenate the encoding results to obtain the original feature sequence of the target grid node.

[0058] Step S250: Perform feature fusion processing based on the structural feature sequence of the target grid node and the original feature sequence to obtain the node feature sequence of the target grid node.

[0059] Structural feature sequence and the original feature sequence After mapping to a unified dimension in their respective linear projection layers, the original feature sequences and structural feature sequences are added and fused element by element. Local features are then extracted through a one-dimensional convolutional layer to finally form the node feature sequence of the target grid node, which serves as the data input for the recursive inference model.

[0060] In some embodiments, step S230, the process of analyzing the neighboring node sequence of the target grid node and its corresponding fire station in each node pair to obtain the multidimensional statistical vector of each node pair, includes: Define target mesh nodes of The skip neighborhood sequence is By calculating the one-hop neighborhood sequence of the target grid node and target nodes (i.e., corresponding fire stations) One-hop neighborhood sequence and two-hop neighborhood sequences The relationships between them are used to identify the "hidden high-order bridges" between the target node and the corresponding fire station.

[0061] (1) Count the total number of neighboring nodes in the neighboring node sequence of the target grid node to obtain the first feature.

[0062] First characteristic (local importance) This involves calculating the total number of neighboring nodes in the target grid node's neighborhood node sequence within the historical time period recorded in the ESCW window. In other words, it represents the frequency of interaction events occurring within the window's record capacity corresponding to the target grid node at the time of the query. This value reflects the node's... Current traffic activity levels.

[0063] (1) (2) Count the frequency of each neighboring node in the neighboring node sequence of the target grid node in the corresponding fire station's neighboring node sequence to obtain the second feature.

[0064] Second feature (one-hop co-occurrence feature) : Compute the compute node At the fire station One-hop neighborhood The frequency of occurrence in [the context]. This is a traditional common neighbor indicator, representing explicit direct connections: (2) (3) Count the number of intersection nodes between the neighboring node sequence of the target grid node and the neighboring node sequence of the corresponding fire station to obtain the third feature.

[0065] The third characteristic (potential high-order bridge characteristics) : Statistics show that the same node appears in the target grid. One-hop neighborhood list and corresponding fire stations One-hop neighborhood list The number of intermediate nodes (i.e., the size of the intersection of the one-hop neighborhood sequences, or the frequency with which nodes in the target grid node's neighborhood sequence appear in the corresponding fire station's two-hop neighborhood sequence). This metric quantifies the likelihood of the two being connected via intermediate steps.

[0066] (3) The physical meaning of this indicator is: even risk points and fire station There are no direct common neighbors (i.e.) ),if This means that a path exists. (Potential intermediate point) This "cross-neighborhood" matching mechanism can uncover deeper rescue channels.

[0067] (4) Count the frequency of occurrence of the corresponding fire station in the neighboring node sequence of the target grid node to obtain the fourth feature; count the frequency of occurrence of the target grid node in the neighboring node sequence of the corresponding fire station to obtain the fifth feature; calculate the product of the fourth feature and the fifth feature to obtain the sixth feature.

[0068] Fourth feature (direct interaction feature) Recording the fire station At the target grid node A leap in neighboring history The frequency of direct occurrences in the data, capturing historical records of direct rescue efforts: (4) Fifth feature (direct interaction feature) Record target mesh nodes. At the fire station A leap in neighboring history Direct frequency of occurrence in: (5) Sixth characteristic (reciprocity characteristic) Calculate the product of two-way interactions to measure the stability of two-way accessibility of rescue routes: (6) (5) Based on the first feature, second feature, third feature, fourth feature, fifth feature and sixth feature, a multidimensional statistical vector of node pairs is constructed.

[0069] In some embodiments, step S300, the comprehensive risk characterization of the target grid node among several grid nodes is obtained in the following way: Step S310: Calculate the time difference between the time when the historical traffic interaction event occurs between the target grid node and its neighboring nodes and the query time. Map the calculation result through sinusoidal position coding to obtain the time interval sequence of the target grid node.

[0070] Record the time of each historical interaction between the target grid node and its neighboring grounds, and the current query time. Time difference between The time interval sequence of the target grid nodes is obtained by mapping the sinusoidal position encoding to a multi-frequency time feature vector. .

[0071] Step S320: Multi-frequency decomposition is performed using a recursive residual network based on the node feature sequence and time interval sequence of the target grid node to obtain several layers of frequency components at different frequency levels and the time feature components corresponding to each layer of frequency components.

[0072] The recursive inference model utilizes a recursive residual network to perform multi-frequency decomposition on the input node feature sequence and time interval sequence, obtaining the data foundation for subsequent processing. Here, let the... The input sequence at the layer time scale is After filtering out high-frequency noise using one-dimensional average pooling (AvgPool), the unique frequency components of this layer are separated using residual calculation. : (7) in This is achieved through frequency domain interpolation, ensuring the orthogonality of risk signals in the spectrum across different time scales (such as annual planning cycles and daytime traffic cycles). Layer decomposition decouples the originally entangled fire risk into a series of independent feature subspaces, each corresponding to risk variables of different frequencies, such as the slow variable of urban development and the fast variable of sudden events.

[0073] Simultaneously, the time interval sequence of the target grid nodes By performing layer-by-layer decomposition, the time feature components corresponding to the frequency components at each layer are obtained. .

[0074] Step S330: Generate the discretization step size corresponding to each frequency component based on the time feature component corresponding to each frequency component.

[0075] To model the evolution of each frequency component over time, this method deploys a structure-guided continuous-time state-space model (CT-SSM) at each layer. Traditional discrete models struggle to adapt to the non-uniform time intervals between events such as fire alarms and traffic congestion; therefore, this embodiment introduces a physical time interval. (i.e., discretization step size) is used for continuous modeling; the model utilizes the k-th layer time feature components obtained from the multi-frequency decomposition mentioned above. To dynamically generate the discretization step size, specifically... Derived from time interval sequence In the The decomposition results at the layer scale, where Each element is a sinusoidal encoded vector of the time difference between the historical interaction time and the current query time.

[0076] The formula for calculating the discretization step size is: (8); In the formula, Linear(·) is a learnable linear transformation layer, and Softplus(·) is the activation function to ensure a positive output. This design allows Δt to adaptively adjust according to the actual time interval: when the historical interaction is close to the current time, Δt is small, the state decay is slow, and the system retains more recent memory; when the time interval is large, Δt increases, the state decay is accelerated, and old information is forgotten more quickly. This step size... The generation mechanism can adaptively handle non-uniform time interval events such as fire alarms or traffic congestion.

[0077] Step S340: Based on the preset equation, each frequency component and its corresponding discretization step size, perform state evolution modeling for each frequency component to obtain the state evolution modeling result for each frequency component.

[0078] By introducing a continuous-time state-space model (CT-SSM) for each layer and modeling the evolution of each frequency component over time using a discretized step size for each layer, the state evolution modeling results of each frequency component are obtained.

[0079] Step S350: A bottom-up strategy is used to fuse the state evolution modeling results of several layers of frequency components to obtain a comprehensive risk characterization of the target grid node.

[0080] The model employs a bottom-up approach for state fusion. The hidden states output by each layer of the CT-SSM are shown. Through layer-by-layer upsampling and overlay, a comprehensive risk representation for the current grid node is generated. The specific calculation formula is shown in equation (9): (9) in This represents the gating modulation of the original fine-grained features. This fusion process ensures that the final output risk embedding vector includes both macroscopic urban safety trends and retains extremely high sensitivity to microscopic risk perturbations at the current moment, providing a complete spatiotemporal state input for subsequent accessibility prediction.

[0081] The method in this application is based on linear ordinary differential equations of a state-space model (SSM), the general form of which is: The parameters are defined as follows: : The hidden state vector represents the system's compressed memory of historical information at time t. In the fire protection scenario of this invention, h(t) encodes the risk situation information accumulated by the grid node up to time t, including historical traffic flow patterns, the impact of historical risk events, etc.

[0082] The derivative of the hidden state with respect to time describes the instantaneous rate of change of the risk memory state.

[0083] The state transition matrix controls the natural evolution of hidden states. Physically, it represents the decay rate of risk memory—when a region experiences no new traffic interactions or risk events for an extended period, historical information will gradually decay according to the law determined by A.

[0084] The input projection matrix determines the intensity and method of writing new external information into the system state.

[0085] The input signal at the current moment corresponds to the fused feature sequence described above at time step [missing information]. The feature value carries dynamic information such as the travel time and congestion index of the road segment at that moment.

[0086] This equation describes the fundamental dynamic process of "system memory = natural evolution of old memories + writing of new information".

[0087] In the modeling process based on the above formula, the method in this application introduces discretization step size and node structural feature sequence for adjustment, as detailed below: In some embodiments, step S340 involves performing state evolution modeling on each frequency component based on a preset equation, each frequency component, and its corresponding discretization step size, to obtain the state evolution modeling result for each frequency component. This process includes the following steps: Step S341: Adjust the preset state transition matrix based on the discretization step size corresponding to each frequency component, and determine the state transition matrix corresponding to each frequency component based on the adjustment result. .

[0088] This invention introduces a physical time interval The above continuous equation is discretized to fit the actual discrete time series data. The specific formula for calculating the state transition matrix is ​​shown in equation (10): (10) in, This is the state transition matrix corresponding to the frequency components of the corresponding layer. This represents the discretization step size corresponding to the frequency components of the corresponding layer. This is the preset state transition matrix.

[0089] In fire safety scenarios, the physical meaning of this formula is that the decay rate of risk memory is strictly controlled by physical time. When an area has not experienced a fire or traffic interaction for a long period (…), Increase), historical risk status The rapid decay according to an exponential law reflects the natural decay mechanism of urban systems for outdated information.

[0090] Step S342: Calculate the discretized input projection matrix corresponding to each frequency component based on the structural feature sequence of the target grid node, the preset state transition matrix, and the discretization step size corresponding to each frequency component.

[0091] To address the "spatiotemporal entanglement" problem—that is, the risk of critical road network nodes has stronger persistence—the model introduces structural feature sequence embedding. For the input matrix and output matrix Dynamic parameterized modulation is performed. The discretized input projection matrix... The calculation is as follows: (11) Simultaneously, output projection matrix Defined as: (12) The above formula shows that the impact intensity of external risk signals (such as sudden traffic accidents) on the latent state of the system is no longer constant, but is affected by the current local road network structure. Weighted. For nodes located at transportation hubs or hidden evacuation routes, their larger weights are weighted. The value will be significantly amplified. and This allows the potential risk status to be quickly activated and output even after a long time interval.

[0092] Step S343: Based on the preset equation, each layer of frequency components and their corresponding state transition matrix and discretized input projection matrix, calculation is performed to obtain the hidden state vector corresponding to each layer of frequency components.

[0093] Based on the adjusted state transition matrix and the discretized projection matrix, the hidden state vector corresponding to each frequency component is calculated using ordinary differential equations. .

[0094] Step S344: Construct the output projection matrix based on the structural feature sequence of the target grid nodes; calculate the state evolution modeling results of each frequency component based on the output projection matrix and the hidden state vector corresponding to each frequency component.

[0095] Based on the output projection matrix and the hidden state vector corresponding to each frequency component, the state evolution modeling results for each frequency component are obtained. The state evolution modeling results are... The output is calculated according to equation (13): (13) Traditional fire accessibility analysis typically relies on static road networks to calculate the coverage area using a two-step movement search method, which fails to reflect the nonlinear impact of dynamic factors such as traffic congestion and road network vulnerability on rescue timeliness. This application's method utilizes the modeling capabilities of the DyGHydra model to reconstruct the fire accessibility problem as a link prediction task on a dynamic graph, i.e., predicting the path forward at a given future time. Fire station nodes With target mesh node Whether an effective rescue connection can be established between them. In some embodiments, step S300, the fire accessibility assessment process for the target grid node among several grid nodes is implemented through the following steps: Step S361: Construct several joint node pairs corresponding to the target grid node; the joint node pairs indicate the correspondence between the grid node and the corresponding fire stations in the several grid nodes.

[0096] Construct joint node pairs between the target grid node and several fire stations as the objects of subsequent analysis.

[0097] Step S362: Based on the comprehensive risk characterization of the target grid nodes and their corresponding fire stations in each joint node pair, construct the joint risk characterization corresponding to each joint node pair.

[0098] Based on the comprehensive risk representation of each joint node to the target grid node and its corresponding fire station, the joint risk representation of each joint node is constructed. , used to characterize the dynamic interaction possibilities between the two.

[0099] Step S363: Input the joint risk representation of each joint node into the preset fire accessibility prediction model for calculation, and determine the fire accessibility probability between the target grid node and each corresponding fire station based on the calculation results.

[0100] This application's method decodes the joint risk representation of joint nodes based on the reachability prediction head of a multilayer perceptron, thereby determining the fire accessibility probability between the target grid node and each corresponding fire station. The specific probability calculation formula is as follows: (14) In the formula, This represents the Sigmoid activation function. This represents a vector concatenation operation. and These are the learnable weight matrix and bias term, respectively. Output value. This quantitatively characterizes the confidence level of the probability that rescue forces will arrive at the risk area within a specified time threshold, given the traffic situation and road network structure at the current query time. Furthermore, during the model training phase, the binary cross-entropy loss function is used to calculate the predicted connectivity probabilities. The error between the model and historical actual access records (accessible / inaccessible) is calculated, and the model parameters are optimized through backpropagation.

[0101] Step S364: Analyze the fire accessibility probability between the target grid node and each corresponding fire station to obtain the fire accessibility assessment result of the target grid node.

[0102] The fire accessibility probability between the target grid node and each corresponding fire station is analyzed. For example, the highest probability of reaching the rescue within a preset time step is analyzed to determine whether the construction of fire protection facilities around the grid node needs to be adjusted.

[0103] In some embodiments, step S362, the process of constructing the joint risk representation for each joint node pair based on the comprehensive risk representation of the target grid nodes and their corresponding fire stations, includes: (1) Based on the weighted aggregation mechanism, the comprehensive risk representation of the target grid node and its corresponding fire station in the joint node pair is compressed to obtain the node embedding vector of the target grid node and its corresponding fire station in the joint node pair; Risk representation of grid nodes obtained from recursive deduction model The sequence is in sequential form, while link prediction tasks require a single node-level representation vector. Therefore, this application's method compresses the sequence into node embedding vectors based on a learnable weighted aggregation mechanism. For the target grid node... Node embedding vector The specific calculation formula is shown in equation (15): (15) In the formula, This represents the sequence representation of CT-HMamba output when the target grid node u is the center node. To learn weight vectors The calculated attention weights ( = This is used to adaptively weight the importance of each time step in the sequence. Similarly, the corresponding fire stations can be obtained. Node embedding vector .

[0104] (2) Based on the node embedding vectors of the target grid nodes and their corresponding fire stations in the joint node pair, the joint risk representation of the joint node pair is obtained by vector splicing.

[0105] The joint risk representation is obtained by concatenating the node embedding vectors of the target grid nodes and their corresponding fire stations in the joint node pair. , This indicates a vector concatenation operation.

[0106] The dynamic fire risk assessment results of the grid nodes in this application include a dynamic fire risk index obtained by mapping the comprehensive risk representation of the grid nodes based on a pre-set multilayer perceptron. To this end, this application's method incorporates a multilayer perceptron (MLP) as a decoding head at the end of the recursive derivation model, thus transforming the high-dimensional comprehensive risk representation... Mapped to scalar risk index This index directly reflects the comprehensive fire risk level of each grid cell under the current spatiotemporal context and can be used to generate dynamic risk heat maps. Additionally, the fire accessibility assessment results indicate the fire accessibility probability between the grid node and each corresponding fire station in the dynamic graph model; in some embodiments, step S300, the process of coupling analysis based on the dynamic fire risk assessment and fire accessibility assessment results of each grid node, includes: Step S371: Analyze the fire accessibility probability between the grid node and each corresponding fire station in the dynamic model to determine the fire accessibility probability between the corresponding fire station and the grid node that meets the preset requirements.

[0107] Step S372: The dynamic fire risk index of the grid node is judged according to the preset fire risk judgment threshold to obtain the first judgment result.

[0108] The dynamic fire risk index of a node is determined based on a preset fire risk assessment threshold to determine whether it meets the preset requirements, and the corresponding judgment result is output.

[0109] Step S373: Based on the preset accessibility threshold, the fire accessibility probability between the corresponding fire station and the grid node that meets the preset requirements is judged to obtain the second discrimination result.

[0110] Based on the preset accessibility threshold, the accessibility probability between each fire station and the target grid node is determined and the corresponding judgment result is output.

[0111] Step S374: Calculate the coupling analysis results of the grid nodes based on the preset coupling discrimination function, the first discrimination result, and the second discrimination result.

[0112] Construct the coupled discriminant function as shown in equation (16). For global grid nodes Perform a scan: (16) In the formula, For indicator functions, This indicates iterating through all available fire stations. Select the points that reach the risk points The value representing the highest probability (i.e., the fastest rescue response) indicates the best level of rescue support available in that area. This is a dynamic fire risk index. The high-risk threshold is set to 0.8 in this embodiment. The reachability threshold is set to 0.3 in this embodiment, meaning a connection probability below 30% is considered unreachable. When At that time, the grid was identified as a hidden service blind spot. Using this method, the study can accurately locate key areas that are physically close but face difficulties in actual rescue due to fragile road network structures or traffic flow interference, providing a forward-looking quantitative basis for the precise allocation of fire-fighting resources and road network optimization.

[0113] Ultimately, the method of this application can construct a dynamic fire risk distribution map based on the dynamic fire risk index of each grid node and a fire rescue accessibility probability map based on the accessibility prediction probability of each grid node. By overlaying the 'dynamic fire risk distribution map' and the 'fire rescue accessibility probability map', the method can identify the spatial mismatch areas between the two (i.e., high-risk-low accessibility areas), which can be used as the final conclusion for fire safety assessment of peripheral urban areas.

[0114] In summary, the embodiments of this method include, but are not limited to, the following beneficial effects: (1) The traffic situation data in this application is recorded through a preset window and based on a time elimination mechanism, which records historical traffic interaction events that occurred within a preset time capacity before the query time. The data recorded by this data recording method can be updated in real time. Traditional GIS or AHP models are mostly based on "snapshot" static data. However, the method in this application introduces real-time updated traffic situation data, which can improve the ability to capture spatiotemporal entanglement. That is, it simultaneously considers static road network data, risk characteristic data and dynamic traffic situation data, thereby effectively improving the comprehensiveness and accuracy of urban fire risk assessment.

[0115] (2) In view of the characteristics of sparse road network data and low node degree in marginal urban areas, traditional graph neural networks (such as GCN and GAT) often suffer from feature smoothing or failure due to the lack of neighbor information because they rely heavily on the message aggregation mechanism of one-hop neighbors. In the process of constructing the feature sequence of grid nodes, the scheme of this application breaks through the traditional bottleneck and is based on the target grid nodes. One-hop neighborhood sequence and target node (i.e., the corresponding fire station) The association between the one-hop neighborhood sequence and the two-hop neighborhood sequence of a target node is used to identify the "hidden high-order bridge" between the target node and the corresponding fire station.

[0116] (3) The method of this application can analyze the fire risk of grid nodes based on the correspondence between different assessment results by coupling the dynamic fire risk assessment results and the fire accessibility assessment results of the nodes, so as to improve the effectiveness of the fire risk assessment results of the grid nodes. For example, if a certain grid node has a high dynamic fire risk and poor fire accessibility, i.e. a hidden service blind spot, then the grid node needs to be given more attention in terms of fire protection deployment in the urban area.

[0117] (4) Compared with the central urban area, the level of fire protection and transportation facilities in the peripheral urban area may be lower. Therefore, introducing real-time traffic interaction data into the fire risk assessment process of the peripheral urban area and considering the correspondence between the assessment results of dynamic fire risk and fire accessibility can effectively improve the effectiveness of the fire risk assessment results of the peripheral urban area.

[0118] Figure 5 and Figure 6 This application provides a flowchart and process architecture diagram of a method for urban fire risk assessment and accessibility prediction based on dynamic graph learning, which includes the following steps: Step 1: Multi-source data mapping and gridding First, the geographic space of the target urban area is divided into regular grid cells, and each grid cell is mapped to a spatial node V in the graph model. The physical road network connections between nodes are mapped to initial edges E, thus completing the transformation from geographic space to graph topology and obtaining a dynamic graph of the target urban area. Second, multi-source data of the target urban area is acquired, including disaster mitigation factors, exposure factors, and disaster-causing factors. Disaster-causing and exposure factor data are used to characterize the risk baseline of the nodes themselves, and disaster-causing factors ( This mainly refers to point hazards that may cause fires, including key fire safety units and fire hazard point (POI) data such as chemical plants, gas stations, and charging stations obtained from the open platform of map software; exposure factor (…). The data refers to objects that may suffer fire damage, including population density data obtained from world population datasets and building outline and density data obtained from public street map websites; disaster mitigation factor data is used to characterize the infrastructure capacity of nodes to resist disasters, including the distribution of various fire stations obtained from map software, and road network density and road grade data reflecting regional rescue accessibility.

[0119] Subsequently, based on GIS spatial mapping technology, disaster-causing, disaster mitigation, and exposure factor data across the entire region are allocated to each grid node and encoded: for point data (such as chemical plants, fire stations, POIs), the number or kernel density value falling into each grid is counted; for areal data (such as population distribution), the average density within each grid is calculated. Through the above processing, macro-level urban data is transformed into micro-level node feature vectors (such as... Figure 6 Shown in as well as This allows the model to perform independent risk assessments for each grid.

[0120] In addition, POI data of fire stations and dynamic traffic situation data within the region are acquired to construct the edge attributes of the dynamic map. Specifically, the coordinates and names of fire stations within the urban area are obtained from map software, demand points are created in the GIS, and several paths between fire stations and demand points are determined. Real-time driving route planning data is then obtained from map software, returning the travel time, distance, and number of traffic lights for each path. Considering that fire trucks can ignore traffic lights while ensuring safety during rescue missions, the average waiting time at traffic lights is subtracted from the travel time data for each path to obtain the actual travel time of the fire truck. This application's method will use nodes in the dynamic map... and Edge weights between Defined as the ratio of the physical length of a road segment to the real-time traffic speed. When a road closure event occurs, the weight is set to infinity. Simultaneously, to capture multi-frequency traffic dynamics characteristics, the crawling time granularity is set to one frame every 5 minutes, thereby generating a continuous time-stamped traffic data sequence. This sequence serves as the time-varying input to the edge attributes of the dynamic graph model, and after encoding, the edge feature vectors of the nodes are obtained (e.g., ...). Figure 6 Shown in as well as ).

[0121] In summary, by concatenating the node feature vectors and edge feature vectors of the grid nodes, we obtain the basic feature sequence of the grid nodes (e.g., ...). Figure 6 Shown in as well as ).

[0122] Step 2: Dynamic Graph Structure Feature Mining This application introduces an Evolutionary Structure Context Window (ESCW) and a Cross-Neighborhood Interaction Encoder (CNIE) to automatically identify "hidden bridges" in the road network by mining the interaction features between the one-hop neighborhood of the source node (reference point / risk node) and the two-hop neighborhood of the target node (fire station).

[0123] First, an Evolutionary Structure Context Window (ESCW) is constructed to maintain the dynamic topology of the road network through dynamic updates. Considering the real-time changes in traffic flow and fire risk, a static adjacency matrix cannot reflect the current accessibility status. The ESCW is designed as a dynamic memory window with a capacity of W, where W is set to correspond to a physical time of 2 hours (i.e., covering a typical cycle of urban traffic congestion from occurrence to dissipation). It caches recent traffic interaction events in the road network in real time and uses a time-based replacement mechanism to maintain a fixed window capacity. When a new interaction occurs, the ESCW not only updates the one-hop neighbor list (e.g., ...) Figure 6 As shown in the middle: For In terms of This includes ud and ua, as well as for In terms of (including uh and ua), and can also quickly index the two-hop neighbor sequence of a node through the cached topology (e.g. Figure 6 As shown in the article, for In terms of , including dh and av, for In terms of (including hd and au), providing a complete local topological view for subsequent structural feature extraction. Let... For window capacity, with new traffic flow data or risk events... The input automatically removes the oldest history records, thus maintaining a timely awareness of the current road network topology in the edge urban areas.

[0124] Secondly, the cross-neighborhood interaction feature encoder (CNIE) is used to quantify the potential correlation strength between regions. Given the sparse road network data and low node degree in peripheral urban areas, traditional graph neural networks (such as GCN and GAT) often suffer from feature smoothing or failure due to the lack of neighbor information, as they heavily rely on one-hop neighbor message aggregation mechanisms. To overcome this non-generality bottleneck, the CNIE module provided in this application's method architecture is no longer limited to directly connected (1-hop) nodes, but explicitly compares the source node (risk point). ) and target node (fire station) The historical neighborhood structure (including one-hop neighbor sequences and two-hop neighbor sequences) of each grid node is used to identify potential risk transmission paths, thereby obtaining multi-dimensional statistical vectors for each grid node. These vectors are then further encoded using a multilayer perceptron to obtain the structural feature sequences of the nodes (e.g., ...). Figure 6 Shown in as well as This feature vector explicitly contains multi-level topological information from direct connections to higher-order bridges, effectively solving the problem of missing structural information under sparse road networks in peripheral urban areas.

[0125] Step 3: Spatiotemporal risk dynamic evolution modeling The evolution of fire risk in peripheral urban areas is a typical non-stationary stochastic process, characterized by the superposition of "long-term urban expansion trends" and "short-term sudden accident disturbances." To mathematically decouple these two distinct risk dynamics and address the nonlinear modulation of risk propagation timeliness by road network topology, this application constructs a method as follows: Figure 7 The continuous-time hierarchical state-space model (CT-HMamba) is shown. This model achieves dynamic prediction of fire risk for each grid node through recursive frequency decomposition and structure-guided state evolution mechanism.

[0126] (1) Input structure of the CT-HMamba module: For a central node (such as a risk point u or a fire station v), the model samples its L historical one-hop neighbors and constructs the following three feature sequences of length L: Original feature sequence Each historical neighbor's characteristics are composed of two parts: the first is the static attributes of the neighbor node (derived from the disaster-causing, disaster-mitigating, and exposure factor data mentioned above, such as population density, building density, and hazard source distribution), and the second is the edge attributes of the interaction (derived from the dynamic traffic data mentioned above, such as the real-time travel time and congestion index of the road segment at that moment).

[0127] Structural feature sequence The high-dimensional embedding obtained by encoding the multidimensional statistical vector r calculated by the CNIE module for each historical neighbor with a multilayer perceptron (MLP) represents the structural importance of the interaction in the road network topology.

[0128] Time interval sequence Record the time of each historical interaction and the current query time. Time difference between (like Figure 6 As shown And so on), and mapped to multi-frequency time feature vectors through sinusoidal encoding.

[0129] After the three sequences are mapped to a unified dimension through their respective linear projection layers, the original feature sequence Z and the structural feature sequence are compared. The elements are added together and fused one by one, and then local features are extracted through a one-dimensional convolutional layer to finally generate a fused feature sequence that enters the multi-frequency decomposition module (CT-HMamba module).

[0130] (2) Multi-frequency risk decomposition The CT-HMamba module utilizes a recursive residual network to perform multi-frequency decomposition on the input fused feature sequence. Through K-layer recursive residual decomposition, the input sequence... Decomposed into Frequency components ,in: The highest frequency component represents short-term, sudden risk disturbances at the hourly level (such as traffic accidents or temporary traffic control). This is an intermediate frequency component, representing periodic risk fluctuations from daily to weekly levels (such as morning and evening peak hours, weekday / weekend differences). The lowest frequency component represents the long-term trend risk base (such as urban expansion, population growth, and infrastructure aging) from the monthly to the grade level.

[0131] (3) Structure-guided state evolution Based on this, in order to model the evolution of each frequency component over time, the method in this application deploys structure-guided features such as... Figure 8 The continuous-time state-space model (CT-SSM) is shown. Traditional discrete models struggle to adapt to the non-uniform time intervals of events such as fire alarms and traffic congestion. Therefore, this method introduces a physical time interval Δt for continuous modeling, based on linear ordinary differential equations of the state-space model (SSM). Its general equation is: Specifically, During the state evolution phase, the model utilizes the k-th layer time feature components obtained from the multi-frequency decomposition described above. To dynamically generate the discretization step size, specifically, T_k is derived from the time interval sequence. The decomposition results at the k-th level, where Each element is a sinusoidal encoded vector representing the time difference between the historical interaction time and the current query time. The formula for calculating the discretization step size is: In the formula, Linear(·) is a learnable linear transformation layer, and Softplus(·) is the activation function to ensure a positive output. This design allows Δt to adaptively adjust according to the actual time interval: when the historical interaction is close to the current time, Δt is small, the state decay is slow, and the system retains more recent memory; when the time interval is large, Δt increases, the state decay is accelerated, and old information is forgotten more quickly. This step size... The generation mechanism can adaptively handle non-uniform time interval events such as fire alarms or traffic congestion.

[0132] Calculate the state transition matrix based on the discretization step size. : (17) In fire safety scenarios, the physical meaning of this formula is that the decay rate of risk memory is strictly controlled by physical time. When an area has not experienced a fire or traffic interaction for a long period (…), Increase), historical risk status The rapid decay according to an exponential law reflects the natural decay mechanism of urban systems for outdated information.

[0133] To address the "spatiotemporal entanglement" problem—that is, the risk of critical road network nodes has stronger persistence—the model introduces a sequence of structural features extracted by the CNIE module. For the input matrix and output matrix Dynamic parameterized modulation is performed. The discretized input projection matrix... The calculation is as follows: (18) Simultaneously, output projection matrix Defined as: (19) The above formula shows that the impact intensity of external risk signals (such as sudden traffic accidents) on the latent state of the system is no longer constant, but is affected by the current local road network structure. Weighted. For nodes located at transportation hubs or hidden evacuation routes, their larger weights are weighted. The value will be significantly amplified. and This allows the potential risk status to be quickly activated and output even after a long time interval.

[0134] (4) Hierarchical integration Finally, the model employs a bottom-up approach for state fusion. The hidden states H_k output by each layer of CT-SSM are upsampled and superimposed layer by layer to generate a comprehensive risk representation for the current grid node. : (20) in This represents the gating modulation of the original fine-grained features. This fusion process ensures that the final output risk embedding vector includes both macroscopic urban safety trends and retains extremely high sensitivity to microscopic risk perturbations at the current moment, providing a complete spatiotemporal state input for subsequent accessibility prediction.

[0135] Step 4: Dual-link prediction output (1) Regional dynamic risk prediction To achieve intuitive quantification and visualization of risk, the method in this application uses a multilayer perceptron (MLP) as the decoding head to represent high-dimensional comprehensive risk. Mapped to scalar risk index This index directly reflects the comprehensive fire risk level of each grid cell under the current spatiotemporal context, and is used to generate dynamic fire risk heat maps. For example... Figure 9 The image shown is a dynamic fire risk heat map of a certain urban area provided in an embodiment of this application. Figure 9 It can be seen that the high-risk fire areas identified by the model (red and orange blocks) exhibit significant spatial clustering characteristics, mainly concentrated in the southern and southwestern parts of the study area. Specifically, the typical characteristic of high-risk fire areas is the "double high superposition": on the one hand, high-density buildings and population constitute an extremely strong low-frequency risk base (high exposure E); on the other hand, dense commercial activities and industrial production lead to high-frequency instantaneous risk disturbances (strong hazard H). More importantly, the model prediction results reveal the strong guiding role of road network structure on risk distribution. Overall, the red high-risk areas exhibit a clear "extending along traffic arteries" elongated shape, which verifies the effectiveness of the cross-neighborhood interaction mechanism (CNIE) in the DyGHydra model—that is, risks tend to be transmitted and diffused along the main road network with high traffic accessibility and frequent logistics interaction. Simultaneously, this can be combined with kernel density analysis of hazardous sources such as chemical plants and gas stations in the urban area (…). Figure 10A , 10B And 10C is a schematic diagram of the nuclear density of a hazardous source in a certain urban area provided in an embodiment of this application; Figure 10A , Figure 10B as well as Figure 10CThe images show kernel density maps of urban charging stations, chemical plants, and gas stations (with light pink to red indicating increasing machinations). It can be seen that the model accurately captures the radiation impact of these point-like hazards on the surrounding areas. Figure 11 This is an overlay diagram of the distribution of fire stations and dynamic fire risk heat map of a certain urban area provided in the embodiments of this application. The correspondence between the distribution of fire stations and fire risk can be analyzed by combining the overlay diagram. The fire risk assessment score is from small to large, from green to red. The above example proves the superiority of dynamic graph model in handling complex spatial heterogeneity and nonlinear risk coupling.

[0136] (2) Fire and rescue accessibility prediction First, a joint interaction representation of node pairs is constructed. This relates to the comprehensive risk representation output by the CT-HMamba module mentioned earlier. The sequence is in sequence form, while link prediction tasks require a single node-level representation vector. Therefore, this invention employs a learnable weighted aggregation mechanism to compress the sequence into node embeddings: (twenty one) In the formula, This represents the sequence output by CT-HMamba when the criterion grid node u is the center node. To learn weight vectors The calculated attention weights ( = This is used to adaptively weight the importance of each time step in the sequence. Similarly, the corresponding fire stations can be obtained. Node embedding vector .

[0137] To capture the dynamic interaction possibilities between the two vectors, they are concatenated and then input into a multilayer perceptron-based reachability prediction head for decoding. The specific probability calculation formula is as follows: (twenty two) In the formula, This represents the Sigmoid activation function. This represents a vector concatenation operation. and These are the learnable weight matrix and bias term, respectively. Output value. This quantitatively characterizes the confidence level of the probability that rescue forces will arrive at the risk area within a specified time threshold, given the traffic situation and road network structure at the current query time. Furthermore, during the model training phase, the binary cross-entropy loss function is used to calculate the predicted connectivity probabilities. The error between the model and historical access records (accessible / inaccessible) is analyzed, and the model parameters are optimized through backpropagation. Secondly, dynamic blind spot identification is performed based on probability thresholds. An accessibility determination threshold is set. When the model outputs the connection probability At that time, determine the region at the current moment. This indicates an unreachable state. This mechanism allows the model to overcome the limitations of traditional static buffer analysis, outputting a dynamic service boundary that fluctuates with traffic tides or sudden events, accurately reflecting the time-varying characteristics of road network accessibility in peripheral urban areas. For example... Figure 12 The diagram shown is a schematic diagram of the isochronous distribution of fire stations in a certain urban area, provided in an embodiment of this application. The diagram illustrates the shortest travel time required from any fire station in the urban area to every point of need. For example... Figure 13A , Figure 13B as well as Figure 13C As shown in the embodiment of this application, the fire accessibility distribution map of a certain urban area at 5 minutes, 10 minutes, and 15 minutes (where yellow to red indicates the fire accessibility probability from small to large) can be analyzed from the map to obtain the fire rescue accessibility probability of each grid node in the urban area, and rescue islands can be found, so as to make targeted fire protection construction deployment.

[0138] Step 5: Spatiotemporal Coupling Decision Analysis Construct the coupled discriminant function as shown in equation (16). For global grid nodes Perform a scan: (twenty three) In the formula, For indicator functions, This indicates iterating through all available fire stations. Select the points that reach the risk points The value representing the highest probability (i.e., the fastest rescue response) indicates the best level of rescue support available in that area. This is a dynamic fire risk index. The high-risk threshold is set to 0.8 in this embodiment. The reachability threshold is set to 0.3 in this embodiment, meaning a connection probability below 30% is considered unreachable. When At that time, the grid was identified as a hidden service blind spot.

[0139] Ultimately, the method of this application can construct a dynamic fire risk distribution map based on the dynamic fire risk index of each grid node and a fire rescue accessibility probability map based on the accessibility prediction probability of each grid node. By overlaying the 'dynamic fire risk distribution map' and the 'fire rescue accessibility probability map', the method can identify the spatial mismatch areas between the two (i.e., high-risk-low accessibility areas), which can be used as the final conclusion for fire safety assessment of peripheral urban areas.

[0140] In some embodiments, such as Figures 14A to 14C The diagram shown is a schematic representation of the results of a coupled analysis of fire risk and fire accessibility in a certain urban area, provided by an embodiment of this application. Figure 14A , Figure 14B as well as Figure 14C This is a coupled map of the high-risk fire zones in the city (marked by red lines) and the fire accessibility distribution maps at 5 minutes, 10 minutes, and 15 minutes (where yellow to red indicates increasing fire accessibility). This allows us to analyze the spatiotemporal disconnect between the city's fire service "supply capacity" and "demand pressure": the orange and red areas in the map represent high-risk clusters identified by the model, mainly distributed in densely populated built-up areas, industrial parks, and around transportation hubs. However, when we map these "hotspots" onto the 5-minute accessibility probability map, we find a significant spatial mismatch: many grid cells assessed as high-risk fire zones have extremely low 5-minute connectivity probabilities predicted by the model. These areas are often located entirely within service blind spots. Model analysis shows that although these areas may be physically close to fire stations, the lack of effective "hidden high-order bridges" or the influence of high-frequency traffic flow disturbances make it difficult to establish stable rescue channels within the golden rescue time. This structural contradiction is the most prominent safety hazard in urban fringe areas, meaning that the areas most likely to experience major fires are often also the weakest points in rescue response. Local administrators can adjust fire deployments in urban areas in advance based on the results of coupling analysis, thereby reducing the occurrence of fire accidents.

[0141] In summary, the method of this embodiment has, but is not limited to, the following beneficial effects: (1) The method in this embodiment introduces dynamic graph learning and state-space model into the field of fire protection planning, breaking through the lag of traditional GIS static analysis in capturing time-varying features. By revealing the entangled spatiotemporal effects between road network structure and fire risk, it expands the application boundaries of deep learning technology in urban infrastructure resilience assessment, and provides new algorithmic support and theoretical paradigm for urban and rural fire protection planning in complex dynamic environments.

[0142] (2) The dynamic evaluation model constructed by the method in this embodiment can directly serve the precise scheduling and forward-looking layout of fire-fighting resources. By performing dynamic link prediction on fire stations and risk areas, the model can intuitively present the blind spots of rescue services under different time windows (such as traffic tidal periods), and accurately identify the hidden shortcomings of poor accessibility due to dynamic road conditions despite close physical distances. This decision support based on dynamic prediction provides quantitative guidance with practical value for the site selection and patrol route optimization of micro fire stations in peripheral urban areas, which helps to significantly shorten the rescue response time, effectively reduce casualties and property losses caused by rescue delays, and promote the intelligent and refined upgrading of regional safety governance capabilities.

[0143] Please see Figure 15 This application also provides a fire risk assessment system for outlying urban areas, which can implement the above-mentioned method. The system includes: The dynamic graph construction module is used to construct a dynamic graph model of the target urban area based on the road network of the target urban area. The dynamic graph model includes several grid nodes and the interaction edges between the grid nodes. The types of grid nodes include demand points and fire stations. The data acquisition module is used to acquire static risk assessment data of the target urban area and its traffic situation data at the time of query. The static risk assessment data includes data on disaster-causing factors, exposure factors, and disaster mitigation factors related to fire risk. The traffic situation data is acquired based on a preset window. The preset window records historical traffic interaction events between grid nodes within a preset time capacity before the time of query based on a time elimination mechanism. The module calculates the node feature sequence of each grid node based on the dynamic graph model of the target urban area, the static risk assessment data, and the traffic situation data at the time of query. The risk assessment module is used to perform multi-frequency decomposition on the node feature sequence of each grid node based on a preset recursive deduction model, and analyze the fire spatiotemporal risk of each grid node based on the multi-frequency decomposition results to obtain a comprehensive risk characterization of each grid node. The multi-frequency decomposition results include frequency components at several different frequency levels. Based on the comprehensive risk characterization of each grid node, dynamic fire risk assessment and fire accessibility assessment are performed on each grid node. Based on the results of the dynamic fire risk assessment and fire accessibility assessment of each grid node, a coupling analysis is performed, and the fire risk assessment results of the target urban area are determined based on the coupling analysis results of each grid node.

[0144] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0145] Please see Figure 16 This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0146] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0147] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0148] This application provides a method, system, and electronic equipment for fire risk assessment in peripheral urban areas. The scheme first constructs a dynamic graph model of the target urban area based on its road network. The model includes several grid nodes and interaction edges between these nodes, serving as the foundation for urban risk assessment. Then, it acquires static risk assessment data related to fire risk in the target urban area and its traffic situation data at the query time. Based on the dynamic graph model and the acquired data, it constructs a node feature sequence for each grid node, which serves as input to a recursive deduction model. In this method, the traffic situation data is recorded through a preset window, using a time-elimination mechanism to record historical traffic interaction events occurring within a preset time capacity before the query time. This data recording method allows for real-time updates. Traditional GIS or AHP models are mostly based on "snapshot" static data, while this method introduces real-time updated traffic situation data, improving the ability to capture spatiotemporal entanglement. It simultaneously considers static road network data, risk characteristic data, and dynamic traffic situation data, thereby effectively improving the comprehensiveness and accuracy of urban fire risk assessment. Furthermore, the recursive deduction model performs... This method employs multi-frequency decomposition, analyzing frequency components at several different frequency levels to determine the comprehensive risk characterization of grid nodes. By analyzing frequency components at different frequency levels, the model can analyze the evolution of risks at different frequency levels, thus more accurately determining the comprehensive risk characterization of grid nodes. Finally, dynamic fire risk assessment and fire accessibility assessment of the nodes are performed based on their comprehensive risk characterization. Coupled analysis of the two assessment results yields the final fire risk assessment result for the target urban area. By coupling the dynamic fire risk assessment results and fire accessibility assessment results, this method can analyze the fire risk of grid nodes based on the correspondence between different assessment results, thereby improving the effectiveness of the fire risk assessment result for the grid node. Furthermore, compared to central urban areas, peripheral urban areas may have lower levels of fire protection and transportation infrastructure. Therefore, introducing real-time traffic interaction data into the fire risk assessment process of peripheral urban areas and considering the correspondence between dynamic fire risk and fire accessibility assessment results can effectively improve the effectiveness of the fire risk assessment results for peripheral urban areas.

[0149] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of this application, and do not constitute a limitation on the technical solutions provided in this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in this application are also applicable to similar technical problems.

[0150] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0151] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0152] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0153] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0154] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0155] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0156] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0157] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A fire risk assessment method for peripheral urban areas, characterized in that, The method includes the following steps: A dynamic graph model of the target urban area is constructed based on the road network of the target urban area; the dynamic graph model includes several grid nodes and interaction edges between grid nodes; the types of grid nodes include demand points and fire stations; The static risk assessment data of the target urban area and its traffic situation data at the query time are obtained. The static risk assessment data includes data on disaster-causing factors, exposure factors, and disaster mitigation factors related to fire risk. The traffic situation data is obtained based on a preset window. The preset window records historical traffic interaction events between grid nodes within a preset time capacity before the query time based on a time elimination mechanism. The node feature sequence of each grid node is obtained by calculating based on the dynamic graph model of the target urban area, the static risk assessment data, and the traffic situation data at the query time. Based on a pre-defined recursive deduction model, the node feature sequence of each grid node is decomposed using multiple frequencies. The spatiotemporal fire risk of each grid node is analyzed based on the multiple frequency decomposition results to obtain a comprehensive risk characterization of each grid node. The multiple frequency decomposition results include frequency components at several different frequency levels. Dynamic fire risk assessment and fire accessibility assessment are performed on each grid node based on its comprehensive risk characterization. A coupling analysis is then conducted based on the results of the dynamic fire risk assessment and fire accessibility assessment of each grid node. Finally, the fire risk assessment result of the target urban area is determined based on the coupling analysis results of each grid node.

2. The method according to claim 1, characterized in that, The node feature sequence of the target grid node in the plurality of grid nodes is obtained in the following manner: The static risk assessment data of the target urban area is standardized, and the standardized data results are mapped to several grid nodes. The static feature vector of each grid node is obtained based on the mapping results. Based on the dynamic graph model of the target urban area and its traffic situation data at the query time, the neighbor node sequence and interaction edge sequence of each grid node are determined; the neighbor node sequence includes several neighbor nodes that have historical traffic interaction events with the grid node; the interaction edge sequence indicates the interaction edges between the grid node and its neighbor nodes; the attributes of the interaction edges are determined by the traffic path planning between the nodes. Construct several node pairs corresponding to the target grid node; the node pairs indicate the correspondence between the target grid node and the fire stations in the several grid nodes; based on the neighboring node sequence of the target grid node and its corresponding fire station in each node pair, analyze to obtain a multidimensional statistical vector of each node pair; The structural feature sequence of the target grid node is obtained by encoding and fusing the multidimensional statistical vectors of several node pairs; The original feature sequence of the target grid node is constructed based on the static feature vectors of the neighboring nodes in the neighboring node sequence and the attributes of the interactive edges in the interactive edge sequence. Based on the structural feature sequence of the target grid node and the original feature sequence, feature fusion processing is performed to obtain the node feature sequence of the target grid node.

3. The method according to claim 2, characterized in that, The analysis of the neighboring node sequence of the target grid node and its corresponding fire station in each node pair yields a multidimensional statistical vector for each node pair, including: The first feature is obtained by counting the total number of neighboring nodes in the neighboring node sequence of the target grid node; The second feature is obtained by statistically analyzing the frequency of occurrence of each neighboring node in the neighboring node sequence of the target grid node in the neighboring node sequence of the corresponding fire station. The third feature is obtained by counting the number of intersection nodes between the neighboring node sequences of the target grid node and the neighboring node sequences of the corresponding fire station. The frequency of occurrence of the corresponding fire station in the neighboring node sequence of the target grid node is counted to obtain the fourth feature; the frequency of occurrence of the target grid node in the neighboring node sequence of the corresponding fire station is counted to obtain the fifth feature; the product of the fourth feature and the fifth feature is calculated to obtain the sixth feature. A multidimensional statistical vector of the node pair is constructed based on the first feature, the second feature, the third feature, the fourth feature, the fifth feature, and the sixth feature.

4. The method according to claim 2, characterized in that, The comprehensive risk characterization of the target grid node among several grid nodes is obtained through the following method: Calculate the time difference between the time of the historical traffic interaction event between the target grid node and its neighboring nodes and the query time, and map the calculation result through sinusoidal position coding to obtain the time interval sequence of the target grid node; By performing multi-frequency decomposition based on the node feature sequence and time interval sequence of the target grid node through a recursive residual network, several layers of frequency components at different frequency levels and the time feature components corresponding to each layer of frequency components are obtained. The discretization step size corresponding to the frequency component of each layer is generated based on the time feature component corresponding to the frequency component of each layer. Based on the preset equation, the frequency components of each layer and their corresponding discretization step size, the state evolution modeling of the frequency components of each layer is performed to obtain the state evolution modeling result of the frequency components of each layer; A bottom-up strategy is used to fuse the state evolution modeling results of several layers of frequency components to obtain a comprehensive risk characterization of the target grid node.

5. The method according to claim 4, characterized in that, The process involves performing state evolution modeling on each layer of frequency components based on a preset equation, the frequency components at each layer, and their corresponding discretization step size, to obtain the state evolution modeling results for each layer of frequency components, including: The preset state transition matrix is ​​adjusted based on the discretization step size corresponding to each frequency component in each layer, and the state transition matrix corresponding to each frequency component in each layer is determined according to the adjustment result; the discretization input projection matrix corresponding to each frequency component in each layer is calculated based on the structural feature sequence of the target grid node, the preset state transition matrix, and the discretization step size corresponding to each frequency component in each layer. Based on the preset equations, the frequency components of each layer and their corresponding state transition matrices and discretized input projection matrices, the hidden state vectors corresponding to the frequency components of each layer are calculated. An output projection matrix is ​​constructed based on the structural feature sequence of the target grid nodes; the state evolution modeling result of each frequency component is obtained by calculating based on the output projection matrix and the hidden state vector corresponding to each frequency component.

6. The method according to claim 1, characterized in that, The fire accessibility assessment process for target grid nodes among several grid nodes is achieved through the following steps: Construct several joint node pairs corresponding to the target grid node; the joint node pairs indicate the correspondence between the grid node and the corresponding fire stations in the several grid nodes; Based on the comprehensive risk characterization of the target grid nodes and their corresponding fire stations in each joint node pair, a joint risk characterization corresponding to each joint node pair is constructed. Each joint node is input into a preset fire accessibility prediction model for calculation, and the fire accessibility probability between the target grid node and each corresponding fire station is determined based on the calculation results. The fire accessibility assessment result of the target grid node is obtained by analyzing the fire accessibility probability between the target grid node and each corresponding fire station.

7. The method according to claim 6, characterized in that, The joint risk representation for each joint node pair is constructed based on the comprehensive risk representation of the target grid node and its corresponding fire station in each joint node pair, including: The comprehensive risk representation of the target grid nodes and their corresponding fire stations in the joint node pair is compressed based on the weighted aggregation mechanism to obtain the node embedding vector of the target grid nodes and their corresponding fire stations in the joint node pair. By concatenating the node embedding vectors of the target grid node and its corresponding fire station in the joint node pair, the joint risk representation corresponding to the joint node pair is obtained.

8. The method according to claim 1, characterized in that, The dynamic fire risk assessment results of the grid nodes include a dynamic fire risk index obtained by mapping the comprehensive risk characterization of the grid nodes based on a preset multilayer sensor. The fire accessibility assessment results indicate the fire accessibility probability between grid nodes and each corresponding fire station in the dynamic graph model; The coupled analysis based on the results of dynamic fire risk assessment and fire accessibility assessment for each grid node includes: The fire accessibility probability between each corresponding fire station in the dynamic graph model and the grid node is analyzed to determine the fire accessibility probability between the corresponding fire station that meets the preset requirements and the grid node. The dynamic fire risk index of the grid node is judged according to the preset fire risk judgment threshold to obtain the first judgment result; The fire accessibility probability between the corresponding fire station that meets the preset requirements and the grid node is judged according to the preset accessibility judgment threshold to obtain a second judgment result; The coupling analysis results of the grid nodes are obtained by calculating based on the preset coupling discrimination function, the first discrimination result, and the second discrimination result.

9. A fire risk assessment system for peripheral urban areas, characterized in that, The system includes: A dynamic graph construction module is used to construct a dynamic graph model of the target urban area based on the road network of the target urban area; the dynamic graph model includes several grid nodes and interaction edges between grid nodes; the types of grid nodes include demand points and fire stations; The data acquisition module is used to acquire static risk assessment data of the target urban area and its traffic situation data at the query time; the static risk assessment data includes data on disaster-causing factors, exposure factors, and disaster mitigation factors related to fire risk; the traffic situation data is acquired based on a preset window; the preset window records historical traffic interaction events between grid nodes within a preset time capacity before the query time based on a time elimination mechanism; and the node feature sequence of each grid node is obtained by calculation based on the dynamic graph model of the target urban area, the static risk assessment data, and the traffic situation data at the query time. The risk assessment module is used to perform multi-frequency decomposition on the node feature sequence of each grid node based on a preset recursive deduction model, and analyze the fire spatiotemporal risk of each grid node based on the multi-frequency decomposition results to obtain a comprehensive risk characterization of each grid node. The multi-frequency decomposition results include frequency components at several different frequency levels. Based on the comprehensive risk characterization of each grid node, dynamic fire risk assessment and fire accessibility assessment are performed on each grid node. Coupled analysis is performed based on the results of the dynamic fire risk assessment and fire accessibility assessment of each grid node, and the fire risk assessment result of the target urban area is determined based on the coupled analysis results of each grid node.

10. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 8.

Images