Urban road network flood risk assessment method based on Bayesian and complex network modeling

Abstract

The invention provides an urban road network flood risk assessment method based on Bayesian and complex network modeling, and relates to the technical field of flood disaster prevention and reduction. Model parameters are optimized by analyzing and researching spatial heterogeneity distribution of historical flood in a region, and flood susceptibility is predicted based on the optimized Bayesian network model; carrying out modeling analysis on the urban road network topological structure by adopting a complex network theory framework to obtain the road network vulnerability of the research region; and comprehensively scoring the flood risk according to the flood susceptibility and the road network vulnerability, and obtaining the dynamic change of the comprehensive score of the flood risk by adjusting the model and the road network structure. According to the method, the Bayesian network and the Domirank algorithm are combined, the complex causal relationship between each influence factor and the flood disaster is deeply analyzed and quantified, and the capability of coping with the flood disaster of a city can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of flood disaster prevention and mitigation, and in particular to a method for assessing urban road network flood risk based on Bayesian and complex network modeling. Background Art

[0002] Climate change and urbanization are not only reshaping the physical mechanisms of extreme precipitation and flooding, but are also exacerbating the dynamic complexity of urban flooding and the unpredictability of transportation network disruptions. While traditional disaster risk assessment systems have developed diverse approaches, including indicator-based comprehensive evaluations, flood risk monitoring using remote sensing and geographic information systems (GIS), and flood dynamics modeling, these methods still lack quantitative accuracy and often overlook the potential causal relationships between multiple hazard-causing factors.

[0003] Against this backdrop, obtaining continuous, high-precision historical flood data has become a major challenge for urban flood risk prevention and control. With the widespread adoption of social media data, which contains vast amounts of user-generated content, it is becoming an emerging data resource for disaster management. Research has shown that leveraging social media data to rapidly extract real-time disaster information is particularly effective and practical for small, economically developed, and densely populated urban areas. Furthermore, modern urban infrastructure networks are highly interconnected and interdependent. Transportation networks, as the critical backbone of urban operations, are deeply coupled with subsystems such as water supply, power supply, and communications. Road flooding and traffic disruptions caused by floods not only directly restrict the flow of people and the allocation of supplies, but also easily trigger cascading failures. Therefore, to build a more proactive disaster response mechanism and enhance the resilience of social systems and transportation networks to disasters, there is an urgent need to establish a systematic flood risk assessment framework based on social media data and develop analytical techniques specifically for urban road transportation systems. This framework should be able to accurately quantify flood risk levels under multiple scenarios, providing scientific and reliable decision-making support for urban transportation management departments.

[0004] The paper "A Dynamic Bayesian Network Assessment Method for Urban Stormwater Flooding Resilience" (Publication No. CN118607192A) introduces dynamic Bayesian networks (DBNs) to the field of urban stormwater flooding resilience assessment. By constructing a multidimensional, time-varying resilience indicator system and a dynamic probabilistic reasoning framework, it quantitatively analyzes the resilience evolution process and identifies critical thresholds throughout the entire lifecycle, from disaster onset to recovery. However, the model's overreliance on idealized data assumptions and static parameter configurations limits the applicability of the assessment results in real, complex urban systems. Furthermore, the method fails to consider the impact and risks of stormwater flooding on transportation networks.

[0005] "A Method for Assessing Urban Flood Disaster Levels" (Publication No. CN117575319A) addresses the challenges of data integration and simulation prediction in urban flood disaster assessment by using digital twin technology to construct a flood simulation model and integrate multi-source information (such as topographic information, meteorological forecasts, and drainage systems). However, this method still has some shortcomings: 1. The simulation model may not fully reproduce the complex factors of real-world transportation networks. For example, the impact of dynamic changes on the transportation network is not fully considered; 2. The simulation process lacks consideration of dynamic factors (such as real-time traffic flow and emergencies), which have a significant impact on transportation network vulnerability and emergency response strategies.

[0006] In general, the current flood risk assessment methods for urban transportation nodes have the following deficiencies: (1) they fail to deeply characterize the complex nonlinear relationship between influencing factors and flood disasters; (2) they lack comprehensive integration of key information such as transportation networks and spatial dimension data such as terrain and meteorology; and (3) they lack quantitative analysis of the vulnerability of transportation networks. Summary of the Invention

[0007] In view of this, the present invention proposes an urban road network flood risk assessment method based on Bayesian and complex network modeling to solve the problems existing in the above-mentioned prior art.

[0008] To achieve the above objectives, the present invention proposes an urban road network flood risk assessment method based on Bayesian and complex network modeling, which is characterized by comprising: Obtain flood-related social media data and build a flood database; Obtain the spatial heterogeneity distribution of historical floods in the study area based on the flood database; A Bayesian network model was constructed based on the influencing factor data in the flood database; Testing and verifying the Bayesian network model and optimizing its parameters based on the influencing factors and the spatial heterogeneity distribution, and calculating the probability of flood occurrence based on the optimized Bayesian network model; Constructing the urban road network topology, modeling and analyzing the urban road network topology based on the complex network theory framework, and obtaining the road network vulnerability of the study area; Obtaining a comprehensive risk score based on the flood occurrence probability and the road network vulnerability; The Bayesian network model is adjusted by adjusting the influencing factors, and the topological structure of the urban road network is adjusted by simulating the changes in traffic node connections under the road network change scenario, so as to generate the latest comprehensive risk score and evaluate the dynamic changes in flood risk.

[0009] Furthermore, the flood database includes social media data, regional terrain data, meteorological data, river network data, underlying surface coverage data, influencing factor data, socioeconomic data and transportation network data; the influencing factor data include elevation, slope, proximity to water bodies, river density, road density, impervious area, population density, GDP density, average annual heavy rain intensity and average annual heavy rain frequency, and the transportation network data includes road length, road flow and road grade.

[0010] Furthermore, the process of obtaining the spatial heterogeneity distribution of historical floods in the study area based on the flood database includes: using a deep learning classification model to perform text classification and data cleaning on the social media data to obtain historical urban flood inundation data and waterlogging records; The historical urban flood inundation data is matched with database data based on a spatial position association algorithm to obtain inundation longitude and latitude coordinates, and the spatial heterogeneity distribution of the historical flood is obtained based on the inundation longitude and latitude coordinates.

[0011] Furthermore, while using a deep learning classification model to perform text classification on the social media data, a manual verification strategy is used to conduct a secondary verification of the text retained after classification. During the verification process, the report text published by official media is retained, and some user texts with reference value are used as supplements.

[0012] Furthermore, the Bayesian network model includes 10 evidence nodes and 1 target node. The evidence nodes include direct influencing factors and indirect influencing factors. The direct influencing factors include elevation, slope, river density, road density, impervious area, and average annual heavy rainfall intensity. The indirect influencing factors include water body proximity, population density, GDP density, and average annual heavy rainfall frequency. The target node is set to represent the state of flood occurrence and no flood occurrence.

[0013] Furthermore, the process of testing, verifying and optimizing the parameters of the Bayesian network model includes: sampling according to the geographical stratification of the study area, adopting a stratified random sampling strategy, and matching flood and non-flood samples in a 1:1 ratio. Based on the matched samples, the AUC-ROC index is used to evaluate the model performance; based on the evaluation results, combined with the influencing factors and the spatial heterogeneity of historical floods, the conditional discrete rules and state classification thresholds of the model are parameterized and corrected until the model error is lower than the threshold.

[0014] Furthermore, the process of constructing the urban road network topology structure includes: filtering the original road network data by attributes, extracting the road section set and key intersection nodes that conform to the dynamic characteristics of urban flood propagation, constructing a weighted directed topology structure, and constructing the urban road network topology structure with the road section length, flow, and type as edge weight parameters.

[0015] Furthermore, the process of modeling and analyzing the urban road network topology structure based on the complex network theory framework includes: Taking road section length as the spatial constraint term, type as the functional weight term, and traffic flow as the dynamic disturbance term, a three-dimensional weighted adjacency matrix was constructed. The Domirank centrality index was used to evaluate the vulnerability of the urban road network topology. The random walk model was used to reflect the global influence of nodes in the network and to obtain the key vulnerable nodes in the road network.

[0016] Furthermore, the method for obtaining the comprehensive risk score is as follows:

[0017] in, For nodes The comprehensive risk score of For nodes The Domirank centrality value of For nodes PageRank centrality, For nodes probability of flood susceptibility.

[0018] Compared with the prior art, the present invention has the following beneficial effects: This paper adopts a multi-source data fusion strategy to integrate multiple data sources such as social media data, terrain data, meteorological data, and socioeconomic data, and uses deep learning technology to accurately screen flooded street data, successfully constructing a high-precision flood inundation event dataset, providing a solid and reliable data foundation for subsequent analysis.

[0019] This paper innovatively combines Bayesian networks with the Domirank algorithm to deeply analyze and quantify the complex causal relationships between various influencing factors and flood disasters. Based on this, it accurately calculates the flood susceptibility and vulnerability of nodes in Wuhan's transportation network, providing a highly targeted and refined decision-making tool for urban flood resilience planning, helping to enhance the city's ability to cope with flood disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings: Figure 1 This is a flow chart of the urban road network flood risk assessment method based on Bayesian and complex network modeling of the present invention; Figure 2 A schematic diagram of the Bayesian network model structure constructed for an embodiment of the present invention; Figure 3 This is the flood susceptibility assessment result of the flood risk assessment conducted in Jianghan District, Wuhan City, according to Example 2 of the present invention; Figure 4 This is the road network vulnerability assessment result of flood risk assessment conducted in Jianghan District, Wuhan City, in Example 2 of the present invention; Figure 5 This is the comprehensive risk level classification result of flood risk assessment conducted in Example 2 of the present invention with Jianghan District of Wuhan City as the research area. DETAILED DESCRIPTION

[0021] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, unless there is a conflict, the embodiments of the present disclosure and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0022] Example 1 This embodiment proposes a method for assessing urban road network flood risk based on Bayesian and complex network modeling. Figure 1 As shown, including: (1) Obtain flood-related social media data, regional topographic data, meteorological data, river network data, underlying surface coverage data, socioeconomic data, and transportation network data to establish a database; (2) Based on social media data, we used the BERT deep learning method to perform text classification, and further manually inspected and cleaned the data to extract historical urban flooding and waterlogging records. By matching addresses, we obtained the longitude and latitude information of the waterlogging points. (3) Screen the influencing factors of floods and waterlogging, and construct a topological network based on the causal relationship between the influencing factors and flood disasters; discretize and classify the influencing factors based on expert experience; use the discretized flooding data and the index data of each influencing factor to drive the Bayesian network model to carry out model training and parameter learning, infer the conditional probability distribution of each node in the Bayesian network in different states, and construct a complete Bayesian network model; (4) Test and verify the Bayesian network model using an independent test set, and adjust the model parameters based on the test results until the model performance meets the standards; calculate the regional flood susceptibility probability based on the verified Bayesian network model; (5) Screening traffic network data and extracting intersection nodes to construct a road network topology structure; (6) Based on the framework of complex network theory, topological modeling and network characteristic analysis of the traffic network are carried out, taking into account the length, type and flow of roads, to identify key vulnerable nodes in the road network that are prone to global paralysis; (7) Conduct flood risk assessment of the transportation network based on the flood susceptibility and network vulnerability of the road network; (8) Conduct scenario analysis on model parameters, simulate and deduce the probability of flooding and network vulnerability of the transportation network under different scenarios such as precipitation and road network structure, and evaluate the risk changes of transportation nodes.

[0023] In the above technical solution, in step (1), the flood disaster social media data is obtained by the following methods: using web crawler technology to crawl data through the public access interface of the Weibo platform to obtain blog information related to rainstorms and floods; the influencing factor data covers multiple dimensions of flood-influencing factor data such as elevation, slope, proximity to water bodies, river density, road density, impervious area, population density, GDP density, average annual heavy rain intensity and average annual heavy rain frequency; and the traffic network data includes road length, road flow and road grade.

[0024] In step (2), the historical flood extraction operation for Weibo text data is as follows: construct and train a deep learning classification model based on the BERT architecture, use the model to classify and process Weibo text data, and accurately extract blog information directly related to rainstorm and flood disasters; at the same time, use a manual verification strategy to conduct a secondary verification of the Weibo content retained after screening by the deep learning model. During the verification process, focus on retaining authoritative and accurate reporting blog posts released by official media, and appropriately supplement a small number of user blog posts with reference value, and summarize and organize the flood information; finally, match the summarized and organized flood information with the point of interest (POI) database, and obtain the latitude and longitude geographic coordinates corresponding to the flood information through a matching operation based on the spatial location association algorithm.

[0025] In step (3), the Bayesian network model constructed in this embodiment is as follows Figure 2 As shown in the figure, it contains 10 evidence nodes and 1 target node. Among the evidence nodes, 6 factors (annual average heavy rainfall intensity, elevation, slope, river density, impervious area, and road density) are direct factors affecting floods, and the remaining 4 factors (annual average heavy rainfall frequency, water body proximity, population density, and GDP density) indirectly affect floods by directly affecting other factors. The status of the target node is set to yes and no, indicating the occurrence of floods and no floods respectively. The calculation of parameter learning is based on the Bayesian theorem in probability theory. Flood susceptibility probability Depends on a set of parent nodes : Then the joint probability based on the Bayesian network can be written as:

[0026] According to the chain rule of Bayesian networks, the joint probability of the parent nodes can be decomposed into:

[0027] The final formula is:

[0028] in, Represents conditional probability, driven by historical data.

[0029] In step (4), a stratified random sampling strategy was implemented, based on the geographical stratification of the study area to ensure spatial uniformity of the samples and avoid local overload. At the same time, flood and non-flood samples were matched at a 1:1 ratio. The Bayesian network model was tested with independent samples, and the AUC-ROC index (a comprehensive indicator of true / false positive rate under the classification threshold) was introduced to quantitatively evaluate the model performance and ensure the reliability of the risk assessment results. Based on the test feedback, the conditional discrete rules and state classification thresholds were parameterized and modified based on the spatial heterogeneity of the influencing factors and historical flood events, reducing the verification error to an acceptable range. At the same time, the AUC-ROC test result exceeded the 0.9 threshold. After the parameter adjustment, the model can effectively support the assessment of urban flood susceptibility.

[0030] In step (5), the original road network data is preprocessed by attribute filtering (such as eliminating bridges, elevated roads, etc.), and then a set of road sections and key intersection nodes that meet the dynamic characteristics of urban flood propagation are extracted, and then a weighted directed topological structure is constructed. The road section length, flow rate, and type are used as edge weight parameters to form a refined road topology network that supports the simulation of the spatiotemporal evolution of floods.

[0031] In step (6), the road length (L) is used as the spatial constraint term, the type (T) as the functional weight term, and the traffic (Q) as the dynamic disturbance term to construct a three-dimensional weighted adjacency matrix. The Domirank centrality index is used to assess the network vulnerability of road network flooding. At the same time, the random walk model PageRank is used to reflect the global influence of the node in the network. When constructing the traffic network graph, the weight of the edge is calculated by combining the road length, road traffic, and road level:

[0032] in: Represents an edge The weight of Indicates the length of the road, represents the road traffic volume, Indicates the road level, : The weight coefficient of each indicator, satisfying .

[0033] Hypothesis Diagram represents a network where is a collection of nodes, is the edge set, the adjacency matrix Represents the structure of the graph, where Representation node and nodes The connection weight between nodes. is a vector representing the centrality value of each node, is a parameter that controls the convergence of the algorithm, is the identity matrix, is a vector of all ones.

[0034] Kinetic equation:

[0035] Analytical method to obtain the Domirank centrality value of nodes :

[0036] Based on the random walk PageRank model, it reflects the global influence of a node in the network. The formula is:

[0037] in Points to the node The neighbor set of is a node The out-degree.

[0038] In step (7), the road network vulnerability is combined with flood susceptibility to generate a comprehensive risk score, where For nodes The comprehensive risk score of For nodes The Domirank centrality value of For nodes PageRank centrality, For nodes The flood susceptibility probability of is calculated as follows:

[0039] In step (8), based on the constructed Bayesian network model, different scenarios of factors such as precipitation and impervious surface are set, and new node evidence is set to infer the change in the probability of flood susceptibility; in addition, the dynamic changes in the connectivity of traffic nodes under road network change scenarios (such as road network planning adjustment, emergency repair temporary access, etc.) are simulated, the topological structure of the road network is adjusted, and the change in road network vulnerability is evaluated; the changes in flood susceptibility and road network vulnerability under different scenarios are comprehensively considered, and the dynamic changes in flood risk are evaluated, breaking through the limitation of the separation of topological structure and risk in traditional static analysis.

[0040] Example 2 This example uses Wuhan as the research area and uses multi-source data fusion to comprehensively analyze the factors affecting flooding in Wuhan. It also combines the Bayesian network model and complex network modeling methods to achieve a dynamic assessment of flood risk. Specifically, the following steps are included: Step 1: Use crawler technology to crawl Weibo blog post data from 2012 to 2023, and collect topographic, socioeconomic, meteorological, and traffic network data.

[0041] Step 2: Screen and clean social media data. First, use the BERT deep learning classification model to classify and process Weibo text data, accurately extracting post information directly related to rainstorm and flood disasters. Then, based on the principle of "news reports as a supplement, and general blog posts as a supplement", further clean the data, match the summarized inundation information with the point of interest (POI) database, and obtain the latitude and longitude geographic coordinates corresponding to the inundation information through a matching operation based on a spatial location association algorithm.

[0042] Step 3: Based on the triggering relationship between influencing factors and flood disasters, a topological network is constructed. The discrete inundation data extracted from social media and the influencing factor data are used to drive model parameter learning, obtain the conditional probability table, and construct a Bayesian network model.

[0043] In step 4, the 1572 samples were divided into a training set (n=1372) and a validation set (n=200) using a stratified random sampling method. The sampling process was based on the spatial stratification standard of the study area to ensure that the sample proportions of each sub-region were consistent with the overall distribution. The geographic coordinate distribution of the sample points was further verified through spatial analysis to ensure the spatial discreteness of the samples. The excessive aggregation of samples in a certain area of ​​the city was avoided, and a 1:1 balanced ratio of flood data to non-flood data was ensured. Finally, the AUC-ROC index was introduced to quantitatively evaluate the performance of the model, and the result was AUC-ROC=0.95, indicating that the model performance met the requirements. The Bayesian network model after verification was used for flood prediction, and the probability of flood susceptibility was output, as shown below: Figure 3 shown.

[0044] In step 5, a set of road sections and key intersection nodes that meet the dynamic characteristics of urban flood propagation are extracted in ArcGIS 10.4, and then a weighted directed topological structure is constructed. The road section length, flow rate, and type are used as edge weight parameters to form a refined topological network that supports the simulation of the spatiotemporal evolution of floods.

[0045] Step 6: Taking road length (L) as the spatial constraint term, type (T) as the functional weight term, and flow (Q) as the dynamic disturbance term, a three-dimensional weighted adjacency matrix is ​​constructed. The Domirank centrality index is used to evaluate the network vulnerability of road network flooding. The evaluation results are as follows: Figure 4 As shown, at the same time, based on the random walk model PageRank, it reflects the global influence of the node in the network. When constructing the traffic network graph, the weight of the edge is calculated by combining the road length, road flow and road level:

[0046] in: Represents an edge The weight of Indicates the length of the road, represents the road traffic volume, Indicates the road level, : The weight coefficient of each indicator, satisfying .

[0047] Hypothesis Diagram represents a network where is a collection of nodes, is the edge set, the adjacency matrix Represents the structure of the graph, where Representation node and nodes The connection weight between nodes. is a vector representing the centrality value of each node, is a parameter that controls the convergence of the algorithm, is the identity matrix, is a vector of all ones.

[0048] Kinetic equation:

[0049] Analytical method to obtain the Domirank centrality value of nodes :

[0050] Based on the random walk PageRank model, it reflects the global influence of a node in the network. The formula is:

[0051] in Points to the node The neighbor set of is a node The out-degree.

[0052] Step 7: Combine road network vulnerability with flood susceptibility to generate a comprehensive risk score, calculated as follows:

[0053] in For nodes The comprehensive risk score of For nodes The Domirank centrality value of For nodes PageRank centrality, For nodes probability of flood susceptibility.

[0054] The flood risk level of the node is determined based on the comprehensive risk score. The results are as follows: Figure 5 shown.

[0055] Step 8: Set new node evidence in the Bayesian network model to infer the probability of flood susceptibility. For example, set the state of the impervious surface node to "high" and the heavy rain intensity to "very_high" to obtain the corresponding flood susceptibility probability. In addition, simulate the dynamic changes in the connectivity of traffic nodes under different extreme scenarios (such as heavy rain flooding cross-river channels and the activation of temporary emergency repair access roads), adjust the topological structure of the road network, and break through the limitations of the separation of topological structure and risk in traditional static analysis.

[0056] The contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. An urban road network flood risk assessment method based on Bayesian and complex network modeling, characterized by: include: Obtain flood-related social media data and build a flood database; Obtain the spatial heterogeneity distribution of historical floods in the study area based on the flood database; A Bayesian network model was constructed based on the influencing factor data in the flood database; Testing and verifying the Bayesian network model and optimizing its parameters based on the influencing factors and the spatial heterogeneity distribution, and calculating the probability of flood occurrence based on the optimized Bayesian network model; Constructing the urban road network topology, modeling and analyzing the urban road network topology based on the complex network theory framework, and obtaining the road network vulnerability of the study area; Obtaining a comprehensive risk score based on the flood occurrence probability and the road network vulnerability; The Bayesian network model is adjusted by adjusting the influencing factors, and the topological structure of the urban road network is adjusted by simulating the changes in traffic node connections under the road network change scenario, so as to generate the latest comprehensive risk score and evaluate the dynamic changes in flood risk.

2. The method according to claim 1, characterized in that The flood database includes social media data, regional terrain data, meteorological data, river network data, underlying surface coverage data, influencing factor data, socioeconomic data and transportation network data; the influencing factor data includes elevation, slope, proximity to water bodies, river density, road density, impervious area, population density, GDP density, average annual heavy rain intensity and average annual heavy rain frequency; the transportation network data includes road length, road flow and road grade.

3. The method according to claim 1, characterized in that The process of obtaining the spatial heterogeneity distribution of historical floods in the study area based on the flood database includes: using a deep learning classification model to perform text classification and data cleaning on the social media data to obtain historical urban flood inundation data and waterlogging records; The historical urban flood inundation data is matched with database data based on a spatial position association algorithm to obtain inundation longitude and latitude coordinates, and the spatial heterogeneity distribution of the historical flood is obtained based on the inundation longitude and latitude coordinates.

4. The method according to claim 3, characterized in that While using a deep learning classification model to perform text classification on the social media data, a manual verification strategy is used to conduct a secondary verification of the text retained after classification. During the verification process, the report texts released by official media are retained, and some user texts with reference value are used as supplements.

5. The method according to claim 1, wherein The Bayesian network model includes 10 evidence nodes and 1 target node. The evidence nodes include direct influencing factors and indirect influencing factors. The direct influencing factors include elevation, slope, river density, road density, impervious area, and average annual heavy rainfall intensity. The indirect influencing factors include water body proximity, population density, GDP density, and average annual heavy rainfall frequency. The target node is set to represent the state of flood occurrence and non-flood occurrence.

6. The method according to claim 1, wherein The process of testing, verifying, and optimizing the parameters of the Bayesian network model includes: sampling according to the geographical stratification of the study area, adopting a stratified random sampling strategy, and matching flood and non-flood samples in a 1:1 ratio. Based on the matched samples, the AUC-ROC index is used to evaluate the model performance; based on the evaluation results, combined with the influencing factors and the spatial heterogeneity of historical floods, the conditional discrete rules and state classification thresholds of the model are parameterized and modified until the model error is lower than the threshold.

7. The method according to claim 1, characterized in that The process of constructing the urban road network topology structure includes: filtering the original road network data by attributes, extracting the road section set and key intersection nodes that conform to the urban flood propagation dynamics characteristics, constructing a weighted directed topology structure, and constructing the urban road network topology structure using the road section length, flow, and type as edge weight parameters.

8. The method according to claim 1, characterized in that The process of modeling and analyzing the urban road network topology based on the complex network theory framework includes: Taking road section length as the spatial constraint term, type as the functional weight term, and traffic flow as the dynamic disturbance term, a three-dimensional weighted adjacency matrix was constructed. The Domirank centrality index was used to evaluate the vulnerability of the urban road network topology. The random walk model was used to reflect the global influence of nodes in the network and to obtain the key vulnerable nodes in the road network.

9. The method according to claim 1, characterized in that The method for obtaining the comprehensive risk score is as follows: ， in, For nodes The comprehensive risk score of For nodes The Domirank centrality value of For nodes PageRank centrality, For nodes probability of flood susceptibility.

Images