Coastal city storm surge disaster toughness evaluation method based on graph neural network

Abstract

The invention discloses a coastal city storm surge disaster toughness evaluation method based on a graph neural network, and the method comprises the steps: collecting the multi-source heterogeneous data, including geographic information data, infrastructure data and historical storm surge data, of a coastal city related to a storm surge disaster; constructing coastal city infrastructure network structure data according to the geographic information data and the infrastructure data, and then constructing a graph neural network model; fusing characteristic parameters of historical storm surge data with coastal city infrastructure network structure data, inputting the fused data into the trained graph neural network model, and simulating step-by-step influences of storm surge disasters on coastal city infrastructure nodes; based on a graph neural network propagation result, carrying out toughness evaluation on state evolution of each node in the coastal city infrastructure network under the influence of storm surge; and based on the obtained toughness evaluation index data of each node, summarizing and analyzing different functional areas to form an overall coastal city toughness evaluation result.

Description

Technical Field

[0001] The present invention belongs to the technical field of meteorological disaster risk prediction and assessment on coastlines, and specifically relates to a method for assessing the resilience of coastal cities to storm surge disasters based on graph neural networks. Background Art

[0002] With the intensification of global climate change, frequent storm surges have become a serious threat to public safety and the operational stability of infrastructure systems in coastal cities. Storm surges often lead to cascading failures in multiple systems, including damage to coastal seawalls, disruption of port operations, and increased flooding in residential areas. This in turn leads to reduced urban resilience and difficulties in post-disaster recovery. Therefore, modeling resilience to storm surge hazards, analyzing impact propagation, and optimizing assessments based on complex urban infrastructure networks has become a research hotspot in coastal disaster prevention and mitigation.

[0003] Traditional urban resilience assessment methods typically rely on expert experience or static indicator evaluation systems, which struggle to reflect the dynamic interactions between infrastructure and the impact of disasters. These methods also lack the versatility and predictive power of these models. Therefore, there is an urgent need to design an efficient and accurate resilience assessment solution for coastal cities in storm surge scenarios. Summary of the Invention

[0004] In order to overcome the shortcomings of the existing technology, the present invention provides a storm surge disaster resilience assessment method for coastal cities based on graph neural networks.

[0005] The present invention is achieved through the following technical solutions: A method for assessing the resilience of coastal cities to storm surge disasters based on graph neural networks includes the following steps: Step 1: Collect multi-source heterogeneous data related to storm surge disasters in coastal cities, including geographic information data, infrastructure data, and historical storm surge data; Step 2: Preprocess the data collected in step 1; Step 3: Construct coastal city infrastructure network structure data based on geographic information data and infrastructure data; Step 4: Build a graph neural network model based on the constructed coastal city infrastructure network structure data; Step 5: After integrating the characteristic parameters of historical storm surge data with the coastal city infrastructure network structure data, the data is input into the trained graph neural network model. Based on the cascading impact propagation mechanism of the graph neural network model, the step-by-step impact of storm surge disasters on coastal city infrastructure nodes is simulated to obtain the dynamic changes in the time series state of each node in the coastal city infrastructure network under the impact of storm surge. Step 6: Based on the graph neural network propagation results of step 5, conduct a resilience assessment on the state evolution of each node in the coastal city infrastructure network under the influence of storm surges; based on the resilience assessment index data obtained for each node, summarize and analyze different functional areas separately to form the overall coastal city resilience assessment results.

[0006] In the above technical solution, in step 1, geographic information data: including high-resolution urban topography, landform data and coastline shape; infrastructure data: including port facility layout, seawall structure parameters and residential area distribution data; historical storm surge data: including the occurrence time, tide height, path trajectory, duration, wind speed and direction at landing and central air pressure value of historical storm surges covering many years.

[0007] In the above technical solution, in step 2, preprocessing includes: outlier removal, missing value filling, format unification and standardization.

[0008] In the above technical solution, in step 3, entities such as ports, seawalls, and residential areas are abstracted as nodes, and edge relationships are established based on their spatial proximity or functional associations to construct a coastal city infrastructure network diagram, in which: residential areas are connected by roads; edges are established between ports and the infrastructure that supports their transportation; and edges are established between seawalls and the areas affected by their protection.

[0009] In the above technical solution, step 4 includes: Step 4.1: First define the node and edge features; Step 4.2: Using a graph neural network model based on a supervised graph sparsification mechanism, we introduce label information to guide edge sampling for the coastal city infrastructure network structure, and construct a task-related sparse subgraph. The model design includes the following steps: ① Edge probability learning: By introducing structural priors and label supervision information, the importance of each edge is learned; ② Sparse subgraph sampling: Based on the edge probability distribution, sparse subgraphs are generated, retaining the key structures that have the greatest impact on the evaluation results; ③ Graph neural network training: Node representation learning and resilience prediction are performed on the sparse graph structure. In the above technical solution, in step 4.1, a distinctive feature vector is defined for each type of node, including: residential area nodes: building density, total population, average building age, and shelter coverage; port nodes: annual throughput, cargo types, and facility disaster resistance level; seawall nodes: seawall length, height, structure type, service life, and maintenance cycle; edge features include: connection strength, connection reliability, and response time.

[0010] In the above technical solution, in step 4.2, for edge probability learning: combined with node feature embedding, a multi-layer perceptron (MLP) is used to extract the edge relevance score.

[0011] In the above technical solution, in step 4.2, for sparse subgraph sampling: in order to make the sampling process supported by gradient optimization, Gumbel noise is introduced to achieve approximate discrete sampling, and Softmax normalization is introduced to construct a differentiable sparse edge sampling distribution, and combined with the temperature annealing mechanism and temperature parameter Control the smoothness of the distribution to achieve a dynamic balance between exploration and determinism.

[0012] In the above technical solution, in step 4.2, for graph neural network training: the METIS graph partitioning algorithm is used to divide the original large graph into several local subgraphs to achieve distributed batch training and effectively reduce memory overhead; the overall training process is divided into three stages: teacher model training, student model distillation learning, and inference deployment.

[0013] In the above technical solution, in step 6, the resilience evaluation indicators of each node include: Node recovery time: refers to the time required for node functions to recover from a damaged state to a normal state; Node function loss rate: refers to the proportion of node function that decreases during the disaster impact period; Node connectivity interruption time: refers to the duration of interruption of functions such as communication or transportation between nodes caused by cascading impact.

[0014] The advantages and beneficial effects of the present invention are: The present invention can effectively simulate the cascading impact process of complex infrastructure networks in coastal cities under disaster impacts, quantify system resilience indicators, and generate optimized disaster prevention strategies.

[0015] This paper proposes a graph neural network model based on a supervised graph sparsification mechanism: it uses label information to guide the edge sampling process, dynamically generates sparse subgraphs related to the prediction task from the full graph, retains key structural information while reducing computational overhead, and adapts to the needs of large-scale heterogeneous graph modeling.

[0016] The present invention constructs the probability distribution of edge existence, and accurately estimates the retention importance of each edge in the evaluation task by fusing node embedding features and introducing degree priors.

[0017] This paper proposes a sparse subgraph sampling method that combines Gumbel noise with a temperature annealing mechanism: while retaining the differentiability of the model, a controllable exploration-exploitation balance is achieved, so that the sampling results are both diverse and close to the optimal structure.

[0018] This paper proposes a label attention distillation mechanism: constructing a teacher-student framework, introducing graph-level label embedding, and guiding node representations to align with graph-level labels through the attention mechanism, significantly improving the semantic consistency of node representations in classification tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flow chart of the method for assessing the resilience of coastal cities to storm surge disasters based on graph neural networks of the present invention.

[0020] For ordinary technicians in this field, other relevant drawings can be obtained based on the above drawings without any creative work. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention are further described below with reference to specific embodiments.

[0022] This paper designs a storm surge disaster resilience assessment method for coastal cities based on graph neural network, see the attached Figure 1 , the specific steps are as follows.

[0023] Step 1: Collect multi-source heterogeneous data related to storm surge disasters in coastal cities. Specifically, it includes: Geographic information data: including high-resolution urban topography, landform data and coastline shapes.

[0024] Infrastructure data: including port facility layout (such as terminal location, number of berths and throughput capacity), seawall structural parameters (such as height, width, material, and designed wave resistance level), and residential area distribution (such as building density, population density, building type, and construction year).

[0025] Historical storm surge data: including the occurrence time, tide height, path trajectory, duration, wind speed and direction at landfall, and central pressure values of historical storm surges covering many years.

[0026] The above data sources include government geographic information databases, real-time data platforms of ocean monitoring stations, meteorological department records and urban planning archives.

[0027] Step 2: Preprocess the data collected in step 1.

[0028] Specifically, the preprocessing includes: Outlier Removal: Identify and remove noisy data caused by equipment failure or transmission errors.

[0029] Missing value filling: Use linear interpolation, spline interpolation or K-nearest neighbor algorithms to fill in the missing values.

[0030] Format unification: Convert data from different sources and different coordinate systems into a unified format and perform coordinate calibration.

[0031] Standardization: Normalize all types of data to the range of [0,1] for subsequent model training.

[0032] Step 3: Based on geographic information data and infrastructure data, construct infrastructure network structure data of coastal cities.

[0033] Specifically, based on geographic information and infrastructure data, entities such as ports, seawalls, and residential areas are abstracted as nodes in a graph. Edge relationships are established based on their spatial proximity or functional associations to construct a coastal urban infrastructure network diagram. Residential areas are connected by roads; edges are established between ports and the infrastructure (roads and railways) that support their transportation; and edges are established between seawalls and the areas they protect.

[0034] Edge attributes include connection strength (such as traffic flow), reliability (such as wind and wave resistance), response time (such as the shortest travel time from residential areas to shelters), etc. The resulting graph is stored in the form of an adjacency matrix and edge list to meet the needs of graph neural network modeling.

[0035] Step 4: Construct a graph neural network (GNN) model based on the constructed coastal city infrastructure network structure data.

[0036] Step 4.1: Define node and edge features.

[0037] Define a discriminative feature vector for each type of node, including but not limited to: 1. Residential area nodes: building density, total population, average building age, and shelter coverage; 2. Port nodes: annual throughput, cargo types, and facility disaster resistance rating; 3. Seawall nodes: seawall length, height, structure type, service life, maintenance cycle, etc.

[0038] Edge features include: connection strength, connection reliability, response time, etc.

[0039] The above features constitute the input of the graph neural network model, which is used to characterize the interaction between nodes and structural fragility.

[0040] Step 4.2: Model selection and design.

[0041] A graph neural network model based on a supervised graph sparsification mechanism is adopted. For the highly heterogeneous coastal urban infrastructure network structure, label information is introduced to conduct guided sampling of edges, thereby constructing task-related sparse subgraphs, which can effectively reduce the model's computational complexity and improve prediction performance.

[0042] Model design includes the following core processes: 1. Edge probability learning: By introducing structural priors and label supervision information, the importance of each edge is learned; 2. Sparse subgraph sampling: Generate sparse subgraphs based on edge probability distribution, retaining the key structures that have the greatest impact on the evaluation results; 3. Graph neural network training: Node representation learning and resilience prediction on sparse graph structures.

[0043] This closed-loop structure implements the adaptive graph modeling mechanism of "structure learning-efficient modeling-performance optimization", which is suitable for processing large-scale, low-homogeneity complex urban network structures.

[0044] Specifically, step 4.2 includes the following steps: Step 4.2.1: Perform edge probability learning.

[0045] The probability distribution of edges used for graph sparsification in the storm surge resilience assessment task is constructed. The importance of each edge retained in constructing the sparse subgraph is estimated through the characteristic relationship between nodes.

[0046] Combined with node feature embedding, a multi-layer perceptron (MLP) is used to extract the edge relevance score. , its corresponding node and The embedding vector of and , calculate the edge weight by fusing features through concatenation and element-wise multiplication : ; , :respectively nodes and Embedding vector of : Vector splicing operation; : vector element-wise multiplication; : Activation function (such as Sigmoid) to ensure that edge weights fall in the interval [0,1]; : Multilayer perceptron, used to model nonlinear feature interactions.

[0047] To prevent high-connectivity nodes from occupying all sampled edges, we introduce a prior distribution based on node degree to guide the model to pay more attention to the connection relationship between low-degree nodes and avoid missing key connections: ; :node and The degree (i.e., the number of connected edges); :side The prior sampling tendency of is used to encourage the preservation of potentially critical but sparse connections in edge structures.

[0048] : A relationship that is proportional to .

[0049] Step 4.2.2: Perform sparse subgraph sampling.

[0050] This step is based on the probability distribution of edges , sampling sparse subgraphs that preserve key structures from the original infrastructure graph , in order to reduce the computational complexity of the graph neural network and improve the generalization performance of the model in storm surge scenarios. (e.g. 20%) Sample a fixed number of edges to form a subgraph structure: ; : sparse retention ratio; : the total number of edges in the original graph; : The actual number of sampled edges.

[0051] To make the sampling process supportable by gradient optimization, Gumbel noise is introduced to achieve approximate discrete sampling. To construct a differentiable sparse edge sampling distribution, Softmax normalization is introduced. The temperature annealing mechanism and temperature parameter are combined to control the smoothness of the distribution, thereby achieving a dynamic balance between exploration and determinism: ; :side The normalized probability of ; :side Gumbel noise to improve the randomness of sampling; : The logarithm of the edge weight, which helps to enhance numerical stability; : the set of all edges in the graph; : Temperature parameter, controlling the smoothness of the distribution ( Approaching hard sampling, approaching a uniform distribution).

[0052] In order to make the model training more exploratory in the early stage and more deterministic in the later stage, a temperature annealing mechanism is introduced. The formula is: ; ; : initial temperature; : minimum temperature; : Annealing rate, controls the temperature drop rate; : Current training round number; max_epochs: total number of rounds; : Maximum value operation.

[0053] Through theoretical analysis and empirical verification, we ensure that the sampled sparse subgraph retains the key edges of the original graph structure while not destroying the connectivity between key paths and nodes, and has a high structural similarity with the "ideal subgraph".

[0054] Step 4.2.3: Graph neural network training.

[0055] This step aims to perform node classification based on the sampled sparse subgraphs and optimize model parameters to achieve efficient learning and representation of graph structure information. Considering the scalability of large graph processing, the METIS graph partitioning algorithm is used to divide the original large graph into several local subgraphs, enabling distributed batch training and effectively reducing memory overhead.

[0056] To improve the model's discriminative performance in graph classification tasks, this step introduces a label attention distillation mechanism, integrating the label guidance mechanism with the distillation learning strategy to effectively alleviate the problem of insufficient alignment between node embeddings and graph-level labels in traditional GNNs. The overall training process is divided into three stages: teacher model training, student model distillation learning, and inference deployment. The details are as follows: Step 4.2.3.1: Teacher model training.

[0057] The GNN backbone network adopts mainstream GNN architectures such as GCN and GAT. The GNN layer update formula is: ; : No. Layer Node Embedding vector of :node The set of neighbors of : Neighbor aggregation function (such as mean, attention mechanism); : Node update function (such as MLP + activation function); : The set of all nodes in the graph.

[0058] Through the adjacency matrix and node features Construct the initial node embedding: ; : adjacency matrix of the graph; : node feature matrix (initial features); : learnable parameters of the GNN backbone; : The overall mapping function of the GNN backbone, outputting the node embedding matrix .

[0059] The label attention distillation mechanism is introduced to build the ideal embedding. The label attention encoder uses the MLP network to distill the image labels. Embed and obtain the label embedding vector: ; :picture The label embedding vector of :picture The unique hot label; : Label encoder (such as MLP), which maps labels to latent space; : graph dataset.

[0060] With the help of Transformer-style attention mechanism, node embedding and label embedding are fused to obtain the ideal node representation : ; ; ; Label embedding is projected into the query vector; :Node embedding is projected into Key and Value vectors; : Learnable projection matrix; : Attention temperature coefficient, controlling the sharpness of the Softmax distribution (adjusting the exploration-exploitation balance); : Layer normalization to alleviate internal covariate shift; : Feedforward neural network, introducing nonlinear transformation; : The ideal node embedding generated by the teacher model (the result after fusing label information).

[0061] Step 4.2.3.2: Student model distillation learning.

[0062] The student model adopts the same GNN backbone structure and classification head as the teacher, inheriting the label-guided node knowledge to ensure that the embedding is learned in the same semantic space.

[0063] During the training process, the teacher model parameters remain fixed, and the student model is minimized through backpropagation. , learn high-quality node embeddings with label guidance from the teacher model. Loss function Using multi-loss joint optimization: ; Weight parameter, used to weigh the proportion of each loss; : Homogeneity loss encourages a higher probability of edge connection between nodes with the same label, thereby strengthening the consistency of the intra-class structure. The formula is: ; Indicator function, 1 if the condition is met, otherwise 0; The set of all edges in the graph; :side The edge weight of For the set of labeled nodes, only nodes with labels are counted Edges between labels, penalizing the low probability of edges with the same label; :node and The true category label of :when If the labels are the same, it is 1, indicating that only nodes of the same type are considered; : Consistency loss, which forces the weights of structural edges to be consistent with the semantic similarity between node embeddings (such as cosine similarity), thereby enhancing the coupling between the structure and features of the graph: ; : No. Layer Node cosine similarity of embeddings; :respectively nodes and Embedding vector of : Cross entropy loss function for graph-level label classification, measuring classification error: ; : number of training samples; : No. The true label of each image (0 or 1); :Teacher model for The predicted probability of each graph; : Distillation loss, which measures the difference between the teacher and student embeddings by minimizing the mean squared error loss between the student model and the teacher's ideal embedding: ; :The first The node embedding matrix of the graph; :The first The node embedding matrix of the graph; : Euclidean distance squared (i.e., mean square error).

[0064] Furthermore, during the inference phase, the trained student model is used to perform the image classification task. Because the student model inherits the label perception capabilities of the teacher model through the distillation process, it can achieve accurate classification while maintaining low computational costs. The prediction process includes: 1. Input subgraph , is the adjacency matrix, is the node feature matrix; 2. GNN backbone generates node embeddings , The node embedding matrix generated for the student model; 3. Generate graph-level representation: ; : graph-level representation vector; : The set of all nodes in the graph; : Pooling function (such as mean pooling, maximum pooling or attention pooling); :node Embedding vector of 4. The classification head outputs the class probability: ; : Graph the predicted probability distribution belonging to each category; : weight matrix of the classification head; : bias term of the classification head; 5. Finally, the one with the highest probability is selected as the prediction result: ; : total number of categories; : Indicates the predicted probability that the graph belongs to the cth class; : The final image category predicted by the model.

[0065] Step 5: After fusing the characteristic parameters of historical storm surge data with the coastal city infrastructure network structure data, the data are input into the trained graph neural network model. Based on the cascading impact propagation mechanism of the graph neural network model, the step-by-step impact of storm surge disasters on coastal city infrastructure nodes is simulated to obtain the dynamic changes in the time series state of each node in the coastal city infrastructure network under the impact of storm surges.

[0066] Step 5.1: Input disaster scenario simulation data.

[0067] In this step, the characteristic parameters of historical storm surge data (storm surge occurrence time, tide height, path trajectory, duration, wind speed and direction at landfall, and central pressure) are integrated with the coastal city infrastructure network structure data established in Step 3. The data is then fed into the graph neural network model trained in Step 4 to construct storm surge impact scenarios. By inputting multiple storm surge intensity levels and path combinations, the system response behavior of coastal cities under different storm surge scenarios can be simulated.

[0068] Step 5.2: Calculation of the cascade shock propagation mechanism based on the graph neural network model (GNN).

[0069] A specific information propagation mechanism built into a graph neural network model is employed to simulate the cascading impact of storm surge hazards on various types of urban infrastructure nodes. In the model's initial state, the seawall node, acting as a coastal defense barrier, first receives the external disturbance input of the storm surge. The extent of damage to the seawall node is determined by both its characteristics (such as structural type and wave resistance rating) and storm surge parameters. The model then transmits this damage information to downstream functional nodes, such as port nodes and residential areas, through an edge propagation mechanism. Each node updates its state characteristics based on its own attributes (such as the port's storage and transportation capacity or the building's disaster resistance rating in a residential area), and further propagates the impact to other connected nodes, forming a cascading chain of effects. Through multiple rounds of graph information propagation iterations, the dynamic changes in the temporal state of each node in the coastal urban infrastructure network under the impact of the storm surge are obtained, characterizing the cascading damage process of the system.

[0070] Step 6: Based on the graph neural network propagation results in step 5, conduct a resilience assessment on the state evolution of each node in the coastal urban infrastructure network under the influence of storm surges.

[0071] The resilience evaluation indicators of each node include but are not limited to the following categories: Node recovery time: refers to the time required for node functions to recover from a damaged state to a normal state; Node function loss rate: refers to the proportion of node function that decreases during the disaster impact period; Node connectivity interruption time: refers to the duration of interruption of functions such as communication or transportation between nodes caused by cascading impact.

[0072] Based on the resilience assessment indicators of the above-mentioned nodes, different functional areas (such as residential areas, ports, transportation hubs, etc.) are summarized and analyzed to form the overall urban resilience assessment results.

[0073] Furthermore, based on the evaluation results of step 6, an optimization strategy for urban resilience can be designed.

[0074] As a preferred approach, based on the evaluation results from step 6, a multi-objective intelligent optimization algorithm was used to construct a search model for strategies to enhance urban storm surge resilience. Optimization methods with global search capabilities, such as genetic algorithms and particle swarm optimization, were selected to generate the optimal resilience strategy combination while considering resource constraints (such as fiscal budgets and spatial limitations). Using residential areas as an example, policy variables such as strengthening building structures, increasing emergency shelter capacity, and optimizing drainage systems were encoded, and objective functions (such as minimizing mean recovery time and reducing functional loss rate) were set. Through an iterative search, the optimal strategy output within the constraints was obtained.

[0075] The optimization process incorporates realistic feasibility constraints, including implementation cycle requirements and land use compatibility, to ensure the executable nature of the generated results in real-world scenarios. The final output is a set of engineering-feasible resilience enhancement solutions, providing technical support for proactive defense and rapid recovery in coastal cities under high-risk storm surge scenarios.

[0076] The above is an exemplary description of the present invention. It should be noted that, without departing from the core of the present invention, any simple deformation, modification or other equivalent replacement that can be made by other skilled in the art without expending creative labor falls within the scope of protection of the present invention.

Claims

1. A method for assessing the resilience of coastal cities to storm surge disasters based on graph neural networks, characterized by: The following steps are involved: Step 1: Collect multi-source heterogeneous data related to storm surge disasters in coastal cities, including geographic information data, infrastructure data, and historical storm surge data; Step 2: Preprocess the data collected in step 1; Step 3: Construct coastal city infrastructure network structure data based on geographic information data and infrastructure data; Step 4: Build a graph neural network model based on the constructed coastal city infrastructure network structure data; Step 5: After integrating the characteristic parameters of historical storm surge data with the coastal city infrastructure network structure data, the data is input into the trained graph neural network model. Based on the cascading impact propagation mechanism of the graph neural network model, the step-by-step impact of storm surge disasters on coastal city infrastructure nodes is simulated to obtain the dynamic changes in the time series state of each node in the coastal city infrastructure network under the impact of storm surge. Step 6: Based on the graph neural network propagation results of step 5, conduct a resilience assessment on the state evolution of each node in the coastal city infrastructure network under the influence of storm surges; based on the resilience assessment index data obtained for each node, summarize and analyze different functional areas separately to form the overall coastal city resilience assessment results.

2. The method for assessing storm surge disaster resilience in coastal cities based on graph neural networks according to claim 1 is characterized by: In step 1, geographic information data: including high-resolution urban topography, landform data and coastline shape; infrastructure data: including port facility layout, seawall structure parameters and residential area distribution data; historical storm surge data: including the occurrence time, tide height, path trajectory, duration, wind speed and direction at landfall and central pressure value of historical storm surges covering many years.

3. The method for assessing storm surge disaster resilience in coastal cities based on graph neural networks according to claim 1 is characterized by: In step 2, preprocessing includes: outlier removal, missing value filling, format unification and standardization.

4. The method for assessing coastal city storm surge disaster resilience based on graph neural network according to claim 1 is characterized by: In step 3, the ports, seawalls, and residential areas are abstracted as nodes, and edge relationships are established based on their spatial proximity or functional associations to construct a coastal city infrastructure network diagram, in which: residential areas are connected by roads; edges are established between ports and the infrastructure that supports their transportation; and edges are established between seawalls and the areas affected by their protection.

5. The method for assessing storm surge disaster resilience in coastal cities based on graph neural networks according to claim 1 is characterized by: Step 4 includes: Step 4.1: First define the node and edge features; Step 4.2: Using a graph neural network model based on a supervised graph sparsification mechanism, we introduce label information to guide edge sampling for the infrastructure network structure of coastal cities and construct task-related sparse subgraphs. The model design includes the following processes: ① Edge probability learning: By introducing structural priors and label supervision information, the importance of each edge is learned; ② Sparse subgraph sampling: Sparse subgraphs are generated based on edge probability distribution, retaining the key structures that have the greatest impact on the evaluation results; ③ Graph neural network training: Node representation learning and resilience prediction are performed on the sparse graph structure.

6. The method for assessing storm surge disaster resilience in coastal cities based on graph neural networks according to claim 5 is characterized by: In step 4.1, a distinguishing feature vector is defined for each type of node, including: residential area nodes: building density, total population, average building age, and shelter coverage; port nodes: annual throughput, cargo types, and facility disaster resistance level; seawall nodes: seawall length, height, structure type, service life, and maintenance cycle; edge features include: connection strength, connection reliability, and response time.

7. The method for assessing coastal city storm surge disaster resilience based on graph neural network according to claim 5 is characterized by: In step 4.2, for edge probability learning: combined with node feature embedding, a multi-layer perceptron is used to extract the edge relevance score.

8. The method for assessing storm surge disaster resilience in coastal cities based on graph neural networks according to claim 5 is characterized by: In step 4.2, for sparse subgraph sampling: In order to make the sampling process supported by gradient optimization, Gumbel noise is introduced to achieve approximate discrete sampling. In order to construct a differentiable sparse edge sampling distribution, Softmax normalization is introduced, and the temperature annealing mechanism and temperature parameter are combined. Control the smoothness of the distribution to achieve a dynamic balance between exploration and certainty.

9. The method for assessing storm surge disaster resilience in coastal cities based on graph neural networks according to claim 5, characterized in that: In step 4.2, for graph neural network training: the METIS graph partitioning algorithm is used to divide the original large graph into several local subgraphs to achieve distributed batch training; the overall training process is divided into three stages: teacher model training, student model distillation learning, and inference deployment.

10. The method for assessing storm surge disaster resilience in coastal cities based on graph neural networks according to claim 1, characterized in that: In step 6, the resilience evaluation indicators of each node include: Node recovery time: refers to the time required for node functions to recover from a damaged state to a normal state; Node function loss rate: refers to the proportion of node function that decreases during the disaster impact period; Node connectivity interruption time: refers to the duration of interruption of communication or transportation functions between nodes caused by cascading impact.

Images