Method for constructing urban building flood vulnerability curves considering typhoon cascade disasters

Abstract

This application relates to the field of disaster risk early warning and big data processing technology, specifically disclosing a method for constructing urban building flood vulnerability curves considering typhoon cascading disasters. First, it constructs a multi-dimensional building prototype classification system encompassing modern features such as high-rise buildings, glass curtain wall systems, and underground spaces, and automatically quantifies the functional importance weights of each component through a multi-modal fusion inference network. Second, it introduces a spatial aerodynamic coupling topology diagram and a cascading failure propagation model, performing cascading amplification calculations of wind-induced failure probability based on wind directionality and the wind pressure transmission relationship between components, and replacing the traditional Boolean cliff-style threshold judgment with a soft logic aggregation mechanism. Based on this, using the overall building damage level as an intermediate bridge variable, it dynamically corrects the flood vulnerability parameters of each component and couples multi-source flood calculations to determine the equivalent inundation depth. Finally, it constructs a vulnerability curve with both pre-event prediction and post-event diagnosis capabilities through log-normal regression fitting and Bayesian back inference.

Description

Technical Field

[0001] This application relates to the field of disaster risk early warning and big data processing technology, and more specifically, to a method for constructing urban building flood vulnerability curves that take into account typhoon cascading disasters. Background Technology

[0002] Currently, scientifically quantifying the vulnerability characteristics of urban buildings under combined disasters and constructing building flood vulnerability curves that reflect the actual physical processes causing disasters plays an irreplaceable and fundamental role in the construction of urban disaster prevention and mitigation planning and resilience assessment systems. However, the impact of typhoon-heavy rainfall coupled disasters on urban buildings is not a simple superposition of the effects of each disaster. Damage to the building envelope caused by strong winds provides an efficient intrusion path for subsequent continuous heavy rain, causing rainwater to not only flood from the bottom up but also penetrate and seep into the building's internal components from the top down, causing severe damage far exceeding the expectations of a single disaster. This cascading amplification effect of wind-induced damage-rainwater intrusion-flood inundation makes it difficult for traditional methods to accurately characterize the true vulnerability of modern urban buildings under combined disasters.

[0003] The traditional vulnerability assessment methods widely used in the industry, represented by the HAZUS model, establish a single mapping relationship between flood disaster impacts based on a pre-set inundation depth-loss function. These methods rely heavily on post-disaster survey data, and the function varies significantly across different building types and regions, resulting in serious deficiencies in universality and transferability. Furthermore, these models are typically deterministic, making it difficult to effectively handle multi-source uncertainties such as disaster propagation intensity, building resistance, and construction quality standards. More critically, existing models generally employ Boolean cliff-like threshold aggregation logic in the overall assessment of wind-induced damage, treating each building envelope component as a physically isolated statistical node. The overall damage level is determined solely based on whether each component independently exceeds a pre-set damage rate threshold, completely ignoring the crucial spatial aerodynamic cascading interactions in fluid dynamics. In real-world scenarios… In strong wind disaster environments, the failure of building envelope components is by no means an isolated event. Once a window sash or curtain wall on the windward side suffers physical damage, a large amount of strong airflow will rush into the building's interior through the breach, inducing a surge in internal transient wind pressure. This superposition effect of internal positive pressure and external negative pressure will cause components on the lateral or leeward sides, which were originally in a safe stress state, to bear a multiplied load, thus triggering a domino effect of chain failures. However, existing methods sever the aerodynamic transmission link at the spatial level, which not only leads to a serious underestimation bias in risk assessment results, but also fails to provide scientific support for the continuous collapse early warning of building envelope structures. The resulting deviation in the overall damage level determination will continue to propagate downstream along the cascading transmission link to vulnerability correction and economic loss calculation, ultimately causing a systematic deviation between the constructed vulnerability curve and the building's actual response under combined disasters.

[0004] Therefore, an optimized scheme for constructing urban building flood vulnerability curves is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a method for constructing urban building flood vulnerability curves that consider typhoon cascading disasters.

[0006] According to one aspect of this application, a method for constructing urban building flood vulnerability curves considering typhoon cascading disasters is provided, comprising: S1: Perform prototype matching and classification on the building to be evaluated to obtain the building prototype category, and perform functional importance quantification inference on each building component to obtain the vulnerability parameter vector of each component. S2: Based on the physical wind speed and vulnerability parameter vector of the typhoon in the disaster scenario, calculate the wind-induced failure probability of each building envelope component, and perform logical analysis and level aggregation on the wind-induced failure probability of each component to obtain the wind-induced failure probability of each building envelope component and the overall damage level of the building. S3: Based on the overall damage level and vulnerability parameter vector of the building, the original flood vulnerability parameters of each component are dynamically corrected by cascading effect to obtain the corrected vulnerability parameters. The maximum effective flood depth of each component is calculated by combining the multi-source flooding caused by external backflow and internal breach intrusion. S4: Based on the corrected vulnerability parameters and the maximum effective flooding depth, calculate the cascade failure probability of each component, and combine the vulnerability parameter vector and functional importance weight to perform weighted loss aggregation on the cascade failure probability to obtain the weighted relative economic loss and comprehensive dataset. S5: Based on the disaster intensity and weighted relative economic loss in the comprehensive dataset, log-normal regression is performed on the discrete simulation lattice to obtain the building flood vulnerability curve.

[0007] Compared with existing technologies, this application provides a method for constructing urban building flood vulnerability curves considering typhoon cascading disasters. First, it constructs a multi-dimensional building prototype classification system encompassing modern features such as high-rise buildings, glass curtain wall systems, and underground spaces. It then automatically quantifies the functional importance weights of each component through a multi-modal fusion inference network, overcoming the shortcomings of outdated traditional prototype libraries and the strong subjectivity of manual assessments. Second, it introduces a spatial aerodynamic coupling topology diagram and a cascading failure propagation model. Based on wind directionality and the wind pressure transmission relationship between components, it performs cascading amplification calculations of wind-induced failure probabilities. Furthermore, it replaces the traditional Boolean cliff-style threshold judgment with a soft logic aggregation mechanism, fundamentally solving the problem of underestimating the overall damage level caused by isolated component modeling. On this basis, using the overall building damage level as an intermediate bridge variable, it dynamically corrects the flood vulnerability parameters of each component and couples multi-source flood calculations to determine the equivalent inundation depth, thus opening up the entire cascading transmission path from wind-induced damage to flood failure. Finally, it constructs a vulnerability curve with both pre-event prediction and post-event diagnosis capabilities through log-normal regression fitting and Bayesian back-inference. Attached Figure Description

[0008] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0009] Figure 1 This is a flowchart illustrating a method for constructing urban building flood vulnerability curves considering typhoon cascading disasters, according to an embodiment of this application. Figure 2 This is a data flow diagram illustrating the method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to an embodiment of this application. Figure 3 This is a flowchart of step S1 of the method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to an embodiment of this application; Figure 4 This is a flowchart of step S2 of the method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to an embodiment of this application; Figure 5 This is a flowchart of step S6 of the method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to an embodiment of this application.

[0010] Figure 6 This diagram illustrates a comparison of simulation results between the A-LSH method and the ordinary Monte Carlo method for constructing urban building flood vulnerability curves considering typhoon cascading disasters, according to embodiments of this application.

[0011] Figure 7 This is a schematic diagram illustrating the physical process and key parameter correction of typhoon cascading disasters in the method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to an embodiment of this application. Detailed Implementation

[0012] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0013] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0014] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0015] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0016] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0017] To address the core technical issues revealed in the background, such as the inability of traditional vulnerability assessment methods to quantify the cascading amplification effect among multiple hazards including typhoons, rainfall, and flooding; the systematic underestimation of the overall damage level in the assessment of wind-induced damage to building envelopes due to neglecting spatial aerodynamic cascading interactions; and the inability of existing building prototype databases to match the complex characteristics of modern urban buildings, this solution designs a complete process for constructing cascading disaster vulnerability curves. Specifically, the scheme first performs prototype dimensionality reduction matching based on the building's functional attributes, structural features, environmental location, and underground space attributes. The buildings to be evaluated are categorized into a prototype system encompassing modern features such as high-rise and super high-rise buildings, glass curtain walls, and multi-level underground spaces. For each instantiated building component, features are independently extracted through semantic encoding, spatial encoding, and graph neural network topology encoding. An attention fusion mechanism dynamically infers the functional importance weights of each component, forming a complete component-level vulnerability parameter vector. Subsequently, the typhoon's physical wind speed is normalized by wind pressure effects, and adaptive Latin hypercube sampling is used to calculate the wind-induced failure probability of each building envelope component. An anisotropic aerodynamic coupling topology graph is constructed to quantify the cascading failure relationships between components caused by wind pressure transmission. Through cascaded amplification calculations and soft logic aggregation, the overall building damage is assessed and output. The damage level is then used as a bridge to identify cascading effect correction factors from two dimensions: damage degree and feasibility of rainwater intrusion paths. The original flood vulnerability parameters of each component are dynamically scaled and corrected. At the same time, the maximum effective inundation depth of each component is calculated by comprehensively considering the multi-source flooding caused by external backflow and internal breach intrusion. The corrected parameters and equivalent inundation depth are substituted into the vulnerability integral formula to solve for the cascading failure probability. Finally, the component replacement cost and functional importance weight are combined to aggregate the weighted economic loss. A comprehensive dataset is generated through multi-scenario Monte Carlo simulation. The vulnerability curve is output by fitting the discrete simulation lattice with log-normal regression through maximum likelihood estimation. Based on Bayesian inference, the inverse probability mapping from economic loss to physical damage state is realized, thereby constructing a full-link modern urban building flood vulnerability assessment method.

[0018] Figure 1 This is a flowchart of a method for constructing urban building flood vulnerability curves considering typhoon cascading disasters, according to an embodiment of this application. Figure 2 A data flow diagram illustrating the method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to embodiments of this application is shown below. Figure 1 and Figure 2As shown, the method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to an embodiment of this application includes: S1, performing prototype matching and classification on the building to be evaluated to obtain building prototype categories, and performing functional importance quantification inference on each building component to obtain vulnerability parameter vectors for each component; S2, calculating the wind-induced failure probability of each building envelope component based on the typhoon physical wind speed and vulnerability parameter vectors in the disaster scenario, and performing logical analysis and level aggregation on the wind-induced failure probability of each component to obtain the wind-induced failure probability of each building envelope component and the overall damage level of the building; S3, based on the overall damage level of the building and the vulnerability parameter vectors, performing analysis on each component... The original flood vulnerability parameters are dynamically corrected for cascading effects to obtain the corrected vulnerability parameters. The maximum effective inundation depth of each component is calculated by integrating multiple sources of flooding, including external backflow and internal breach intrusion. S4: Based on the corrected vulnerability parameters and the maximum effective inundation depth, the cascading failure probability of each component is calculated. The cascading failure probability is then weighted and aggregated using the vulnerability parameter vector and functional importance weights to obtain the weighted relative economic loss and a comprehensive dataset. S5: Based on the disaster intensity and weighted relative economic loss in the comprehensive dataset, a log-normal regression is performed on the discrete simulation lattice using the maximum likelihood estimation algorithm to obtain the building flood vulnerability curve.

[0019] Specifically, in step S1, the building to be evaluated is classified using prototype matching to obtain the building prototype category, and the functional importance of each building component is quantitatively inferred to obtain the vulnerability parameter vector of each component. It should be understood that the vulnerability of urban buildings under typhoon cascading disasters is highly dependent on the building's structural form, functional layout, and the spatial distribution characteristics of its internal components. Modern urban high-rise and super high-rise buildings, large-area glass curtain wall systems, and multi-story underground spaces with complex functions are fundamentally different from traditional low-rise and mid-rise residential buildings. Different building prototypes exhibit drastically different failure modes and loss paths under the same disaster intensity. Furthermore, the contribution of each internal component to the overall functional maintenance also varies significantly. Without differentiation and quantification, it is impossible to accurately characterize the true vulnerability response of each component under cascading disasters. Therefore, in the technical solution of this application, the building to be evaluated is classified by prototype matching based on its functional attributes, structural characteristics, environmental location, and underground space attributes. A multimodal fusion inference network is then used to quantify the functional importance of each instantiated building component. This provides a precise component-level parameterized input basis for subsequent calculations of wind-induced failure probability, dynamic correction of cascading effects, and weighted economic loss aggregation. In this way, the building to be evaluated can be reduced from a macroscopic overall structure to a computable component level, ensuring that each building component carries complete type identification, spatial location, disaster resistance parameters, economic value, and functional weight information, laying a data foundation for end-to-end cascading vulnerability assessment.

[0020] Figure 3 This is a flowchart of step S1 of the method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to an embodiment of this application. Figure 3 As shown in the embodiment of this application, step S1 includes: S11, performing dimensionality reduction matching and component instantiation on the functional attributes, structural features, environmental location, and underground space attributes of the building to be evaluated to obtain the building prototype category and initial component information set; S12, performing multimodal feature independent encoding extraction on the initial component information set to obtain semantic feature vector, spatial feature vector, and topological feature vector; S13, performing feature fusion and weight inference based on attention mechanism on the semantic feature vector, spatial feature vector, and topological feature vector to obtain the functional importance weight of each component; S14, performing high-dimensional concatenation on the type, spatial location, vulnerability parameter set, and reset cost of each component to obtain the vulnerability parameter vector of each component.

[0021] Accordingly, in step S11, the functional attributes, structural features, environmental location, and underground space attributes of the building to be evaluated are subjected to building prototype dimensionality reduction matching and component instantiation to obtain the building prototype category and initial component information set. It should be understood that, due to the extremely complex and diverse forms of modern urban buildings, traditional vulnerability assessments rely excessively on outdated and simplistic building databases, resulting in an inability to accurately reflect the damage patterns and defense shortcomings of buildings with modern characteristics such as high-rise buildings, large-area glass curtain walls, and multi-level underground spaces in complex disasters. Therefore, in the technical solution of this application, by performing building prototype dimensionality reduction matching and component instantiation on the functional attributes, structural features, environmental location, and underground space attributes of the building to be evaluated to obtain the building prototype category and initial component information set, a systematic standard mapping and parameter deconstruction of external macroscopic structural features and internal microscopic physical composition is achieved. This ensures that the evaluation model accurately corresponds to the real physical form of modern urban buildings, constructing a unified and complete underlying data foundation for subsequent multimodal feature extraction and disaster cascade effect simulation.

[0022] Specifically, in this embodiment, the system first receives the original engineering data of the building to be evaluated and extracts functional attributes, structural features, environmental location, and underground space attributes as independent input variables. Functional attributes are used to classify the building's macro-use scenarios, thereby defining the distribution density of internal assets and the expected level of functional importance. Structural features directly determine the aerodynamic response threshold of the building under extreme strong wind loads and the distribution pattern of vulnerable nodes on the windward side. Environmental location anchors the external disaster-causing physical environment in which the building is located, characterizing the benchmark wind pressure setpoint and benchmark hydrological backflow conditions for flooding. Underground space attributes specifically characterize the most vulnerable underground flood control blind spots in modern urban buildings to accurately assess the runoff path and deep flooding risk after rainwater intrusion. Subsequently, the variables of the above four dimensions are substituted into a preset building prototype matching function. A = f ( F, S, E, U The function performs dimensionality reduction calculations, using rule matching or table lookup mapping to reduce the dimensionality of the four-dimensional attribute combination to a specific prototype label, outputting a unique corresponding building prototype category. Here, A represents the building prototype category; The system uses the following parameters: F represents the building prototype matching function; S represents the structural characteristics of the building; E represents the environmental location of the building; and U represents the underground space of the building. Through this dimensionality reduction mapping process, the system can eliminate computational interference from redundant building features under multidimensional data constraints, reducing highly complex real-world building units to standard building prototype categories that reflect the commonalities of disaster degradation. After obtaining a clear building prototype category, the system retrieves a standard component configuration template from the database that matches that category. Based on this structured template, the system performs a traversal instantiation operation on each physical component within the building to be evaluated, extracting the independent attributes of each component from the building information model library. These attributes include the text description of the component name, its spatial coordinates in the building's three-dimensional coordinate system, its wind-induced vulnerability parameter set, its original flood vulnerability parameter set, and its replacement cost. The system integrates and merges all the instantiated building component data, ultimately outputting an initial component information set covering all building elements.

[0023] More specifically, in a particular example of this application, the building to be evaluated is a large commercial complex located in a coastal strong wind zone. The calculation system first extracts key attributes from the building's electronic engineering files, defining its functional attribute as commercial office space, its structural features as a 20-story frame structure with a fully covered glass curtain wall system, its environmental location as a 7-degree wind-resistance zone and its corresponding low-lying survey elevation point, and its underground space attribute as a two-story underground centralized parking garage. Subsequently, the system inputs the extracted functional attributes, structural features, environmental location, and underground space attributes as core independent variables into a building prototype matching function for mapping calculation. The output building prototype category is a comprehensive template identifier containing the features of commercial offices, a high-rise glass curtain wall, 7-degree wind resistance, and a two-story basement. Next, the system delves into the interior of the commercial complex to perform component instantiation operations, specifically generating independent component entities including glass curtain wall units on the exterior of the first to twentieth floors, various ventilation and exhaust vents, indoor office network equipment, and various models of cars parked on the second basement level. The system accurately records key data for each component, including its specific type name, the specific floor and spatial coordinates, the set of wind-induced vulnerability parameters, the set of original flood vulnerability parameters, and the replacement cost. All independently generated component data objects are then written into the data bus to obtain an initial set of component information for subsequent feature extraction modules.

[0024] Accordingly, in step S12, multimodal feature independent encoding is performed on the initial component information set to obtain semantic feature vectors, spatial feature vectors, and topological feature vectors. It should be understood that, since the component attribute data contained in the initial component information set has significant heterogeneous multimodal characteristics—textual descriptions are discrete natural language symbol sequences, spatial location coordinates are continuous numerical geometric data, and the physical connections between components are non-Euclidean graph topological data—these three types of information respectively carry vulnerability features of different dimensions, such as the inherent semantic attributes of components, spatial exposure risks, and disaster cascading propagation paths. Using a single encoder for unified processing would inevitably lead to semantic confusion and feature annihilation among heterogeneous information. Therefore, in the technical solution of this application, multimodal feature independent encoding is further performed on the initial component information set to obtain semantic feature vectors, spatial feature vectors, and topological feature vectors, thereby capturing the inherent semantic attributes, spatial risk exposure features, and physical cascading propagation features of components from three orthogonal dimensions. This ensures that each type of heterogeneous information is fully expressed in its most suitable feature space without causing cross-modal interference, providing high-quality multi-dimensional feature input for subsequent attention fusion mechanisms.

[0025] Specifically, in this embodiment of the application, multimodal feature independent encoding is performed on the initial component information set to obtain semantic feature vectors, spatial feature vectors, and topological feature vectors, including: performing deep semantic feature extraction on the component text description through a semantic encoder to obtain semantic feature vectors; performing high-dimensional risk feature mapping on spatial location data through a spatial encoder to obtain spatial feature vectors; and performing neighborhood aggregation modeling on the physical connection relationship of components through a graph neural network to obtain topological feature vectors.

[0026] More specifically, in a concrete example of this application, firstly, a semantic encoder performs deep semantic feature extraction on the component text descriptions to obtain a semantic feature vector. The system extracts the text type descriptions of each component from the initial component information set and inputs them into a Transformer-based semantic encoder. By calculating the attention scores between words, the system captures the deep semantic relationships contained in the component type descriptions and outputs a semantic feature vector rich in deep semantic information. Secondly, a spatial encoder performs high-dimensional risk feature mapping on the spatial location data to obtain a spatial feature vector. The system extracts the spatial location data of each component from the initial component information set and inputs it into a multilayer perceptron-based fully connected neural network spatial encoder. Through layer-by-layer nonlinear transformation, the spatial location vectors of the components are mapped to a high-dimensional feature space that can represent risk features, and the output is a spatial feature vector with a clear risk orientation. Its mathematical expression is as follows: in, This represents the spatial feature vector of the output; Represents a non-linear activation function; The weight matrix represents the first layer of the fully connected network; Represents the spatial location vector of component i; This represents the bias vector of the first fully connected layer. The weight matrix represents the second layer of the fully connected network; This represents the bias vector of the second fully connected layer. In the above computational model, the inner layer operations... The original low-dimensional spatial location vector is mapped to the intermediate hidden layer feature space after linear transformation and nonlinear activation by the first-layer weight matrix, completing the initial encoding from physical coordinates to abstract risk features; outer layer operations Based on the intermediate hidden layer features, a higher-order nonlinear transformation is further performed to refine the initially encoded risk features into the final spatial feature vector. Taking a car component near the entrance of the first basement level as an example, the negative floor information encoded in its spatial location vector will be given a feature expression highly correlated with the risk of rainwater backflow in a high-dimensional feature space after the two layers of nonlinear mapping, rather than simply retaining the original floor values. This allows the output of the spatial encoder to directly reflect the disaster exposure risk level of the spatial location of the component.

[0027] Finally, a graph neural network is used to model the physical connections between components through neighborhood aggregation to obtain topological feature vectors. The system calculates the spatial Euclidean distance between components based on the 3D spatial coordinates of each component in the initial component information set, and constructs a graph structure connection matrix describing the physical connections between components by combining preset contact type coefficients. The dimension of this matrix is ​​the product of the total number of components. The input to the topology encoder includes a node feature matrix X (N×F) describing the features of each component and the aforementioned graph structure connection matrix A (N×N). Taking component 1 as an example, its final initial feature vector... Composed of semantic feature vectors, spatial feature vectors, vulnerability feature vectors, and cost feature vectors, it is a standardized, purely numerical initial feature vector that can be directly processed by a GNN. All N components of the building are processed into initial feature vectors, and these are combined into an N×F matrix, which is the node feature matrix. The graph structure connection matrix indicates whether the N components are interconnected; the interconnection is defined as a value between 0 and 1 based on distance. In the crucial topology coding stage, the system constructs a graph neural network based on the physical connections between components. Through a neighborhood aggregation algorithm, it learns and quantifies the risk level of the car due to changes in the physical state of its adjacent components. The formula for calculating the graph structure connection matrix describing the strength of the interconnection between component i and component j is: in, This represents the initial association weight values ​​between component i and component j in the graph structure connection matrix. This represents the normalization mapping function, used to transform the calculation result into a standardized value between zero and one. The contact type coefficient is used to classify and weight the physical dependency logic between components. It is defined according to the physical relationship between components. For vehicle component 1 in the underground parking garage, its relationship with the floor slab is a structural support relationship, and the value is [value missing]. =1; It is adjacent to the next vehicle. =0.4; unrelated to the fire protection system. =0; Represents the spatial Euclidean distance between the centroids of components ii and jj. The contact type coefficient of the molecular part. This component performs the physical logic determination function. For a car component in an underground parking garage, there is a direct structural support relationship between it and the ground floor slab. In this case, the coefficient is set to one, indicating a strong physical coupling. However, if the car is only adjacent to other vehicles, the coefficient is set to 0.4. If it is unrelated to the remote fire protection system, the weight is zero. This setting ensures that the algorithm can model risk propagation along the actual physical support relationships or spatial neighborhood clues. The denominator is the spatial Euclidean distance. This introduces a damping characteristic where the physical effect attenuates with spatial distance, ensuring that the intensity of the disaster factor's influence is inversely proportional to the spatial distance. This is achieved by... With geometric properties Perform ratio calculations, and by After standardization, the graph connection matrix accurately characterizes the actual risk exposure points of the car component within the complex physical grid of the building. Subsequently, the graph neural network receives the node feature matrix composed of semantic, spatial, and topological features and performs multiple rounds of feature aggregation operations based on the graph connection matrix, ultimately outputting a topological feature vector. This process ensures that, even in underground spaces not directly exposed to strong winds, the component can sense and predict the cascading amplification effect of typhoon damage on the risk of underground backflow through its physical topological links with drainage outlets, ventilation systems, or the main structure under extreme combined disasters.

[0028] Graph neural networks (GNNs) iterate through multiple rounds of neighborhood aggregation, updating the feature vector of the current node with the feature vectors of its neighboring nodes in each round. This simulates and predicts the propagation path and impact range of physical hazards within a building, ultimately outputting a topological feature vector containing cascading physical risks. The mathematical expression for this neighborhood aggregation update process is as follows: in, The node feature vector of component i after aggregation in the (k+1)th round; This represents the feature vector of component i at the current node in round k; The node feature vector of neighboring component j of component i in the kth round; This represents the set of neighboring components that have non-zero connection weights with component i in the graph structure connection matrix; This represents a neighborhood aggregation function, used to summarize the feature information of neighboring nodes; This represents the node update function, used to fuse aggregated neighborhood information with its own features. After K iterations, the final output topological feature vector Egraph= In the above calculation model, The function is responsible for collecting the current feature vectors of all neighboring components in the graph structure connection matrix that have non-zero connection weights with the target component, and summing them according to the connection weights. This process ensures that the neighboring components with higher physical association strength contribute more to the feature of the target node. The function then fuses the aggregated neighborhood information with the feature vector of the target component itself to generate a new feature vector containing neighborhood cascading risk information. As the aggregation rounds increase, the feature vector of each component node gradually encodes a wider range of neighborhood cascading risk information, so that the final output topological feature vector not only includes the component's own physical attributes but also the indirect impact of disaster factors propagating along physical connection paths. Taking a car component parked on the basement level as an example, this component has a structural support relationship with the floor slab below it, corresponding to a high connection weight in the graph structural connection matrix. It has a close proximity relationship with adjacent parked vehicles, corresponding to a moderate connection weight. It has no direct physical connection with the fire protection system, corresponding to a zero connection weight. After multiple rounds of neighborhood aggregation, the topological feature vector of this car component encodes its sensitivity to floor water accumulation and the associated risk of flooding of adjacent vehicles, among other physical cascading propagation information.

[0029] Accordingly, in step S13, the semantic feature vector, spatial feature vector, and topological feature vector are fused and weighted based on an attention mechanism to obtain the functional importance weights of each component. It should be understood that since the semantic feature vector, spatial feature vector, and topological feature vector output from the previous sub-step characterize the vulnerability features of the component from three independent dimensions—inherent component attributes, spatial exposure risk, and physical cascading propagation—the dominant roles of these three types of features in assessing the functional importance of different types of components differ significantly. If a simple concatenation or equal-weighted summation is performed using manually set fixed fusion weights, it will be impossible to adaptively adjust the contribution ratio of each modality information according to the characteristics of each component, resulting in a lack of contextual objectivity and intelligent adaptability in the inference results of the functional importance weights. Therefore, in the technical solution of this application, the semantic feature vector, spatial feature vector, and topological feature vector are further fused and weighted based on an attention mechanism to obtain the functional importance weights of each component. This achieves dynamic learning of the importance allocation of the three modalities based on the input data characteristics of each component, transforming the implicit decision-making logic of domain experts into a computable and scalable objective evaluation model. This approach overcomes the fundamental shortcomings of traditional single-expert scoring methods in estimating the functional importance parameters of building components, such as strong individual subjectivity, inconsistent results, high costs, and difficulty in scaling. It makes weight allocation more intelligent and context-aware, enabling the batch calculation of the functional importance weights of all components in a large commercial complex with tens of thousands of components in a very short time.

[0030] Specifically, in a concrete example of this application, the implementation process of this step includes three stages: multimodal feature concatenation and dynamic calculation of attention weights, attention weighted fusion, and functional importance weight mapping output. First, the system concatenates the semantic feature vector, spatial feature vector, and topological feature vector output from the previous sub-step to form a concatenated vector with more complete information dimensions. This concatenated vector is then input into an attention scoring network specifically used for weight calculation. The output layer of this network has three neurons corresponding to the original attention scores of the semantic, spatial, and topological feature channels, respectively. After processing by the Softmax normalization function, three attention weights are output, and the sum of the three attention weights is strictly equal to one. Subsequently, the calculated attention weights are used to perform a weighted summation operation on the semantic feature vector, spatial feature vector, and topological feature vector to obtain the fused feature vector. The mathematical expression for the attention weighted fusion is as follows: in, Represents the fused feature vector; This represents the attention weights dynamically assigned to semantic features by the attention scoring network; Represents semantic feature vectors; This represents the attention weights dynamically assigned to spatial features by the attention scoring network; Represents spatial feature vectors; This represents the attention weights dynamically assigned to topological features by the attention scoring network; This represents a topological eigenvector and satisfies the constraints. =1. In the above computational model, the three attention weights are not fixed constants preset by humans, but adaptive parameters dynamically learned by the attention scoring network based on the input data of each component. This partially represents the contribution of the inherent attribute semantics of the component to the fusion features. For components whose type descriptions contain strong functional orientation, this weight will be allocated higher. This partially characterizes the contribution of component spatial exposure risk to fusion features; for components located in high-risk spatial positions, this weight will be dynamically amplified. This partially characterizes the contribution of physical cascading propagation risk to the fused features, with higher weights assigned to components that are key transmission nodes in the building topology network. This adaptive dynamic weight allocation mechanism enables the fused feature vector to intelligently integrate information from three dimensions tailored to the specific characteristics of each component, avoiding the coarse-grained treatment of all components by fixed-weight schemes.

[0031] Finally, the system inputs the fused feature vector into the top-level predictive multilayer perceptron network, and processes it using the Sigmoid activation function to compress the output values ​​to an open interval between zero and one, ultimately outputting the functional importance weights of each component. Its mathematical expression is as follows: in, The value represents the functional importance weight of component i, and its value ranges from zero to one in an open interval. This represents the top-level prediction multilayer perceptron network, used to decode the fused feature vector into a scalar output; This represents the fused feature vector output from the previous stage; The sigmoid activation function is used to smoothly compress the linear calculation results into a probability space of zero to one. In the above mapping process, the multilayer perceptron network is responsible for nonlinearly decoding the multi-dimensional information encoded in the fused feature vector, extracting decision features directly related to functional importance. The sigmoid activation function ensures that the output functional importance weights strictly fall within the effective range of zero to one. The closer the weight value is to one, the greater the contribution of the component to maintaining the overall function of the building, and the higher the loss amplification factor will be obtained in the subsequent weighted economic loss aggregation.

[0032] Accordingly, in step S14, the type, spatial location, vulnerability parameter set, and replacement cost of each component are subjected to high-dimensional concatenation to obtain the vulnerability parameter vector of each component. It should be understood that, since the building to be evaluated contains tens of thousands of heterogeneous components, and the physical properties, geographical exposure, mechanical performance parameters, and economic value data of each component are distributed across different original databases or inference outputs, subsequent disaster cascade simulation and dynamic parameter correction stages face technical bottlenecks such as cumbersome computational addressing, difficulties in multi-dimensional data alignment, and low efficiency in multi-hazard simulation processing. Therefore, in the technical solution of this application, the type, spatial location, vulnerability parameter set, and replacement cost of each component are further subjected to high-dimensional concatenation to obtain the vulnerability parameter vector of each component, thereby constructing a unified and semantically complete digital parameter image for each atomized building component. This ensures that in subsequent complex simulation processes such as typhoon wind pressure calculation, rainwater intrusion path correction, and functionally weighted loss aggregation, the computing system can quickly read and synchronously process the cross-domain feature information of components based on a single vector index, greatly improving the computational robustness and real-time response speed of coupled simulation of multiple consistent disaster factors under cascading disasters.

[0033] Specifically, in a concrete example of this application, the computational system performs structured aggregation processing on the initial component information set from the component instantiation module and the functional importance weights output by the multimodal fusion inference network. First, the system extracts the component type data field for each component and parses its spatial location vector in the building coordinate system. Then, it retrieves the wind-induced vulnerability parameter set and the original flood vulnerability parameter set matching the component type from a pre-defined resistance database, and combines this with the entered component replacement cost data. Finally, according to a pre-defined tensor topology, the system concatenates the aforementioned discrete feature fields and the functional importance weights of each component in a high-dimensional feature space, ultimately generating a vulnerability parameter vector for a specific component. The mathematical expression of this vulnerability parameter vector is as follows: in, This represents the output vector of vulnerability parameters for a specific component i; A high-dimensional splicing operator representing the fusion of performance features; The component type representing component i; The spatial location vector of component i; The set of wind-induced vulnerability parameters for component i characterizes the component's structural resistance to strong wind pressure. The original flood vulnerability parameter set representing component i defines the initial loss probability of the component at this inundation depth; Represents the replacement cost of component i; This represents the functional importance weight of each component in the output. In the physical meaning of the above data packet, and Together, they defined the physical resistance distribution characteristics of components when facing multiple consistent disaster factors. The geometrically exposed nodes of the components were anchored along the infiltration path within the building, while and The combination of these parameters quantifies the economic density level of damage resulting from component failure. Specifically, the building to be evaluated is a prototype high-rise commercial office building. For a car component parked on the basement level, the calculation system identifies its component type as a car and calibrates its spatial location vector to its precise coordinates near the basement entrance. The system automatically extracts the wind-induced vulnerability parameter set (including the mean and standard deviation of critical wind speeds) and the original flood vulnerability parameter set for the car model from the database. Subsequently, the system performs high-dimensional concatenation of the vehicle's replacement cost with the functional importance weights of the component calculated by the multimodal fusion inference network, ultimately generating a holographic vulnerability parameter vector for the car component. Through this vector encapsulation, the calculation system can directly call its spatial location to determine the backflow priority based on this single vector object during subsequent simulations of underground parking garage flooding. Combined with its flood vulnerability index corrected by typhoon damage levels, it can accurately calculate the car's loss probability under cascading disasters and its contribution to the overall economic vulnerability curve of the building.

[0034] Specifically, in step S2, based on the typhoon's physical wind speed and vulnerability parameter vector in the disaster scenario, the wind-induced failure probability of each building envelope component is calculated. Logical analysis and level aggregation are then performed on the wind-induced failure probabilities of each component to obtain the wind-induced failure probability of each building envelope component and the overall damage level of the building. It should be understood that damage to modern urban building envelope components such as glass curtain walls and window sashes under extreme typhoon conditions is the physical source of large-scale rainwater intrusion and subsequent cascading flood disasters. The failure behavior of components under different disaster-causing wind speeds exhibits strong uncertainty and physical correlation. Therefore, in the technical solution of this application, the wind-induced failure probability of each building envelope component is further calculated based on the typhoon's physical wind speed and vulnerability parameter vector in the disaster scenario. Logical analysis and level aggregation are then performed on the wind-induced failure probabilities of each component to obtain the wind-induced failure probability of each building envelope component and the overall damage level of the building, thereby accurately characterizing the physical evolution process of the key intermediate variable of wind-induced damage. In this way, the probability of component damage at the atomic level can be transformed into the building damage state at the system level, providing a logical starting point and quantitative boundary for the identification of subsequent cascading effect correction factors and the simulation of rainwater infiltration problems.

[0035] Figure 4 This is a flowchart of step S2 of the method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to an embodiment of this application. Figure 4As shown in the embodiment of this application, step S2 includes: S21, based on the typhoon physical wind speed and reference wind speed in the disaster scenario, performing wind speed normalization calculation based on wind pressure effect on each building envelope component to obtain a standardized damage intensity index; S22, performing failure probability calculation based on adaptive Latin hypercube sampling on the standardized damage intensity index and the wind-induced vulnerability parameters extracted from the vulnerability parameter vector to obtain the wind-induced failure probability of each building envelope component; S23, performing multiple logical judgments and overall damage level aggregation on the wind-induced failure probability of each building envelope component to obtain the overall damage level of the building.

[0036] Accordingly, in step S21, based on the typhoon physical wind speed and reference wind speed in the disaster scenario, a wind speed normalization calculation based on the wind pressure effect is performed on each building envelope component to obtain a standardized damage intensity index. It should be understood that since the external typhoon physical wind speed does not cause substantial physical damage to the building envelope before reaching the mechanical bearing capacity boundary of a specific component, and because the threshold response of components to resistance forces differs significantly under different fortification levels, directly mapping wind speed without ignoring the critical damage boundary of the component will not accurately depict the abrupt damage effect of the disaster load near the critical point. Therefore, in the technical solution of this application, a wind speed normalization calculation based on the wind pressure effect is further performed on each building envelope component based on the typhoon physical wind speed and reference wind speed in the disaster scenario to obtain a standardized damage intensity index, thereby refining the description of the cumulative damage effect of the remaining wind pressure on the component after exceeding the critical threshold. In this way, the interference of non-damaging wind speed fluctuations can be effectively eliminated by introducing critical threshold determination, making the normalized index more consistent with the physical evolution logic of the building envelope components from being loaded to being damaged, and providing load input with physical threshold awareness for subsequent execution of failure probability calculation based on adaptive sampling.

[0037] More specifically, in a concrete example of this application, the building to be evaluated is a prototype high-rise commercial office building, and strength calculations are performed on its windward glass curtain wall components. The calculation system first obtains the typhoon physical wind speed from the disaster scenario parameters, and retrieves the corresponding critical wind speed for damage from the wind-induced vulnerability parameters in the vulnerability parameter vector based on the fortification level of the component's height. This critical wind speed is determined by a log-normal distribution probability density function constructed from local measured data. The system then performs a difference comparison between the typhoon physical wind speed and the critical wind speed for damage, and performs a power-function-based normalization calculation in conjunction with a preset reference wind speed, outputting a standardized damage intensity index. The standardized calculation formula is defined as follows: in, This represents the standardized damage intensity index generated by calculation for a specific component i; Represents the input physical wind speed of the typhoon; The critical wind speed that causes damage to component i follows a log-normal distribution; Represents the preset reference wind speed; The wind pressure effect index represents the vulnerability parameter extracted from component i. In the above physical calculation model, the molecular part... The logical operator extracts the effective surplus amount of the typhoon's physical wind speed exceeding the critical damage wind speed, achieving precise capture of the component's damage initiation point. This ensures that the index output is zero when the wind speed is below the critical threshold, consistent with the logic of real physical resistance. This surplus amount is then dimensionally standardized using a reference wind speed and nonlinearly mapped by the wind pressure effect index, thus deconstructing the external physical wind speed layer into a standardized damage-causing pressure energy that directly characterizes the glass curtain wall component. Through this calculation process, the system can transform the uncertain, dynamically changing external wind field environment into a normalized intensity data dictionary with a unified mechanical evaluation scale for each specific component, greatly enhancing the cascade disaster assessment model's perception accuracy of sudden damage characteristics caused by extreme strong winds.

[0038] Accordingly, in step S22, the failure probability of each building envelope component is calculated based on adaptive Latin hypercube sampling by performing a failure probability calculation on the standardized damage strength index and the wind-induced vulnerability parameters extracted from the vulnerability parameter vector. It should be understood that the physical resistance of urban building envelope components such as glass curtain walls and window sashes is significantly random due to the influence of material uniformity, construction quality, and dynamic wind pressure effects. Furthermore, their failure probability often exhibits a very high gradient change within a specific critical range as the load increases. If traditional Monte Carlo simulation or conventional Latin hypercube sampling is used, it is not only difficult to guarantee the fitting accuracy near the damage critical point, but also leads to a severe computational bottleneck when modeling modern high-rise buildings with tens of thousands of components due to excessive allocation of computational resources in non-failure areas. Therefore, in the technical solution of this application, the failure probability of each building envelope component is further calculated based on adaptive Latin hypercube sampling by analyzing the standardized damage intensity index and the wind-induced vulnerability parameters extracted from the vulnerability parameter vector. This allows for a refined and efficient quantitative characterization of the component damage process through a dual strategy of global exploration and local refinement. This significantly improves computational efficiency while ensuring overall assessment accuracy, particularly achieving more stable convergence results in the critical failure domain where the vulnerability curve slope is largest. This provides accurate underlying probability input for the subsequent construction of a building-wide cascading damage early warning system.

[0039] More specifically, in a particular example of this application, the building to be evaluated is a commercial office complex employing a modern curtain wall system. After obtaining the standardized damage strength index for the 15th-floor glass curtain wall component on the windward side, the calculation system extracts the logarithmic mean and logarithmic standard deviation reflecting the resistance level from the vulnerability parameter set of this component. First, the implementation process involves the execution of an adaptive sampling strategy. The system uses an initial Latin hypercube sampling with a small sample size to perform a preliminary simulation across the entire resistance parameter space, identifying high-gradient sensitive regions where the failure probability gradient changes most drastically. Subsequently, the system performs local point densification for this identified critical failure domain, and again performs adaptive high-frequency sampling, thereby obtaining a high-density sample point cloud near the critical critical point. The wind-induced failure probability of each building envelope component is calculated and output by solving the limit state equation. The specific probability calculation formula is defined as follows: in, This represents the calculated probability of wind-induced failure of component i; Represents the cumulative function of the standard normal distribution; This represents the standardized damage intensity index for component i, which is the output of the previous sub-step. This represents the log-mean of the resistance distribution of component i extracted from the wind-induced vulnerability parameter; This represents the logarithmic standard deviation of the resistance distribution of the extracted component i. In the above mathematical physics model, the logarithmic term... This method achieves a logarithmic difference comparison between the expected value of external wind pressure load and the median intrinsic resistance of the component, enabling quantification of the physical surplus of load exceeding the component's design strength. (Logarithmic standard deviation of the denominator) This is used to characterize the discrete fluctuations caused by component material strength and construction errors. By scaling the difference in the numerator, the model gains the ability to perceive uncertain fluctuations. Finally, the standard normal distribution cumulative function is used. The integral mapping transforms the discrete, dynamically random wind pressure impact into a continuous wind-induced failure probability between zero and one. Based on this calculation logic, when facing post-disaster assessment scenarios of extreme strong winds, the system can accurately capture the key physical inflection point of each glass curtain wall unit from a stable state to failure, thus accurately depicting the chain damage evolution of the entire building under wind load, laying a solid probabilistic foundation for subsequent cascading disaster analysis.

[0040] In particular, Figure 6 This diagram illustrates a comparison of simulation results between the A-LSH method and the ordinary Monte Carlo method for constructing urban building flood vulnerability curves considering typhoon cascading disasters, according to embodiments of this application. Figure 6As shown, this application first conducts a first-stage global exploratory sampling within the standardized wind-induced damage intensity index (corresponding to the blue circular sampling points in the figure), initially establishing a mapping relationship between damage intensity and failure probability. Further, by identifying intervals where the failure probability changes drastically, the critical region for failure probability transitions (i.e., the ROI region shown in the shaded area in the figure) is automatically determined. Within this ROI region, this application initiates a second-stage adaptive densification sampling (corresponding to the small green dots in the figure), increasing the sampling density within this interval, thereby achieving accurate characterization of the steep segment of the vulnerability curve with a very small total global sample size (e.g., 500 samples). It should be understood that because wind-induced failure often has a significant threshold effect, traditional uniform sampling methods often result in sparse sampling points in the failure abrupt change zone, leading to low fitting accuracy. However, as... Figure 6 As shown, this application encrypts the identified key areas, resulting in a high degree of overlap between the fitted curve (solid line) and the theoretical true curve (dashed line), achieving a fitting accuracy (MSE) of 0.00182 with extremely low computation time. This ensures the high reliability of the overall building damage level generated in subsequent steps, avoiding the underestimation of building damage due to insufficient sampling points.

[0041] Accordingly, in step S23, the wind-induced failure probability of each building envelope component is subjected to multiple logical judgments and overall damage level aggregation to obtain the overall building damage level. It should be understood that since urban buildings are a physical system composed of a massive number of heterogeneous components, the structural failure of a single building envelope component does not directly equate to the loss of macroscopic functionality of the building system. Without a scientific judgment mechanism that can integrate the damage states of each discrete node and map them to the defense level gradient, the calculation system will be unable to accurately define the logical starting point of the transmission of cascading disasters in subsequent spatiotemporal evolution. Therefore, in the technical solution of this application, the wind-induced failure probability of each building envelope component is further subjected to multiple logical judgments and overall damage level aggregation to obtain the overall building damage level, thereby transforming the failure probability distribution at the microscopic component level into macroscopic building physical damage state parameters. In this way, a structured discrete state determination path can be used to establish physical boundaries for the disaster resistance degradation law of buildings under extreme scenarios of typhoons and heavy rainfall, ensuring that subsequent dynamic correction of the cascading effects of internal components has clear state triggering evidence.

[0042] More specifically, in a concrete example of this application, the building to be evaluated is a 20-story commercial complex. After obtaining the wind-induced failure probabilities of all building envelope components, such as exterior glass curtain walls, ventilation openings, and air conditioning units, the calculation system retrieves the preset damage classification logic and damage rate thresholds from the graded diagnostic model. At this point, the system employs Boolean cliff-based aggregation logic, using a multi-level discrimination path constructed from union and intersection operators to perform a hierarchical aggregation task targeting the overall damage level of the building. The mathematical expression of its judgment model is defined as follows: in, K represents the calculated overall damage level of the building; K represents the preset set of building damage level indexes, which includes five levels from undamaged to completely damaged; k represents the specific damage level index value. Represents the total number of preset independent fault determination paths under damage level k; i represents the path index under a specific damage level. represents the set of related components that need to meet the conditions together within the i-th decision path; m represents a specific component in the set. Represents an indicator function used for Boolean mapping of execution conditions; This represents the real-time damage status value of the components extracted from the wind-induced failure probability of each of the building envelope components; This represents a preset damage rate threshold for a specific level and path. For the aforementioned building physics judgment model, this formula utilizes nested product operators to achieve strong coupling at the logical level. The innermost product operation... Corresponding to the calculation process of "AND" logic, its significance lies in simulating the parallel failure characteristics under specific building failure paths, such as when a building is determined to be at a moderate level of damage. If the judgment rule is set as follows: the damage rate of building ventilation openings exceeds 25% and the damage rate of drainage outlets exceeds 40%, then only when the damage status of all specific components within the link exceeds the corresponding... At a threshold, the product of the indicator function sequences will output a value, signifying that the specific damaged scene has been successfully activated. (Intermediate layer) The structure corresponds to the calculation process of "OR" logic. Its significance lies in representing the diversity of damage forms in building systems. That is, for the same damage level, the system may be set to large-area damage to exterior windows, damage to air conditioning unit groups, or coordinated failure of multiple parts, etc. Each path has an independent trigger condition; if any one of these paths meets the judgment criteria, the entire system is marked as entering that damage tier. Through this Boolean cliff-like aggregation mapping, the system can efficiently transform the originally scattered probabilistic inputs into state labels with clear disaster prevention implications. This provides a precise logical entry point for subsequent dynamic parameter correction of the vulnerability of internal components to flooding caused by damage to the retaining structure under the typhoon-rainfall-flood coupling effect.

[0043] In particular, regarding the disaster prevention and mitigation needs of modern urban buildings in extreme typhoon scenarios, step S23 in the first embodiment above exhibits significant limitations. Its Boolean cliff-like aggregation logic can only handle static and isolated failure states, completely ignoring the crucial spatial aerodynamic cascading interactions in fluid mechanics. In real strong wind disaster environments, the failure of building envelope components is not a statistical event that does not interfere with each other. Once a window sash or curtain wall on the windward side physically breaks, a large amount of strong airflow will rush into the building's interior through the breach, inducing a surge in internal transient wind pressure. This superposition effect of internal positive pressure and external negative pressure will cause components on the lateral or leeward sides, which were originally in a safe stress state, to bear a multiplied load, thus triggering a domino effect of chain failures. The first embodiment treats each component as a physically isolated node, severing the aerodynamic transmission link at the spatial level. This not only leads to a serious underestimation of the risk assessment results but also fails to provide scientific support for early warning of continuous collapse of building structures.

[0044] To address the aforementioned technical problems, in another specific example of this application, an improved mechanism for constructing a spatial aerodynamic topology diagram and a cascaded failure propagation model is proposed. Specifically, the wind-induced failure probability of each building envelope component is subjected to multiple logical judgments and overall damage level aggregation to obtain the overall building damage level. This includes: calculating the aerodynamic coupling strength between each pair of building envelope components based on the component spatial coordinate vector and the disaster wind direction vector to obtain an aerodynamic coupling matrix; cascading amplification of the comprehensive survival rate of each component under the internal pressure transmission of its own damage and adjacent node damage based on the wind-induced failure probability and the aerodynamic coupling matrix to obtain a cascaded amplified failure probability; and performing soft logic aggregation and damage level judgment on the cascaded amplified failure probability to obtain the overall building damage level.

[0045] Accordingly, based on the component spatial coordinate vector and the disaster wind direction vector, the aerodynamic coupling strength between each pair of building envelope components is calculated to obtain the aerodynamic coupling matrix. It should be understood that, when modern urban buildings are subjected to extreme typhoons, the failure of building envelope components is not a physically independent random event. Once the glass curtain wall or window sash on the windward side is damaged by pressure, strong external airflow will rush into the interior through the breach, inducing a surge in internal transient wind pressure. This internal pressure bursting effect will cause other components, which were originally in a safe stress state, to bear a multiplied load. Traditional discretized evaluation models focus on static component judgment, completely severing the aerodynamic transmission link at the spatial level, resulting in a significant underestimation bias in the evaluation results. Therefore, in the technical solution of this application, based on the component spatial coordinate vector and the disaster wind direction vector, the aerodynamic coupling strength between each pair of building envelope components is calculated to establish an anisotropic aerodynamic topology structure. By obtaining the spatial coordinate vector of each building envelope component and the wind direction vector at the disaster site, the anisotropic impact correlation of the internal airflow on other components after component damage is quantified. In this way, the directionality of wind pressure transmission in fluid mechanics can be integrated into the static geometry of building components, generating an energy transmission field with topological awareness, thereby laying a physical topological foundation for realizing the reconstruction of the chain failure process from point to surface.

[0046] More specifically, in a concrete example of this application, the computational system obtains the spatial coordinate vectors of each component of the building envelope to be evaluated, and extracts the wind direction vector of the current typhoon disaster based on real-time meteorological data at the disaster site. Subsequently, the system performs a nonlinear mapping operation based on the spatial distance damping and the cosine of the wind direction angle to quantify the pressure conduction coupling weights between nodes, and finally outputs the aerodynamic coupling matrix. The specific calculation formula for its aerodynamic coupling weights is defined as follows: in, Representation Component With components aerodynamic coupling matrix element values ​​between them; Represents the spatial distance damping coefficient; and Components With components The spatial coordinate vector; The Euclidean distance between the centroids of the two components; This represents the wind direction vector for typhoon disasters; The vector pointing from component i to component j; This is an indicator function used to perform a self-feedback effect exclusion decision based on the component's unique index identifier, thus excluding the component's self-feedback effect. In the above aerodynamic coupling physical model, the exponential term... It acts as a distance damping unit in the aerodynamic propagation process, by adjusting the spatial distance damping coefficient. Mapped to Euclidean distance The nonlinear evolution curve provides a physical characterization of the energy loss in wind pressure propagation as geometric distance increases. (Cosine structure term) It then undertook the core task of directional assessment, calculating the wind direction vector of typhoon disasters. Relative position vectors of the two components The phase consistency between them dynamically defines the weight distribution of airflow impact in the spatial topology, ensuring that the coupling strength in the downwind direction is significantly amplified in a physical sense, while the upwind side is reasonably weakened. By integrating distance damping and directional correction, the generated aerodynamic coupling matrix can accurately characterize the wind pressure wave transmission efficiency caused by local damage inside the building, transforming the originally static component lattice into a resilient response network with dynamic air pressure correlation.

[0047] Accordingly, based on the wind-induced failure probability and the aerodynamic coupling matrix, the comprehensive survival rate of each component under the internal pressure transmission caused by its own damage and the failure of adjacent nodes is cascaded and amplified to obtain the cascaded amplified failure probability. It should be understood that, due to the significant non-independence and cascading characteristics of the failure logic of building envelope components under extreme typhoon conditions, the dogmatic assumption of treating each component as a physically isolated node in traditional assessment models completely ignores the secondary impact load on adjacent components caused by the explosive effect of transient wind pressure induced by the violent influx of large-volume external airflow after local damage. This internal pressure transmission mechanism often triggers a domino-like chain reaction of collapse risks; without cascade modeling, the overall risk assessment results will suffer a serious underestimation bias. Therefore, in the technical solution of this application, based on the wind-induced failure probability and the aerodynamic coupling matrix, the comprehensive survival rate of each component under internal pressure transmission caused by its own damage and the failure of adjacent nodes is cascaded and amplified to obtain the cascaded amplified failure probability. This introduces the Markov approximation concept and the joint survival probability model, and utilizes anisotropic coupling weights to perform risk neighborhood propagation processing on the original damage risk. The aim is to simulate the comprehensive failure performance of components under internal pressure bursts caused by their own damage or the failure of adjacent nodes through the joint survival probability model. In this way, the continuous failure process of a building from point to surface can be realistically reproduced through the survival product principle in probability theory, quantifying the additional damage risk brought about by sudden increases in internal wind pressure, and ensuring that the assessment system has extremely high sensitivity to extreme physical chain reactions.

[0048] More specifically, in a concrete example of this application, the computing system obtains the original wind-induced failure probabilities of each building envelope component calculated in the aforementioned sub-steps, as well as the aerodynamic coupling matrix characterizing the intensity of spatial air pressure correlation. Subsequently, the system introduces the internal wind pressure transmission coefficient and performs cascaded convolution calculations of dynamic failure risk based on a joint survival probability model to generate the final cascaded amplified failure probability. The mathematical expression for this process is as follows: in, Refers to components The final wind-induced failure probability after cascaded amplification; and Components With components The probability of original wind-induced failure caused by direct external load; The internal wind pressure conduction coefficient characterizes the efficiency of pressure energy conduction; These are the weights of the aerodynamic coupling matrix output from the aforementioned steps. This represents the total number of components involved in the aerodynamic correlation calculation.

[0049] This formula aims to quantify the all-probability failure boundary of a component under a combined stress field. The first term on the right-hand side of the formula... This reflects the initial survival probability of component i under the initial design conditions against external wind pressure impact. (Multiplication operator) This constructs a dynamic risk propagation field weight, where each component term ( The physical meaning of ) is: only when the risk of damage to adjacent component j is... via aerodynamic coupling matrix Spatial weight amplification and internal wind pressure transmission coefficient After amplitude correction, the induced internal pressure shock still did not cause the target component i to fail, and only then could the component obtain the survival probability under this path. Through the complement operation of this full probability survival product, the system successfully broke the statistical bias of isolated distribution and achieved an accurate characterization of the aerodynamic cascade evolution chain of "wind-induced damage - internal pressure surge - secondary explosion", ensuring that the generated cascaded amplification failure probability can capture the common topological damage evolution of building envelopes under strong wind fields.

[0050] Accordingly, soft logic aggregation and damage level assessment are performed on the cascading amplification failure probability to obtain the overall building damage level. It should be understood that traditional discretized Boolean threshold methods, when dealing with cascading failures caused by multi-hazard coupling, often result in abrupt, non-physical jumps in the overall assessment results due to small fluctuations in the probability of individual components. This system instability severely interferes with the scientific validity of post-disaster emergency decision-making. Therefore, in the technical solution of this application, soft logic aggregation and damage level assessment are further performed on the cascading amplification failure probability to obtain the overall building damage level. This introduces a probability-based T-modulus soft logic aggregation mechanism, incorporating the cascading amplification failure probability into the continuous risk control assessment path. This enables full probability transfer from the probability of underlying components to the state of the top-level system, greatly improving the smoothness and physical robustness of the algorithm under complex disturbance environments.

[0051] More specifically, in a particular example of this application, the building to be evaluated is set to be under specific extreme disaster conditions. The calculation system obtains the cascading amplification failure probability of each affected component by parsing the underlying data architecture, and simultaneously retrieves the set of all damage levels preset for the building prototype category, as well as the corresponding multiple fault judgment paths. For the medium damage level, the system presets two core judgment paths: the first is the vertical wind and rain intrusion path caused by the failure of the main windward curtain wall system, and the second is the internal pressure burst path of the leeward cladding components caused by internal flow field instability. The system first uses the probabilistic T-modulus operator to perform joint trigger judgment on the set of components associated in each path, and then completes the comprehensive trigger assessment across paths through the T-inverse modulus operator, calculates the continuous risk score of the building reaching a specific damage level, and compares it with the system-level confidence threshold set for that level. The mathematical expression for judging the degree of building damage is defined as follows: in, The output is the overall damage level of the building; Represents the preset damage level index; The set of all damage levels; A logical indicator function that satisfies the threshold condition; Refers to a specific fault diagnosis path; For path A collection of related components; For components The cascade amplification of failure probability; This refers to the damage level. The system-level confidence threshold is set. In the above full probability quantization decision model, the innermost product operator... A cascaded AND logic within the probability space was constructed. Its physical significance lies in accurately characterizing the joint probability of failure of all related components in a specific failure path through a joint probability distribution. This achieves probabilistic aggregation of the chain reaction of component failures within the path, realistically reproducing the direct threat of concurrent failures to structural integrity. The intermediate-level operator structure... This constructs a cascaded "OR" logic across paths, essentially a probability and operator from probability theory, to quantify the comprehensive continuous risk score of a system reaching a specific damage level when a building faces multiple heterogeneous disaster evolution paths. This soft logic design overcomes the information loss caused by traditional Boolean judgments. (Logic indicator function) Combined with system-level confidence thresholds set for specific levels It acts as the decision gating system for the entire probabilistic link, accurately mapping continuously distributed risk signals to discrete physical state levels while maintaining physical logical continuity. Finally, it utilizes the maximum value operator. Weighted filtering is performed within the full probability level space to ensure that the overall building damage level output is Level II, i.e., moderate damage. This process successfully eliminates systemic spikes caused by fluctuations in the failure probability of local components, providing accurate state constraints for subsequent correction of flood vulnerability parameters and dynamic calculation of rainwater intrusion intensity based on the actual damage location of the building envelope.

[0052] This improved mechanism, by constructing a spatial aerodynamic topology map and a cascading failure propagation model, fundamentally addresses the technical deficiency of traditional urban building vulnerability assessments by lacking physical mechanisms. It achieves a refined depiction of the entire process of "wind-induced damage - internal pressure burst - cascading failure" under typhoon-rainfall combined disasters. It not only physically recreates the damage logic of extreme disasters on modern high-rise buildings and their maintenance structures, but also eliminates the jump in assessment results caused by local data fluctuations by introducing a fully probabilistic soft logic aggregation algorithm. This significantly enhances the scientific rigor and accuracy of risk warnings, providing a high-precision quantitative decision-making tool for improving the resilience of coastal city disaster prevention systems.

[0053] Specifically, in step S3, based on the overall damage level and vulnerability parameter vector of the building, the original flood vulnerability parameters of each component are dynamically corrected using a cascading effect to obtain the corrected vulnerability parameters. Furthermore, considering the multi-source flooding caused by external backflow and internal breach intrusion, the maximum effective inundation depth of each component is calculated. It should be understood that in typhoon cascading disaster scenarios, wind-induced damage to the building envelope significantly alters the physical exposure boundaries of its internal components, leading to a dynamic deterioration of the original flood resistance under the cascading impact of wind and rain. Moreover, the internal components of high-rise buildings often face multi-source coupled threats from backflow of rainfall introduced by the damage points in the envelope and surface flooding. If traditional static flood vulnerability parameters are used for assessment, the amplifying effect of wind-induced damage on subsequent flood losses cannot be accurately reflected. Therefore, in the technical solution of this application, the original flood vulnerability parameters of each component are dynamically corrected through a cascading effect based on the overall damage level and vulnerability parameter vector of the building to obtain the corrected vulnerability parameters. Furthermore, by integrating multiple sources of flooding, including external backflow and internal breach intrusion, the maximum effective inundation depth of each component is calculated. This quantifies the physical correlation between wind-induced damage and increased flood sensitivity, and accurately simulates the evolution process of infiltration and water accumulation in stratified spaces. This approach can establish a cascading transmission path between typhoon wind pressure damage and indoor flood damage, eliminating the risk underestimation bias caused by single disaster factor assessments and improving the accuracy of loss prediction for modern buildings under extremely complex conditions.

[0054] More specifically, in this embodiment, step S3 includes: identifying a two-dimensional cascade effect correction factor for the overall building damage level and the component spatial location information in the vulnerability parameter vector to obtain a cascade effect correction factor; scaling and adjusting the original vulnerability logarithmic mean based on the cascade effect correction factor and the original flood vulnerability parameter in the vulnerability parameter vector to obtain a corrected vulnerability parameter; and accumulating the multi-source infiltration volume over time based on the breach area determined by the overall building damage level, external rainfall intensity data, and the component spatial location information in the vulnerability parameter vector to obtain the maximum effective inundation depth of each component.

[0055] Specifically, in a particular example of this application, the building to be assessed is a 20-story commercial complex, and its 15th-floor office area was determined to be at the overall building damage level after undergoing the simulation steps described above. (Moderate damage state). The calculation system first analyzes the spatial location information of components stored in the vulnerability parameter vector, identifying a window-adjacent office component on that floor as being adjacent to the nearest broken window. Based on preset rules, the system determines the damage severity factor for the overall damage level of the building. The value is set to 0.5, and the feasibility factor of rainwater intrusion path is determined based on the spatial relative location. The value is set to 0.1. Subsequently, the system performs the identification calculation of the cascade effect correction factor, calculated using the following formula: in, The cascade effect correction factor represents the calculated output; The damage factor represents the degree of damage to the enclosure structure near the component; This represents a rainwater intrusion path feasibility factor that reflects the degree of component exposure relative to the damaged location. Using this formula, the system successfully performs a joint mapping between the macroscopic physical impact intensity on the building and the microscopic risk of component location exposure, generating scalar coefficients for correcting the distribution of physical resistance.

[0056] After obtaining the cascading effect correction factor, the system retrieves the original flood vulnerability parameters from the vulnerability parameter vector and adjusts the logarithmic mean of the original vulnerabilities by scaling. The calculation formula is as follows: in, This represents the calculated corrected logarithmic mean, which is the core component of the corrected vulnerability parameter. This represents the log-mean of the original vulnerabilities extracted from the original flood vulnerability parameters in the vulnerability parameter vector; This represents the cascading effect correction factor identified in the preceding steps. In the above logic, by scaling the logarithmic mean, the system dynamically shifts the position of the flood vulnerability curve, intuitively reflecting the significant increase in vulnerability of components that originally had high flood resistance due to the loss of the retaining structure under the cascading effects of wind and rain. This achieves a shift in physical parameters from "static resistance" to "cascading resistance," improving the model's physical robustness in predicting subsequent cascading failure probabilities.

[0057] Finally, the calculation system performs dynamic simulations of the maximum effective inundation depth for each component. Specifically, based on the breach area determined by the overall building damage level, external rainfall intensity data, and component spatial location information in the vulnerability parameter vector, the multi-source infiltration volume is accumulated over time to obtain the maximum effective inundation depth for each component. This includes: calculating the internal infiltration depth of each component at each time point based on the overall building damage level, external rainfall intensity data, and component spatial location information; calculating the equivalent inundation depth of each component at each time point based on the real-time depth of external surface floodwater, component installation height, and internal infiltration depth; and extracting the extreme values ​​of the equivalent inundation depth of each component during the disaster simulation period to obtain the maximum effective inundation depth for each component.

[0058] More specifically, firstly, based on the overall damage level of the building. Determined internal infiltration rate function Calculate the accumulated depth of rainwater in the local space where the component is located at time t due to wind-induced breach intrusion. Its calculation formula is defined as: in, This represents the calculated depth of internal infiltration water accumulation in component i at time t. The internal infiltration rate function of a building depends on the real-time rainfall intensity, the breach area determined by the overall damage level of the building, and the spatial location information of the components. In this integral logic, The system fully reproduces the process of indoor water accumulation caused by the loss of high-level physical boundaries (such as glass curtain walls) over time. Since modern buildings also face the threat of rising surface water levels, the system further introduces a multi-source coupled discriminant model to calculate the real-time equivalent inundation depth faced by components at time t. : in, Represents the equivalent flooding depth of component i-face at time t; Represents the real-time depth of external surface floodwaters; The installation height (i.e., ground clearance) of component i is represented by the vulnerability parameter vector record. This represents the consistent internal infiltration depth calculated in the aforementioned steps. In the above discrimination model, the terms in square brackets... It depicts the traditional bottom-up threat of flooding, while It depicts the unique top-down intrusive threat of cascading disasters, with both parties participating. The competitive calculation of operators ensures that the system uses the most unfavorable disaster-causing load as the evaluation input, whether it is backflow in a low-rise basement or seepage due to damage in a high-rise building. Finally, the system calculates and outputs the maximum effective inundation depth of each component by traversing the entire disaster simulation period T. Its calculation formula is defined as: in, The maximum effective inundation depth of each component in the final output represents the upper limit of the extreme flood load faced by that component throughout the entire disaster cycle. Through the above three-layer cascaded calculation, the system thoroughly solves the problem of inaccurate loss quantification under typhoon compound disasters in traditional models from two dimensions: physical resistance correction and multi-source load coupling. Thus, the cascaded calculation logic from enclosure damage to internal flood load is successfully closed, providing high-precision structured decision parameters for subsequent calculation of cascade failure probability based on the vulnerability integral formula and the final construction of the building flood vulnerability curve.

[0059] Figure 7 This is a schematic diagram illustrating the physical processes and key parameter corrections for typhoon cascading disasters in the urban building flood vulnerability curve construction method considering typhoon cascading disasters, according to an embodiment of this application. It should be understood that, under typhoon cascading disaster conditions, building flood damage is not only affected by the depth of external surface water, but also by changes in the water intrusion path caused by wind-induced failure of the building envelope. Therefore, the following section utilizes... Figure 7 The physical cascade model shown guides parameter adjustments. For example... Figure 7 As shown, this cascading damage process includes: Figure 7 As shown in (a), the building facade and envelope are subjected to high-frequency damaging loads due to the influence of typhoon wind pressure and pulsating wind speed; Figure 7 As shown in (b), when the wind pressure exceeds the component's bearing capacity limit, it will cause physical damage to the building envelope, such as glass curtain walls and windows, thereby determining the overall damage level of the building. .like Figure 7 As shown in (c), due to the disappearance of the physical barrier of the building envelope, heavy rainfall, driven by wind pressure, enters the building interior through the damaged openings, forming an independent intrusion path distinct from external flooding; as Figure 7 As shown in (d), the accumulation of intrusive moisture alters the actual flooding state of the indoor components. At this point, in step S3 of this application, combined with... Figure 7 The damage level determined in (b) is consistent with Figure 7 The inflow path information in (c) is relevant to the original flood vulnerability parameters. Dynamic correction is performed to obtain Taking into account the competition between external surface water and indoor intrusive water, the maximum effective flood depth at the component level was calculated. This allows for a more accurate representation of the amplified flood effects inside damaged buildings, significantly improving the physical consistency of urban building flood damage assessments under complex weather conditions.

[0060] Specifically, in step S4, based on the corrected vulnerability parameters and the maximum effective inundation depth, the cascading failure probability of each component is calculated. The cascading failure probability is then weighted and aggregated using the vulnerability parameter vector and functional importance weights to obtain a weighted relative economic loss and a comprehensive dataset. It should be understood that after completing the dynamic correction of flood vulnerability caused by wind-induced damage and the calculation of the maximum effective inundation depth under multi-source flooding, the model only obtains the physical disaster conditions of the components under cascading disasters, but has not yet further transformed these physical disaster conditions into failure probabilities and economic consequences that can be used for overall loss assessment. Without this unified mapping process from physical disaster intensity to loss output, the system will be unable to construct the loss sample matrix required for subsequent building flood vulnerability curves, nor will it be able to realize the differentiated value expression of different components within the overall building functional system. Therefore, in the technical solution of this application, the cascading failure probability of each component is further calculated based on the modified vulnerability parameters and the maximum effective flooding depth. The cascading failure probability is then weighted and aggregated using the vulnerability parameter vector and functional importance weights to obtain a weighted relative economic loss and a comprehensive dataset. This achieves a unified quantitative mapping from component-level disaster conditions to system-level economic response. In this way, the component damage process under the combined effects of wind-induced damage, rainwater intrusion, and flooding can be transformed into structured data results that are fitable, statistically sound, and inversely applicable, providing a high-quality data foundation for the subsequent construction of building flood vulnerability curves and post-disaster status diagnosis.

[0061] More specifically, in the embodiments of this application, step S4 includes: performing vulnerability integral calculation on each component based on the corrected vulnerability parameters and the maximum effective flooding depth to obtain the cascade failure probability of each component; performing weighted economic loss aggregation on each component based on functional importance based on the cascade failure probability and the replacement cost and functional importance weight in the vulnerability parameter vector to obtain the weighted relative economic loss; and performing structured association and iterative persistent encapsulation of each simulation result through multi-scenario simulation loops based on the weighted relative economic loss, disaster intensity input parameters, and overall building damage level to obtain the weighted relative economic loss and comprehensive dataset.

[0062] Specifically, in a concrete example of this application, the building to be assessed is a commercial office complex with twenty floors above ground and two floors below ground. The system first reads the corrected vulnerability parameters and maximum effective flooding depth output from the previous sub-step for the office equipment component located near a window on the fifteenth floor and the car component near the entrance on the first basement floor. These parameters are then substituted into the vulnerability integral model to calculate the cascading failure probability of the component under the entire cascading disaster. This calculation process uses a vulnerability integral formula based on a log-normal distribution, the mathematical expression of which is: in, The cascade failure probability of component i represents the calculated output. Represents the standard normal cumulative distribution function; This represents the maximum effective flooding depth of component i obtained from the preceding steps. This represents the log-mean of the modified flood vulnerability parameter; This represents the logarithmic standard deviation of the original flood vulnerability parameters for the corresponding component. In this integral model, It represents the degree of deviation of the actual disaster depth from the corrected median disaster resilience. The discrete fluctuation characteristics used to describe the component's disaster resistance capability, together with the component's failure probability under the current disaster conditions, determine the probability of the component entering a failure state. For the office equipment component on the fifteenth floor near the window, since it is located on the direct intrusion path of the enclosure breach, the logarithmic mean of its flood vulnerability is significantly reduced after correction, and the maximum effective inundation depth is formed by the accumulation of wind and rain intrusion through the breach. Therefore, its cascading failure probability will rapidly increase from the near-zero value in the original single flood model. For the car component near the entrance on the first basement floor, its maximum effective inundation depth is mainly formed by the accumulation of external backflow. Based on this, the system can also obtain its cascading failure probability under the underground space flood scenario.

[0063] After obtaining the cascading failure probabilities of each component, the system continues to read the replacement cost and functional importance weights from the vulnerability parameter vector, and performs weighted economic loss aggregation based on functional importance. The mathematical expression for this process is: in, This represents the weighted relative economic loss calculated; Represents the total number of components in the building to be evaluated; Represents the probability of cascading failure of component i; Represents the reset cost of component i recorded in the vulnerability parameter vector; The functional importance weight of component i is recorded in the vulnerability parameter vector. In this loss aggregation model, the numerator is calculated by multiplying and summing the cascade failure probability, replacement cost, and functional importance weight of each component. This achieves cumulative statistics from component-level risk to system-level weighted economic loss. The replacement cost reflects the physical and economic value of a single component, while the functional importance weight reflects the component's role in the overall building operation system. The coupling of these two factors with the cascade failure probability allows the model to avoid the biases caused by traditional cost-based loss calculations. The denominator is normalized by summing the replacement costs of all components, giving the output weighted relative economic loss a uniform scale, facilitating horizontal comparisons under different buildings and disaster conditions.

[0064] After calculating the weighted relative economic loss for a single disaster scenario, the system further combines disaster intensity input parameters and the overall building damage level to execute a multi-scenario simulation loop, and performs structured association and persistent encapsulation of the input and output results of each simulation. Specifically, after each simulation, the system writes the disaster intensity input parameters, overall building damage level, cascade failure probability of each component, and weighted relative economic loss corresponding to that scenario into a comprehensive dataset according to unified data fields, thus forming a structured sample library that can be continuously accumulated and directly used for regression fitting. Each record in this comprehensive dataset corresponds to a complete cascade disaster simulation process, including both external input disaster intensity quantities such as wind speed and rainfall intensity, as well as the intermediate state of the overall building damage level and the final output weighted relative economic loss. Through this process, the system ultimately obtains a comprehensive dataset that can be used to construct building flood vulnerability curves and support subsequent Bayesian back inference, realizing a data closed loop from single component failure calculation to multi-scenario system loss modeling.

[0065] Specifically, in step S5, based on the disaster intensity and weighted relative economic loss in the comprehensive dataset, a log-normal regression fit is performed on the discrete simulation points using the maximum likelihood estimation algorithm to obtain the building flood vulnerability curve. It should be understood that since the comprehensive dataset generated through multi-scenario simulation cycles is essentially still a series of discrete points between disaster intensity and weighted relative economic loss, although these discrete simulation points record the loss response of buildings under different typhoon and rainfall conditions, they have not yet formed a continuous functional relationship that can be directly used for risk prediction, regional comparison, and decision support. Without further statistical regression fitting, the discrete simulation results cannot be transformed into a building flood vulnerability curve with engineering applicability. Therefore, in the technical solution of this application, a continuous mapping model between disaster intensity and expected building loss is established by further performing a log-normal regression fit on the discrete simulation points using the maximum likelihood estimation algorithm based on the disaster intensity and weighted relative economic loss in the comprehensive dataset. In this way, discrete simulation outputs can be transformed into functional expressions that can be used for risk extrapolation, resilience assessment, and scenario simulation, enabling a unified characterization of the loss evolution of buildings under cascading typhoon disasters of different intensities.

[0066] More specifically, in a concrete example of this application, the building to be evaluated is a commercial office complex with twenty floors above ground and two floors below ground, featuring a glass curtain wall system on its exterior. The system first extracts the disaster intensity input parameters and corresponding weighted relative economic losses from each simulation record in the comprehensive dataset, forming a discrete simulation matrix. The disaster intensity input parameters serve as the independent variable on the horizontal axis, and the weighted relative economic losses serve as the dependent variable on the vertical axis. Subsequently, the system assumes that the building's loss response under cascading disasters follows a log-normal distribution and calls the maximum likelihood estimation algorithm to search for parameters on the scattered samples, solving for the system-level logarithmic mean and system-level logarithmic standard deviation that maximize the overall likelihood value of the observed samples. After completing the parameter estimation, the system substitutes the obtained parameters into a log-normal cumulative distribution model, thereby outputting the building's flood vulnerability curve, the mathematical expression of which is: in, The output building flood vulnerability curve is a measure of disaster intensity. The corresponding loss response value; This represents the disaster intensity input parameter extracted from the comprehensive dataset; Represents the standard normal cumulative distribution function; This represents the system-level log-mean obtained through the maximum likelihood estimation algorithm; This represents the system-level log-standard deviation obtained through the maximum likelihood estimation algorithm. In the regression fitting model described above, The original disaster intensity input parameters are mapped to logarithmic space to eliminate the influence of nonlinear distribution of intensity scale, so that disaster intensity of different orders of magnitude is statistically comparable. It reflects the median position of the building's loss response in logarithmic space. Its physical meaning lies in characterizing the key strength threshold for the building as a whole to transition from the low-loss zone to the high-loss zone. This describes the degree of dispersion of the discrete simulation lattice around the median position, reflecting the systematic uncertainty of a building under the combined effects of disaster propagation intensity, component resistance, and functional layout differences. It is expressed using the standard normal cumulative distribution function. The model can continuously transform disaster intensity measures into the expected response of weighted relative economic losses through mapping, thereby generating a monotonically increasing building flood vulnerability curve with clear engineering interpretation significance.

[0067] In its application to this commercial office complex, the system performs simulations across multiple scenarios, ranging from low wind speeds and light rainfall to extreme typhoons accompanied by high-intensity rainfall. The resulting disaster intensity input parameters and weighted relative economic losses are continuously written into the comprehensive dataset. After parameter fitting, the system generates continuous loss response curves for the building under different disaster intensities. For example, when the disaster intensity input parameters are low, the weighted relative economic loss output by the curve is close to zero, indicating that the building as a whole maintains high resilience. As the disaster intensity input parameters gradually increase, the curve enters a rapid upward trend, indicating that the cascading effects of building envelope damage, rainwater intrusion, and internal flooding begin to significantly amplify the losses. When the disaster intensity input parameters exceed a specific threshold, the curve gradually approaches a plateau, indicating that most critical components have entered a high-probability failure state. Through this process, the system successfully transforms the discrete simulation points in the comprehensive dataset into building flood vulnerability curves that can directly support pre-disaster prediction and resilience assessment, achieving an engineering-based expression from sample data to a continuous risk function.

[0068] Specifically, the method for constructing urban building flood vulnerability curves considering typhoon cascading disasters includes step S6: using a Bayesian inference model to perform an inverse probability mapping between economic losses and physical damage states to obtain the posterior probability distribution of damage states. It should be understood that in post-disaster emergency response and rapid damage assessment scenarios, what is initially obtained is often not a complete list of physical damage to each component within the building, but rather an actual estimate of relative economic losses based on on-site inspections, insurance damage assessments, or remote sensing evaluations. If this known economic loss information cannot be inversely mapped to the possible physical damage state of the building as a whole, it is difficult to provide direct quantitative basis for emergency resource allocation, repair priority determination, and insurance claims. Therefore, in the technical solution of this application, a Bayesian inference model is further used to perform an inverse probability mapping between economic losses and physical damage states to obtain the posterior probability distribution of damage states, thereby establishing a two-way inference channel from observed losses to damage states. In this way, the constructed building flood vulnerability curve model can be expanded from a simple pre-event prediction tool into a comprehensive decision-making model with post-event diagnostic capabilities, thereby significantly improving the accuracy and timeliness of rapid identification of building damage levels in urban disaster management.

[0069] Figure 5 This is a flowchart of step S6 of the method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to an embodiment of this application. Figure 5As shown in the embodiment of this application, step S6 includes: S61, based on the comprehensive dataset, the trigger frequency of each physical damage state is statistically normalized using a frequency statistics method, and the economic loss field is grouped and aggregated according to the state label to obtain the prior probability distribution of the damage state and the economic loss sequence set grouped by state; S62, based on the economic loss sequence set grouped by state, the discrete loss samples under each state are fitted with a continuous probability density to obtain the loss likelihood probability distribution; S63, based on the prior probability distribution of the damage state, the loss likelihood probability distribution, and the actual estimated value of the relative economic loss obtained after the disaster, the joint probability of each state under the observed loss condition is normalized and back-mapped using Bayes' formula to obtain the posterior probability distribution of the damage state.

[0070] More specifically, in a concrete example of this application, the building to be evaluated is a commercial office complex with twenty floors above ground and two floors below ground. The system first performs statistical normalization on the trigger frequency of each physical damage state based on the comprehensive dataset generated in the preceding steps, and then groups and aggregates the economic loss field according to the state label, thereby obtaining the prior probability distribution of the damage state and the economic loss sequence set grouped by state. Subsequently, based on the economic loss sequence set grouped by state, the system performs continuous probability density fitting on the discrete loss samples of each state to obtain the loss likelihood probability distribution. Finally, the system receives the actual estimated value of the relative economic loss obtained after the disaster as the observation condition, and inputs it together with the prior probability distribution of the damage state and the loss likelihood probability distribution into the Bayesian inference model. By normalizing and back-mapping the joint probability of each state under the observed loss condition, the system outputs the posterior probability distribution of the damage state, the mathematical expression of which is as follows: in, The posterior probability distribution of the damage state, representing the calculated output, indicates that the building is in a state of physical damage. The posterior probability; Represents the level of physical damage; Represents the actual estimated relative economic losses obtained after the disaster; Represents a known physical damage state Observation loss occurs under certain conditions The loss likelihood probability distribution takes values; Physical damage state in the prior probability distribution representing the damage state The prior probability; K represents the set of all physical damage state levels; This represents the joint probability term of all states when traversing all damage states. In the Bayesian backmap model described above, the numerator... Used to quantify the joint probability that a specific physical damage state and observed loss both hold true, where the prior probability... This reflects the fundamental tendency of such buildings to naturally reach this damage level in historical simulations and multi-scenario simulations, as shown in the loss likelihood probability distribution. This reflects the degree of matching between the observed economic loss and the damage level at that level; the denominator part The joint probabilities of all damage states are summed and normalized. This process incorporates the competing relationships of all possible damage states into a unified probabilistic framework, ensuring that the final output satisfies the constraints of a complete probability distribution. Through this normalization mapping, the system can deduce the credible probability ranking of a building at each damage level from a single observed loss value.

[0071] In the post-disaster application scenario of this commercial office complex, if the preliminary on-site assessment estimates the relative economic loss of the building to be approximately 30%, the system substitutes this value into the aforementioned Bayesian inference model. After calling the prior probability distribution of the damage state and the loss likelihood probability distribution from the comprehensive dataset, the system obtains that the posterior probability of the building being in a moderate damage state is the highest, followed by the posterior probability of being in a severe damage state, while the posterior probabilities of other damage states are relatively low. Therefore, the system can quickly output a probabilistic judgment of the overall damage state of the building even in the absence of complete component-level survey data, providing direct quantitative decision support for prioritizing the deployment of maintenance teams, dispatching emergency personnel, and post-disaster insurance loss assessment. Through this process, this application achieves a closed-loop reverse inference from the comprehensive dataset to post-disaster state diagnosis, significantly expanding the application depth and practical value of the building flood vulnerability curve model.

[0072] In summary, the method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to the embodiments of this application is clarified. First, it constructs a multi-dimensional building prototype classification system encompassing modern features such as high-rise buildings, glass curtain wall systems, and underground spaces. Then, it automatically quantifies the functional importance weights of each component through a multi-modal fusion inference network, overcoming the shortcomings of outdated traditional prototype libraries and the strong subjectivity of manual assessment. Second, it introduces a spatial aerodynamic coupling topology diagram and a cascading failure propagation model. Based on wind directionality and the wind pressure transmission relationship between components, it performs cascading amplification calculations of wind-induced failure probability. Furthermore, it replaces the traditional Boolean cliff-style threshold judgment with a soft logic aggregation mechanism, fundamentally solving the problem of underestimating the overall damage level caused by isolated component modeling. On this basis, using the overall building damage level as an intermediate bridge variable, it dynamically corrects the flood vulnerability parameters of each component and couples multi-source flood calculations to determine the equivalent inundation depth, thus opening up the entire cascading transmission path from wind-induced damage to flood failure. Finally, it constructs a vulnerability curve with both pre-event prediction and post-event diagnosis capabilities through log-normal regression fitting and Bayesian back-inference.

[0073] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for constructing urban building flood vulnerability curves considering typhoon cascading disasters, characterized in that, include: S1: Perform prototype matching and classification on the building to be evaluated to obtain the building prototype category, and perform functional importance quantification inference on each building component to obtain the vulnerability parameter vector of each component. S2: Based on the physical wind speed and reference wind speed of the typhoon in the disaster scenario, wind speed normalization calculation based on wind pressure effect is performed on each building envelope component to obtain a standardized damage intensity index. The failure probability of each building envelope component is calculated by adaptive Latin hypercube sampling based on the standardized damage intensity index and the wind-induced vulnerability parameters extracted from the vulnerability parameter vector. The overall damage level of the building is obtained by performing multiple logical assessments and overall damage level aggregation on the wind-induced failure probability of each building envelope component. This includes: calculating the aerodynamic coupling strength between each pair of building envelope components based on the component spatial coordinate vector and the disaster wind direction vector to obtain the aerodynamic coupling matrix; cascading amplification of the comprehensive survival rate of each component under the internal pressure transmission of its own damage and adjacent node damage based on the wind-induced failure probability and the aerodynamic coupling matrix to obtain the cascaded amplified failure probability; and performing soft logic aggregation and damage level assessment on the cascaded amplified failure probability to obtain the overall damage level of the building. S3: Based on the overall damage level and vulnerability parameter vector of the building, the original flood vulnerability parameters of each component are dynamically corrected by cascading effect to obtain the corrected vulnerability parameters. The maximum effective flood depth of each component is calculated by combining the multi-source flooding caused by external backflow and internal breach intrusion. S4: Based on the corrected vulnerability parameters and the maximum effective flooding depth, calculate the cascade failure probability of each component, and combine the vulnerability parameter vector and functional importance weight to perform weighted loss aggregation on the cascade failure probability to obtain the weighted relative economic loss and comprehensive dataset. S5: Based on the disaster intensity and weighted relative economic loss in the comprehensive dataset, log-normal regression is performed on the discrete simulation lattice to obtain the building flood vulnerability curve.

2. The method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to claim 1, characterized in that, It also includes step S6: using a Bayesian inference model to perform an inverse probability mapping between economic loss and physical damage status to obtain the posterior probability distribution of damage status.

3. The method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to claim 1, characterized in that, Step S1 includes: The functional attributes, structural features, environmental location, and underground space attributes of the building to be evaluated are used for dimensionality reduction matching and component instantiation of the building prototype to obtain the building prototype category and the initial component information set. Multimodal feature independent encoding is performed on the initial component information set to obtain semantic feature vectors, spatial feature vectors, and topological feature vectors; We perform attention-based feature fusion and weight inference on semantic feature vectors, spatial feature vectors, and topological feature vectors to obtain the functional importance weights of each component; The vulnerability parameter vector of each component is obtained by high-dimensional concatenation of the component's type, spatial location, vulnerability parameter set, and reset cost.

4. The method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to claim 3, characterized in that, Multimodal feature independent encoding is performed on the initial component information set to obtain semantic feature vectors, spatial feature vectors, and topological feature vectors, including: A semantic feature vector is obtained by performing deep semantic feature extraction on the component text description through a semantic encoder. Spatial feature vectors are obtained by mapping high-dimensional risk features to spatial location data through a spatial encoder. The physical connections between components are modeled by neighborhood aggregation using graph neural networks to obtain topological feature vectors.

5. The method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to claim 1, characterized in that, Step S3 includes: A two-dimensional cascade effect correction factor is identified by analyzing the spatial location information of components in the overall damage level and vulnerability parameter vector of the building to obtain the cascade effect correction factor. Based on the cascade effect correction factor and the original flood vulnerability parameters in the vulnerability parameter vector, the original logarithmic mean of the vulnerability is scaled and adjusted to obtain the corrected vulnerability parameters. Based on the breach area determined by the overall damage level of the building, external rainfall intensity data, and component spatial location information in the vulnerability parameter vector, the multi-source infiltration volume is accumulated over time to obtain the maximum effective flooding depth of each component.

6. The method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to claim 1, characterized in that, Step S4 includes: Based on the corrected vulnerability parameters and the maximum effective flooding depth, vulnerability integral calculations are performed on each component to obtain the cascade failure probability of each component. Based on the cascade failure probability and the replacement cost and functional importance weights in the vulnerability parameter vector, a weighted relative economic loss aggregation based on functional importance is performed on each component to obtain the weighted relative economic loss. Based on the weighted relative economic loss, disaster intensity input parameters, and overall building damage level, the simulation results of each simulation are structurally correlated and iteratively persistently encapsulated through multi-scenario simulation loops to obtain the weighted relative economic loss and comprehensive dataset.

7. The method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to claim 2, characterized in that, Step S6 includes: Based on the comprehensive dataset, the trigger frequency of each physical damage state is statistically normalized by frequency statistics method, and the economic loss field is grouped and aggregated according to the state label to obtain the prior probability distribution of damage state and the economic loss sequence set grouped by state. Based on the economic loss sequence set grouped by state, the continuous probability density is fitted to the discrete loss samples under each state to obtain the loss likelihood probability distribution. Based on the prior probability distribution of damage state, the probability distribution of loss likelihood, and the actual estimated value of relative economic loss obtained after the disaster, the joint probability of each state under the observed loss condition is calculated by performing normalized inverse mapping using Bayes' formula to obtain the posterior probability distribution of damage state.

8. The method for constructing urban building flood vulnerability curves considering typhoon cascading disasters according to claim 5, characterized in that, Based on the breach area determined by the overall building damage level, external rainfall intensity data, and component spatial location information in the vulnerability parameter vector, the multi-source infiltration volume is accumulated over time to obtain the maximum effective flooding depth of each component, including: Based on the overall damage level of the building, external rainfall intensity data, and component spatial location information, the depth of internal infiltration water accumulation formed by internal breaches in each component at each time is calculated. Based on the real-time depth of external surface floodwater, component installation height, and internal infiltration water depth, the equivalent inundation depth of each component at each time moment is calculated. Extreme values ​​of the equivalent inundation depth of each component during the disaster simulation period are extracted to obtain the maximum effective inundation depth of each component.

Images