Building structure earthquake dynamic response calculation method and device based on time integrogram network, equipment and medium

Abstract

The invention discloses a building structure seismic dynamic response calculation method and device based on a time integrogram network, equipment and a medium, and relates to the technical field of civil engineering, and the method comprises the steps: obtaining a finite element model and seismic oscillation time history data of a target building structure, constructing a graph structure, and obtaining initial graph data; and then mapping the initial graph data into graph input data, inputting the graph input data into a preset seismic response model for processing, and finally obtaining a seismic response displacement sequence as a dynamic response result. The preset model comprises a graph encoder, a time integral calculation module and a graph decoder, all the modules work cooperatively, efficient and accurate earthquake response prediction is achieved, calculation resource consumption and calculation time are effectively reduced, meanwhile, high precision and generalization are kept, and the real-time simulation requirement is met.

Description

Technical Field

[0001] This invention relates to the field of civil engineering technology, and in particular to a method, apparatus, equipment and medium for calculating the seismic dynamic response of building structures based on time integral graph networks. Background Technology

[0002] In the field of civil engineering, applications such as structural health monitoring, seismic performance assessment, and rapid post-disaster response require the ability to instantly acquire the response (e.g., displacement, acceleration) of large buildings or bridges under seismic dynamics. Traditional numerical simulation methods, represented by the finite element method, discretize continuous structures into a finite number of elements and nodes, establish mass, damping, and stiffness matrices, and then use time integration algorithms to solve the differential equations of motion step by step to obtain the dynamic response of the structure. While these methods offer high accuracy, they are computationally intensive and resource-intensive, making it difficult to meet real-time requirements. In contrast, neural network-based methods utilize models such as multilayer perceptrons, convolutional neural networks, or recurrent neural networks to directly learn the mapping relationship from a large number of input (e.g., load time histories) and output (e.g., displacement response) data pairs. For example, using long short-term memory networks to process time-series load data and predict future structural responses, the accuracy of the model is highly dependent on the quantity and coverage of the training data. When encountering load conditions or structural states not included in the training set, the model's predictive performance drops sharply, exhibiting poor generalization ability. Furthermore, completely "black box" models lack interpretability and it is difficult to guarantee that their predictions conform to basic physical laws.

[0003] Therefore, there is an urgent need for a method to calculate the seismic dynamic response of building structures, so as to achieve high-precision, real-time dynamic response calculation of structures while requiring less training data. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, equipment, and medium for calculating the seismic dynamic response of building structures based on time integral graph networks, aiming to solve the technical problem of how to achieve high-precision, real-time dynamic response calculation of structures while having low requirements for training data.

[0005] To achieve the above objectives, this application proposes a method for calculating the seismic dynamic response of building structures based on time integral graph networks, including: Acquire the finite element model of the target building structure and the ground motion time history data during the prediction period, wherein the finite element model includes the node information, element information, mass matrix, damping matrix and stiffness matrix of the target structure; A graph structure is constructed based on the node and element information of the finite element model, and initial graph data is obtained; Based on the initial graph data, the mass matrix, damping matrix, stiffness matrix, and seismic motion time history data are mapped to obtain graph input data; The graph input data is input into a preset seismic response model for processing to obtain the seismic response displacement sequence of the target building structure. The preset seismic response model includes a graph encoder, a time integration calculation module, and a graph decoder. The graph encoder includes a residual network module and a convolutional residual network module. The residual network module includes a first preset number of residual blocks, and the convolutional residual network module includes a first preset number of convolutional residual blocks. The time integration module includes a parameter calculation unit, an edge update unit, and a node recursion unit. The graph decoder includes a first preset number of fully connected layers and an output layer. The seismic response displacement sequence is used as the seismic dynamic response result of the target building structure.

[0006] In one embodiment, the step of inputting the graph input data into a preset seismic response model for processing to obtain the seismic response displacement sequence of the target building structure, wherein the preset seismic response model includes a graph encoder, a time integration calculation module, and a graph decoder, includes: The graph input data is reduced in dimensionality by the graph encoder to obtain a low-dimensional hidden graph. The time integration calculation module performs multi-step recursive calculations on the low-dimensional hidden graph to obtain the updated hidden graph node features at each time step. The target building structure's seismic response displacement sequence is obtained by decoding and restoring the target building structure based on the updated hidden graph node features using the graph decoder.

[0007] In one embodiment, the step of performing dimensionality reduction processing on the graph input data using the graph encoder to obtain a low-dimensional hidden graph includes: The graph node features in the graph input data are input into the residual network module. The residual block performs feature compression and dimensionality reduction on the graph node features to obtain low-dimensional node features. The dimension of the low-dimensional node features is a preset ratio of the dimension of the graph node features. The graph edge features in the graph input data are input into the convolutional residual network module to perform convolution operations and dimension reduction on the graph edge features to obtain low-dimensional edge features, wherein the dimension of the low-dimensional edge features is a preset ratio of the dimension of the graph edge features. The adjacency matrix in the graph input data is sparsified to obtain the processed adjacency matrix; The low-dimensional node features, low-dimensional edge features, and the processed adjacency matrix are combined to obtain a low-dimensional hidden graph.

[0008] In one embodiment, the step of performing multi-step recursive calculations on the low-dimensional hidden graph through the time integration calculation module to obtain the updated hidden graph node features at each time step includes: Extract low-dimensional edge features and low-dimensional node features from the low-dimensional hidden graph; Set an edge update function and a node update function. The edge update function adopts a convolutional neural network, which includes a second preset number of convolutional layers and a third preset number of activation layers. The node update function adopts a fully connected neural network, which includes a second preset number of fully connected layers and a first preset number of activation layers. The low-dimensional edge features are input into the edge update function to optimize the low-dimensional edge features, resulting in updated edge features. The low-dimensional node features are recursively updated in the first step to obtain the hidden graph node features updated in the first time step. Based on the updated edge features, the low-dimensional node features of adjacent graph nodes in the low-dimensional hidden graph are aggregated through the node update function to obtain the graph nodes for the next time step. The updated hidden graph node features of the first time step are used as the initial node features of the graph nodes of the next time step. The update and aggregation steps are repeated until the calculation of all time steps in the prediction period is completed, and the updated hidden graph node features of each time step are obtained.

[0009] In one embodiment, the step of obtaining the seismic response displacement sequence of the target building structure by decoding and reconstructing the target building structure based on the updated hidden graph node features using the graph decoder includes: The updated hidden graph node features at each time step are sequentially input into the fully connected layer of the graph decoder for dimension recovery and feature mapping to obtain the physical domain features corresponding to each time step. Displacement information is extracted from the physical domain features corresponding to each time step. The displacement information includes the x-direction displacement, y-direction displacement, and z-direction displacement of the target building structure's corresponding structural degrees of freedom at the current time step. The validity of the displacement information at each time step is verified, and the verification results are obtained. When the value of the displacement information is within the preset displacement threshold, the displacement information of all valid time steps is integrated in chronological order to obtain the seismic response displacement sequence of the target building structure.

[0010] In one embodiment, the step of constructing a graph structure based on the node information and element information of the finite element model and obtaining initial graph data includes: Structural degrees of freedom are extracted from the node information of the finite element model, and the physical coordinates and mechanical properties corresponding to each structural degree of freedom are determined. Each structural degree of freedom is mapped to a graph node, and the graph node is associated with the physical coordinates and mechanical properties of the corresponding structural degree of freedom; Extract the structural element connection relationships from the element information of the finite element model, and determine the range of structural degrees of freedom for each structural element connection; Based on the range of structural degrees of freedom connected by the structural units, graph edges are set between the corresponding graph nodes, and the graph edges are associated with the type and size of the corresponding structural units; An adjacency matrix is ​​generated based on the stiffness matrix of the finite element model, wherein the element value of the adjacency matrix is ​​a first preset value indicating that the corresponding graph node has a mechanical connection, and the element value is a second preset value indicating that there is no mechanical connection. By integrating the graph nodes, graph edges, and adjacency matrix, initial graph data is obtained.

[0011] In one embodiment, the step of mapping the mass matrix, damping matrix, stiffness matrix, and seismic motion time history data based on the initial graph data to obtain graph input data includes: Extract the graph edge list from the initial graph data and determine the structural unit corresponding to each graph edge; The sub-matrix in the mass matrix corresponding to the structural unit, the sub-matrix in the damping matrix corresponding to the structural unit, and the sub-matrix in the stiffness matrix corresponding to the structural unit are mapped to the first side feature component, the second side feature component, and the third side feature component of each edge of the graph. The graph edge features are obtained by combining the first edge feature components, the second edge feature components, and the third edge feature components. Extract a list of graph nodes from the initial graph data and determine the structural degrees of freedom corresponding to each graph node; The load time sequence corresponding to the structural degrees of freedom in the earthquake motion time history data is mapped to the nodal feature components of each graph node; Obtain the initial displacement data of the target building structure at the start of the prediction period; The initial displacement data is added to the node feature components corresponding to the graph node to obtain the graph node features; The graph edge features, the graph node features, and the adjacency matrix in the initial graph data are integrated to obtain the graph input data.

[0012] Furthermore, to achieve the above objectives, this application also proposes a building structure seismic dynamic response calculation device based on a time integral graph network, the building structure seismic dynamic response calculation device based on a time integral graph network comprising: The acquisition module is used to acquire the finite element model of the target building structure and the ground motion time history data during the prediction period, wherein the finite element model includes the node information, element information, mass matrix, damping matrix and stiffness matrix of the target building structure. The construction module is used to construct the graph structure based on the node information and element information of the finite element model, and to obtain the initial graph data; The mapping module is used to map the mass matrix, damping matrix, stiffness matrix, and seismic motion time history data based on the initial map data to obtain map input data; The processing module is used to input the graph input data into a preset seismic response model for processing to obtain the seismic response displacement sequence of the target building structure. The preset seismic response model includes a graph encoder, a time integration calculation module, and a graph decoder. The graph encoder includes a residual network module and a convolutional residual network module. The residual network module includes a first preset number of residual blocks. The convolutional residual network module includes a first preset number of convolutional residual blocks. The time integration module includes a parameter calculation unit, an edge update unit, and a node recursion unit. The graph decoder includes a first preset number of fully connected layers and an output layer. The results module is used to use the seismic response displacement sequence as the seismic dynamic response result of the target building structure.

[0013] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the method for calculating the seismic dynamic response of building structures based on time integral graph networks as described above.

[0014] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the method for calculating the seismic dynamic response of building structures based on time integral graph networks as described above.

[0015] This application acquires the finite element model and seismic time history data of the target building structure, constructs a graphical structure, and obtains initial graphical data. Then, the initial graphical data is mapped to graphical input data, which is input into a preset seismic response model for processing, ultimately yielding a seismic response displacement sequence as the dynamic response result. The preset model includes a graphical encoder, a time integration calculation module, and a graphical decoder. These modules work collaboratively to achieve efficient and accurate seismic response prediction, effectively reducing computational resource consumption and computation time while maintaining high accuracy and generalization, meeting the requirements of real-time simulation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the first embodiment of the seismic dynamic response calculation method for building structures based on time integral graph networks in this application. Figure 2 This is a 21-story finite element model frame structure diagram of the first embodiment of the seismic dynamic response calculation method for building structures based on time integral graph networks in this application; Figure 3 This is a training seismic load time history curve of the first embodiment of the seismic dynamic response calculation method for building structures based on time integral graph networks in this application; Figure 4 This is a time history curve of predicted seismic load from the first embodiment of the seismic dynamic response calculation method for building structures based on time integral graph networks in this application. Figure 5 This is a flowchart illustrating the second embodiment of the seismic dynamic response calculation method for building structures based on time integral graph networks in this application. Figure 6 This is a schematic diagram of the modular structure of the building structure seismic dynamic response calculation device based on time integral graph network of this application; Figure 7 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the seismic dynamic response calculation method for building structures based on time integral graph networks in the embodiments of this application.

[0018] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0020] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0021] In the era of Industry 4.0, digital twin technology is playing an increasingly important role in industrial production and operation, with real-time evolution of structural states becoming a crucial element. However, traditional numerical methods suffer from high computational resource consumption and excessive computation time when calculating structural dynamic responses, making it difficult to meet real-time requirements. In recent years, neural network-based methods have gradually emerged, but existing methods are either highly dependent on data or inefficient when dealing with complex structures.

[0022] Therefore, this application proposes a method for calculating the seismic dynamic response of building structures based on time integral graph networks to solve the above problems. The main solution of this application embodiment is as follows: A finite element model of the target building structure and seismic motion time history data within the prediction period are obtained. The finite element model includes node information, element information, mass matrix, damping matrix, and stiffness matrix of the target structure. A graph structure is constructed based on the node and element information of the finite element model to obtain initial graph data. The mass matrix, damping matrix, stiffness matrix, and seismic motion time history data are mapped based on the initial graph data to obtain graph input data. The graph input data is input into a preset seismic response model for processing to obtain the seismic response displacement sequence of the target building structure. The preset seismic response model includes a graph encoder, a time integration calculation module, and a graph decoder. The graph encoder includes a residual network module and a convolutional residual network module. The residual network module includes a first preset number of residual blocks, and the convolutional residual network module includes a first preset number of convolutional residual blocks. The time integration module includes a parameter calculation unit, an edge update unit, and a node recursion unit. The graph decoder includes a first preset number of fully connected layers and an output layer. The seismic response displacement sequence is used as the seismic dynamic response result of the target building structure.

[0023] Based on the above, this application also provides a method for calculating the seismic dynamic response of building structures based on time integral graph networks, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the seismic dynamic response calculation method for building structures based on time integral graph networks according to this application. In this embodiment, the seismic dynamic response calculation method for building structures based on time integral graph networks includes steps S10 to S50: Step S10: Obtain the finite element model of the target building structure and the ground motion time history data within the prediction period.

[0024] It should be noted that the target building structure refers to various civil, public, or industrial building structures for which seismic dynamic response prediction is required, such as bridges and residential buildings. The finite element model includes the node information, element information, mass matrix, damping matrix, and stiffness matrix of the target structure. First, in terms of acquiring the finite element model, the computing system needs to retrieve the corresponding finite element model file from the design database or engineering simulation platform of the target building structure. This finite element model must fully cover the geometric features and mechanical properties of the structure. The node information includes the three-dimensional spatial coordinates (accurate to the millimeter level) of all key nodes of the structure and the structural degrees of freedom (such as translational and rotational degrees of freedom) corresponding to each node, and the node number and the structural component to which it belongs (such as beam node and column node) must be labeled. The element information must clearly specify the finite element type (such as beam element, column element, floor slab element), cross-sectional dimensions (such as the cross-sectional height and width of beam element), material parameters (such as the elastic modulus of steel and Poisson's ratio of concrete), and the connection relationship between elements and nodes for each structural component, ensuring that it can accurately reflect the actual topology of the structure. Simultaneously, the finite element model must also include three core mechanical matrices: the mass matrix, the damping matrix, and the stiffness matrix. The mass matrix, using either a uniform mass matrix or a lumped mass matrix, is generated based on the mass distribution characteristics of the structural components, reflecting the inertial response of the structure under dynamic loads. The damping matrix, using Rayleigh damping, is generated through a pre-defined... , coefficients (e.g.) =0.01、 =0.01) is calculated by combining the mass matrix and stiffness matrix, reflecting the energy dissipation effect during structural vibration; the stiffness matrix is ​​generated by assembling the element stiffness matrix, reflecting the structure's ability to resist deformation, and its dimension is consistent with the total number of degrees of freedom of the structure.

[0025] Furthermore, regarding the acquisition of seismic ground motion time history data, the calculation system needs to select seismic ground motion time history data that conforms to the local seismic fortification intensity (e.g., 7 degrees, 8 degrees) and design earthquake group, based on the seismic ground motion parameter zoning map of the area where the target building structure is located. This data needs to be presented in the form of acceleration time history curves, with a time span covering the preset response prediction period (e.g., 10s, 15s), and a time step set to a preset small step size (e.g., 0.02s) to ensure accurate reflection of the variation of seismic load over time.

[0026] Step S20: Construct a graph structure based on the node information and element information of the finite element model, and obtain initial graph data.

[0027] It should be noted that this process aims to transform the discretization concept in traditional finite element analysis into a topological representation that can be processed by graph neural networks.

[0028] Furthermore, step S20 also includes: first, extracting structural degrees of freedom from the node information of the finite element model, and determining the physical coordinates and mechanical properties corresponding to each structural degree of freedom. Specifically, structural degrees of freedom are independent parameters describing the deformation or motion that a structure may undergo under dynamic loads. For example, in three-dimensional space, each node may have three translational degrees of freedom and three rotational degrees of freedom. Physical coordinates are used to determine the position of the node in space, while mechanical properties include parameters such as mass and damping, which are crucial for describing the response behavior of the structure under dynamic loads.

[0029] Next, each structural degree of freedom is mapped to a graph node, and the graph node is associated with the physical coordinates and mechanical properties of the corresponding structural degree of freedom. Specifically, a graph node is the basic building block of a graph structure, and each graph node represents a structural degree of freedom, with its feature vector containing the physical coordinates and mechanical properties of that degree of freedom. This mapping method enables the graph neural network to directly process physical information related to the structural degree of freedom, providing a foundation for subsequent dynamic response prediction.

[0030] Next, the structural element connection relationships are extracted from the element information of the finite element model to determine the range of structural degrees of freedom for each connection. Specifically, the element information describes the connection relationships between nodes and the physical properties of the elements, such as stiffness and damping. The structural element connection relationships define which nodes are connected by elements, while the range of structural degrees of freedom specifies the specific degrees of freedom involved in these connections. For example, a beam element may connect two nodes and involve the translational degrees of freedom of those nodes.

[0031] Subsequently, graph edges are set between the corresponding graph nodes according to the range of structural degrees of freedom connected by the structural units. These graph edges are associated with the type and size of the corresponding structural units. Specifically, graph edges represent the connection relationships between graph nodes, and their feature vectors contain information such as the type and size of the structural units. This information is crucial for describing the interactions between nodes; for example, different types and sizes of units may have different effects on force transmission between nodes.

[0032] Then, an adjacency matrix is ​​generated based on the stiffness matrix of the finite element model. In this adjacency matrix, elements with a first preset value indicate a mechanical connection between corresponding graph nodes, while elements with a second preset value indicate no mechanical connection. Specifically, the stiffness matrix is ​​a crucial component of the finite element model, describing the deformation characteristics of the structure under stress. Generating the adjacency matrix from the stiffness matrix ensures that the graph structure accurately reflects its mechanical behavior. The adjacency matrix is ​​a two-dimensional matrix whose elements indicate whether a mechanical connection exists between graph nodes. For example, if a mechanical connection exists between two nodes, the corresponding adjacency matrix element value is 1; if there is no mechanical connection between two nodes, the corresponding adjacency matrix element value is 0.

[0033] Finally, the graph nodes, edges, and adjacency matrix are integrated to obtain the initial graph data. Specifically, the initial graph data contains the topological information and physical properties of the structure, providing the necessary input for subsequent graph neural network processing. This process not only preserves the physical characteristics of the finite element model but also provides a structured data format for the learning and prediction of the graph neural network. The graph structure constructed in this way can effectively characterize the mechanical behavior of the structure, laying a solid foundation for achieving efficient, low-data-dependency prediction of structural dynamic responses.

[0034] Step S30: Based on the initial graph data, the mass matrix, damping matrix, stiffness matrix, and seismic motion time history data are mapped to obtain graph input data.

[0035] It should be noted that this process not only requires accurately converting physical information into a form that graph neural networks can process, but also requires ensuring that this information is reasonably represented in the graph structure so that subsequent calculations and predictions can accurately reflect the behavior of the structure under seismic loads.

[0036] Furthermore, step S30 also includes: first, extracting a graph edge list from the initial graph data to determine the structural unit corresponding to each graph edge. The graph edge list is a set of all edges in the graph structure, and each graph edge represents the connection relationship between two graph nodes, corresponding to the two nodes connected by the element in the finite element model.

[0037] Next, the sub-matrices in the mass matrix, the damping matrix, and the stiffness matrix corresponding to the structural units are mapped to the first, second, and third feature components of each graph edge. The mass matrix describes the mass distribution of the structure, the damping matrix describes the energy dissipation characteristics of the structure, and the stiffness matrix describes the elastic characteristics of the structure. These sub-matrices are associated with specific structural units and reflect the physical properties of that unit. By mapping these sub-matrices to the feature components of the graph edges, the physical properties of the structural units can be transformed into a form that can be processed by a graph neural network.

[0038] Next, the first, second, and third edge feature components are combined to obtain the graph edge features. Graph edge features are attribute vectors of graph edges, containing all physical information related to that edge. This combination method allows graph neural networks to consider multiple physical properties such as mass, damping, and stiffness when processing graph edges, thereby more accurately simulating the behavior of structural units.

[0039] Subsequently, a graph node list is extracted from the initial graph data to determine the structural degree of freedom corresponding to each graph node. The graph node list is the set of all nodes in the graph structure, and each graph node represents a structural degree of freedom. By extracting the graph node list, the representation of each structural degree of freedom in the graph structure can be clearly defined, providing a foundation for subsequent feature mapping.

[0040] Then, the load-time sequence corresponding to the structural degrees of freedom in the seismic time history data is mapped to the nodal feature components of each graph node. The seismic time history data describes the change of seismic action over time, while the load-time sequence represents the specific manifestation of seismic motion in a particular degree of freedom. By mapping the load-time sequence to nodal feature components, the external excitation information of seismic motion can be transformed into a form that the graph neural network can process, enabling the graph neural network to receive the corresponding seismic motion information at each time step.

[0041] Finally, the initial displacement data of the target building structure at the start of the prediction period is obtained, and this initial displacement data is added to the node feature components of the corresponding graph nodes to obtain graph node features. Then, the graph edge features, the graph node features, and the adjacency matrix in the initial graph data are integrated to obtain graph input data. Specifically, graph node features are attribute vectors of graph nodes, containing all physical information and initial state information related to that node. By adding the initial displacement data to the node feature components, the graph neural network can consider both the external excitation of seismic motion and the initial state of the structure when processing graph nodes, thereby more accurately simulating the behavior of the structural degrees of freedom. Furthermore, the graph input data is the input to the graph neural network, containing the structure's topological information, physical properties, initial state, and external excitation information. By integrating this information, a comprehensive perspective can be provided to the graph neural network, enabling it to predict dynamic responses while considering all relevant factors.

[0042] Step S40: Input the graph input data into a preset seismic response model for processing to obtain the seismic response displacement sequence of the target building structure.

[0043] It should be noted that the preset seismic response model includes a graph encoder, a time integration calculation module, and a graph decoder. The graph encoder includes a residual network module and a convolutional residual network module. The residual network module includes a first preset number of residual blocks, and the convolutional residual network module includes a first preset number of convolutional residual blocks. The time integration module includes a parameter calculation unit, an edge update unit, and a node recursion unit. The graph decoder includes a first preset number of fully connected layers and an output layer.

[0044] Specifically, the graph encoder is the first part of the model. Its task is to extract key features from the graph input data and encode these features into a more tractable form. The graph encoder consists of two main modules: a residual network module and a convolutional residual network module. The residual network module consists of a first predetermined number of residual blocks. The design of these residual blocks allows the network to learn deeper feature representations during training, while avoiding the gradient vanishing problem, thereby improving the training efficiency and stability of the model. The convolutional residual network module further enhances the model's ability to process graph-structured data. It contains a first predetermined number of convolutional residual blocks. These blocks, through convolution operations, can effectively capture the spatial relationships between nodes and edges in the graph, further enriching the feature representation.

[0045] Then, after processing by the graph encoder, the obtained feature data is fed into the time integration calculation module. This module is the core of the entire model; it is responsible for calculating and updating the state of each node in the graph based on the time series, thereby simulating the dynamic response process of the structure under seismic loading. The time integration calculation module consists of a parameter calculation unit, an edge update unit, and a node recursion unit. The parameter calculation unit is responsible for calculating the various parameters required during the time integration process, which are crucial for accurately simulating the physical process. The edge update unit updates the edge information based on the characteristics of the graph edges and the states of adjacent nodes, ensuring that the interactions between nodes in the graph structure are correctly reflected. The node recursion unit recursively calculates the node state for the next time step based on the updated edge information and the current node state. In this way, the model can gradually construct the changes in node states over the entire time series, i.e., the dynamic response process of the structure.

[0046] Finally, the node state data obtained by the time integration calculation module is passed to the graph decoder. The graph decoder's task is to decode these encoded feature data and restore them into an interpretable seismic response displacement sequence. The graph decoder consists of a first preset number of fully connected layers and an output layer. The fully connected layers are responsible for further transforming and integrating the feature data to extract key information related to the seismic response displacement sequence. The output layer then transforms this information into a specific displacement sequence, i.e., the response result of the target building structure under seismic loading.

[0047] By inputting the graph data into a sophisticated and fully functional pre-defined seismic response model, we can obtain the seismic response displacement sequence of the target building structure. This process not only fully utilizes the advantages of graph neural networks in processing complex structural data, but also accurately simulates the dynamic response process of the structure through a time integration calculation module, ultimately obtaining highly accurate and reliable prediction results through a graph decoder.

[0048] Step S50: The seismic response displacement sequence is used as the seismic dynamic response result of the target building structure.

[0049] It should be noted that the seismic response displacement sequence obtained after processing by the pre-set seismic response model is a detailed dynamic description of the target building structure's response to seismic action during the prediction period. This displacement sequence not only includes the displacement changes of the structure under seismic loading but also implicitly contains dynamic information such as velocity and acceleration, because displacement is obtained from acceleration through time integration, while velocity is the time derivative of displacement. This information is crucial for evaluating the seismic performance of the structure.

[0050] Furthermore, using the seismic response displacement sequence as the seismic dynamic response result of the target building structure means that this sequence can be directly applied to various aspects such as seismic design, performance evaluation, and disaster prevention. For example, in seismic design, engineers can use this sequence to evaluate the structure's response under different seismic intensities, thereby determining whether the structure meets the requirements of seismic codes. In performance evaluation, this sequence can help identify weak points in the structure, providing a basis for structural reinforcement and modification. In disaster prevention, by analyzing this sequence, the potential damage to the structure during an earthquake can be predicted, allowing for the implementation of appropriate protective measures in advance.

[0051] Furthermore, this seismic dynamic response result can be used to validate and improve the finite element model. By comparing the predicted displacement sequence with actual observation data or experimental results, the accuracy and reliability of the finite element model can be evaluated. If there is a significant deviation between the predicted results and the actual data, the finite element model can be adjusted and optimized accordingly to improve the model's prediction accuracy.

[0052] Furthermore, the model proposed in this embodiment is verified, such as... Figure 2 The 21-story finite element model frame structure diagram shown is illustrated. The frame structure is 98m high and has a planar dimension of 48.6m × 48.6m. The structural finite element model uses full beam elements for simulation. The finite element model has 2984 nodes, 7656 elements, 17400 degrees of freedom, and a stiffness matrix dimension of 17400 × 17400. The main parameters of the finite element model are shown in Table 1, the finite element model characteristic table. Table 1. Properties of the Finite Element Model The data from the first 15 seconds of a specific seismic load were applied to the aforementioned finite element model, such as... Figure 3 The training earthquake load time history curve is shown in the figure, and the training dataset is obtained.

[0053] In the preset seismic response model of this embodiment, the relevant parameters of the model are determined based on the degrees of freedom of the finite element model described above. Specific parameter settings are shown in Table 2, the preset seismic response model parameter table: Table 2 Preset Seismic Response Model Parameter Table Using the same seismic load data from the first 15 seconds as described above, applied to the aforementioned preset seismic response model, such as... Figure 4 The predicted earthquake load time history curve is shown in the figure, and the prediction results are obtained.

[0054] By comparing the calculation methods of the preset seismic response model and the finite element model in this embodiment, the following results are obtained: Table 3 shows the prediction result index, and Table 4 shows the calculation efficiency results. Table 3 Prediction Result Indicators Table 4 Calculation Efficiency Results Table 3 lists the average and contrast values ​​of the Normalized Mean Square Error (NMSE) and the coefficient of determination (R²). The average NMSE is 0.00129 with a contrast of 0.00116, and the average R² is 0.986 with a contrast of 0.012, demonstrating that the patented method has high prediction accuracy and good fit. Table 4 compares the performance of the traditional finite element method and the patented method in terms of computation time, number of time steps, and efficiency improvement. The patented method has a computation time of only 8.2 seconds, compared to 1385.7 seconds for the finite element method, representing a 169-fold improvement in efficiency, and processes 5736 time steps, significantly improving computational efficiency. These data indicate that the patented method not only has high accuracy in predicting structural dynamic response but also offers significant improvement in computation speed.

[0055] This embodiment acquires the finite element model and seismic time history data of the target building structure, constructs a graphical structure, and obtains initial graphical data. Then, the initial graphical data is mapped to graphical input data, which is input into a preset seismic response model for processing. Finally, the seismic response displacement sequence is obtained as the dynamic response result. The preset model includes a graphical encoder, a time integration calculation module, and a graphical decoder. These modules work collaboratively to achieve efficient and accurate seismic response prediction, effectively reducing computational resource consumption and computation time while maintaining high accuracy and generalization, meeting the requirements of real-time simulation.

[0056] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 5 The method for calculating the seismic dynamic response of building structures based on time integral graph networks, step S40, further includes steps S201 to S203: Step S201: The graph input data is dimensionality reduced by the graph encoder to obtain a low-dimensional hidden graph.

[0057] It should be noted that the graph encoder, through its internal residual network module and convolutional residual network module, compresses the high-dimensional features contained in the graph input data, extracts key information, and maps it to a low-dimensional space, thus obtaining a low-dimensional hidden graph. This dimensionality reduction process not only reduces data storage requirements and computational complexity, but also improves the model's ability to capture important features, which is helpful for subsequent time integration calculations and earthquake response prediction.

[0058] Further, step S201 includes: First, inputting the graph node features from the graph input data into a residual network module, and then performing feature compression and dimensionality reduction on the graph node features through the residual blocks to obtain low-dimensional node features, wherein the dimension of the low-dimensional node features is a preset proportion of the dimension of the graph node features. Specifically, the graph node features are input into the residual network module. The residual network module consists of multiple residual blocks, each of which introduces skip connections, enabling the network to learn the residual mapping between input features and output features. This design helps alleviate the gradient vanishing problem in deep network training and promotes the network to learn deeper feature representations. In the residual blocks, feature compression and dimensionality reduction are performed on the graph node features through a series of linear transformations and nonlinear activation functions to obtain low-dimensional node features. This process not only reduces the dimensionality of the data but also enhances the expressive power of the features, enabling the low-dimensional node features to effectively represent the original high-dimensional features with a preset proportion of dimensionality.

[0059] Secondly, the graph edge features from the input graph data are input into a convolutional residual network module to perform convolution operations and dimensionality reduction on the graph edge features, resulting in low-dimensional edge features. The dimension of these low-dimensional edge features is a preset proportion of the dimension of the graph edge features. Specifically, the processing of graph edge features is similar to that of graph node features, but it is processed using a convolutional residual network module. The convolutional residual network module uses convolution operations to capture spatial dependencies in the graph structure, which is crucial for understanding the interactions between nodes. By performing convolution operations and dimensionality reduction on the graph edge features, low-dimensional edge features are obtained. These low-dimensional edge features, also with a preset proportion of dimension, retain the key information of the original graph edge features, providing a more compact data representation for subsequent time integration calculations.

[0060] Then, the adjacency matrix in the graph input data is sparsified to obtain the processed adjacency matrix. Specifically, after processing the graph node features and graph edge features, the adjacency matrix in the graph input data is sparsified. The adjacency matrix describes the connection relationships between nodes in the graph, and its sparsification can further reduce the data storage requirements and computational complexity. Sparsification typically involves identifying and removing connections that contribute little to the model's predictions, resulting in a more concise adjacency matrix that still accurately represents the graph's topology.

[0061] Finally, the low-dimensional node features, low-dimensional edge features, and the processed adjacency matrix are combined to obtain a low-dimensional hidden graph. Specifically, the low-dimensional hidden graph is a dimensionality-reduced data representation that not only contains the key physical information of the nodes and edges in the graph but also preserves the graph's topological structure. This data representation provides an efficient and compact input for subsequent time integration calculations and seismic response prediction, enabling the model to significantly reduce computational resource consumption while maintaining high prediction accuracy.

[0062] Step S202: The time integration calculation module performs multi-step recursive calculation on the low-dimensional hidden graph to obtain the updated hidden graph node features at each time step.

[0063] It should be noted that the time integration calculation module is the core component for performing seismic response prediction. It is responsible for performing multi-step recursive calculations on the low-dimensional hidden graph to simulate the evolution of the structural dynamic response under seismic loading. This module achieves time-series updates of the hidden graph node features through the collaborative work of parameter calculation units, edge update units, and node recursive units.

[0064] Furthermore, step S202 also includes: firstly, extracting low-dimensional edge features and low-dimensional node features from the low-dimensional hidden graph. The low-dimensional edge features and low-dimensional node features extracted from the low-dimensional hidden graph are the basis for time integration calculations. These features contain key information about structural elements and degrees of freedom, and they are used to simulate the dynamic response of the structure under seismic loading.

[0065] Secondly, edge update functions and node update functions are set. The edge update function uses a convolutional neural network, which includes a second preset number of convolutional layers and a third preset number of activation layers. The node update function uses a fully connected neural network, which includes a second preset number of fully connected layers and a first preset number of activation layers. Specifically, the convolutional layers are responsible for extracting local features from the edge features, while the activation layers introduce nonlinearity, enabling the network to learn more complex feature representations. In this way, the edge update function can optimize low-dimensional edge features to obtain updated edge features that more accurately reflect the interactions between structural units. The node update function is implemented using a fully connected neural network, which includes a second preset number of fully connected layers and a first preset number of activation layers. The fully connected layers are responsible for mapping node features to a new high-dimensional space, while the activation layers further enhance the network's nonlinear expressive power. The node update function obtains the graph node features for the next time step by aggregating the low-dimensional node features of adjacent graph nodes. This process simulates the dynamic changes in the structural degrees of freedom under seismic loading. Then, the low-dimensional edge features are input into the edge update function to optimize them, resulting in updated edge features. Next, a first-step recursive update is performed on the low-dimensional node features to obtain the hidden graph node features updated at the first time step. Based on the updated edge features, the low-dimensional node features of adjacent graph nodes in the low-dimensional hidden graph are aggregated using the node update function to obtain the graph nodes for the next time step. Specifically, the first-step recursive update of the low-dimensional node features yields the hidden graph node features updated at the first time step. This step is fundamental to subsequent recursive calculations, providing the model with initial node state information. Then, based on the updated edge features, the low-dimensional node features of adjacent graph nodes in the low-dimensional hidden graph are aggregated using the node update function to obtain the graph node features for the next time step. This aggregation process considers the interactions between structural units, enabling the model to more accurately predict the dynamic response of the structure. Finally, the hidden graph node features updated at the first time step are used as the initial node features for the graph nodes in the next time step, and the update and aggregation steps are repeated until the calculations for all time steps within the prediction period are completed, yielding the updated hidden graph node features for each time step. Specifically, the hidden graph node features updated in the first time step are used as the initial node features for the graph nodes in the next time step. This update and aggregation process is repeated until the calculations for all time steps within the prediction period are completed. This multi-step recursive calculation process simulates the evolution of the structural dynamic response under seismic loading, obtaining the updated hidden graph node features for each time step. These features not only contain dynamic information such as displacement, velocity, and acceleration of the structure at each time step, but also reflect the interactions and energy transfer between structural elements.

[0066] In this way, the method of this embodiment can process a large number of time steps in a short time, providing a fast and accurate tool for seismic design and performance evaluation of structures. The efficiency of this method lies in its ability to significantly reduce computational resource consumption while maintaining high prediction accuracy. Furthermore, this method can adapt to different seismic scenarios and structural types, exhibiting excellent generalization ability.

[0067] Step S203: The target building structure's seismic response displacement sequence is obtained by decoding and restoring the target building structure based on the updated hidden graph node features using the graph decoder.

[0068] It's important to note that the graph decoder's role is to convert the updated hidden graph node features from the time integral calculation module back into actual physical quantities, namely, the seismic response displacement sequence of the target building structure. This decoding process is a crucial step in mapping low-dimensional hidden graph node features back to high-dimensional space, involving the reconstruction of spatial information and physical details lost during the encoding process. Graph decoders typically consist of multiple fully connected layers that learn the complex mapping relationship from low-dimensional features to high-dimensional outputs and introduce nonlinearity through activation functions to enhance the model's expressive power.

[0069] Further, step S203 includes: First, the updated hidden graph node features of each time step are sequentially input into the fully connected layer of the graph decoder for dimensionality recovery and feature mapping, thereby obtaining the physical domain features corresponding to each time step. Specifically, the fully connected layer performs dimensionality recovery and feature mapping on the input low-dimensional features through a series of linear combinations of weights and biases, and subsequent nonlinear activation functions. This process not only recovers the high-dimensional characteristics of the original data but also ensures that the mapped features are physically reasonable. Processing the updated hidden graph node features of each time step in chronological order ensures that the decoding process matches the temporal evolution of the seismic action, thereby obtaining the physical domain features corresponding to each time step.

[0070] Secondly, displacement information is extracted from the physical domain features corresponding to each time step. This displacement information includes the x-direction, y-direction, and z-direction displacements of the target building structure's corresponding structural degrees of freedom at the current time step. Specifically, extracting displacement information from the physical domain features is the process of converting the decoded data into specific physical quantities. The displacement information includes the x, y, and z-direction displacements of the target building structure at each time step; this information is crucial for evaluating structural response and conducting structural design.

[0071] Then, the displacement information at each time step is validated to obtain the validation results. Specifically, this step typically involves checking the reasonableness of the displacement values. For example, by comparing the displacement values ​​with preset displacement thresholds, outliers that may be caused by calculation errors or model inaccuracies can be identified. Validation helps improve the reliability of the prediction results, ensuring that only reasonable displacement information is used to construct the final seismic response displacement sequence.

[0072] Finally, when the displacement information values ​​are within a preset displacement threshold, the displacement information from all valid time steps is integrated chronologically to obtain the seismic response displacement sequence of the target building structure. Specifically, when the displacement information values ​​are within the preset displacement threshold, this information is considered valid and integrated into the seismic response displacement sequence. This integration process is performed chronologically, ensuring the temporal continuity and consistency of the displacement sequence. The resulting seismic response displacement sequence is a complete dynamic response description of the target building structure under seismic loading, and can be used for various aspects such as seismic performance assessment, design optimization, and disaster prevention.

[0073] This embodiment uses a graph encoder to reduce the dimensionality of the initial graph data to generate a low-dimensional hidden graph. The time integral calculation module is used to recursively update the node features of the hidden graph in multiple steps, and the graph decoder is used to reconstruct the seismic response displacement sequence of the target structure. This improves computational efficiency while ensuring high accuracy and good generalization ability, and significantly reduces the dependence on large-scale training data.

[0074] Based on the first embodiment of this application, this application also provides a device for calculating the seismic dynamic response of building structures based on a time integral graph network. Please refer to... Figure 6 The device includes: The acquisition module 10 is used to acquire the finite element model of the target building structure and the ground motion time history data during the prediction period, wherein the finite element model includes the node information, element information, mass matrix, damping matrix and stiffness matrix of the target building structure.

[0075] The construction module 20 is used to construct a graph structure based on the node information and element information of the finite element model and obtain initial graph data.

[0076] The mapping module 30 is used to map the mass matrix, damping matrix, stiffness matrix, and seismic motion time history data based on the initial map data to obtain map input data.

[0077] The processing module 40 is used to input the graph input data into a preset seismic response model for processing to obtain the seismic response displacement sequence of the target building structure. The preset seismic response model includes a graph encoder, a time integration calculation module, and a graph decoder. The graph encoder includes a residual network module and a convolutional residual network module. The residual network module includes a first preset number of residual blocks. The convolutional residual network module includes a first preset number of convolutional residual blocks. The time integration module includes a parameter calculation unit, an edge update unit, and a node recursion unit. The graph decoder includes a first preset number of fully connected layers and an output layer.

[0078] Result module 50 is used to use the seismic response displacement sequence as the seismic dynamic response result of the target building structure.

[0079] The seismic dynamic response calculation device for building structures based on time integral graph networks provided in this application employs the seismic dynamic response calculation method for building structures based on time integral graph networks in the above embodiments. It can solve the technical problem of how to achieve high-precision, real-time dynamic response calculation of structures while having low requirements for training data. Compared with the prior art, the beneficial effects of the seismic dynamic response calculation device for building structures based on time integral graph networks provided in this application are the same as those of the seismic dynamic response calculation method for building structures based on time integral graph networks provided in the above embodiments. Furthermore, other technical features in the seismic dynamic response calculation device for building structures based on time integral graph networks are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0080] In one embodiment, the processing module 40 is further configured to perform dimensionality reduction processing on the graph input data through the graph encoder to obtain a low-dimensional hidden graph; perform multi-step recursive calculation on the low-dimensional hidden graph through the time integration calculation module to obtain the updated hidden graph node features at each time step; and decode and restore the target building structure according to the updated hidden graph node features through the graph decoder to obtain the seismic response displacement sequence.

[0081] In one embodiment, the processing module 40 is further configured to: input graph node features from the graph input data into a residual network module; perform feature compression and dimensionality reduction on the graph node features through the residual blocks to obtain low-dimensional node features, wherein the dimension of the low-dimensional node features is a preset proportion of the dimension of the graph node features; input graph edge features from the graph input data into a convolutional residual network module to perform convolution operations and dimensionality reduction on the graph edge features to obtain low-dimensional edge features, wherein the dimension of the low-dimensional edge features is a preset proportion of the dimension of the graph edge features; perform sparsification processing on the adjacency matrix in the graph input data to obtain a processed adjacency matrix; and combine the low-dimensional node features, low-dimensional edge features, and the processed adjacency matrix to obtain a low-dimensional hidden graph.

[0082] In one embodiment, the processing module 40 is further configured to extract low-dimensional edge features and low-dimensional node features from the low-dimensional hidden graph; set an edge update function and a node update function, wherein the edge update function adopts a convolutional neural network, the convolutional neural network includes a second preset number of convolutional layers and a third preset number of activation layers, and the node update function adopts a fully connected neural network, the fully connected neural network includes a second preset number of fully connected layers and a first preset number of activation layers; input the low-dimensional edge features into the edge update function to optimize the low-dimensional edge features, thereby obtaining updated edge features; perform a first-step recursive update on the low-dimensional node features to obtain the hidden graph node features updated at the first time step; based on the updated edge features, aggregate the low-dimensional node features of adjacent graph nodes in the low-dimensional hidden graph through the node update function to obtain the graph nodes at the next time step; use the hidden graph node features updated at the first time step as the initial node features of the graph nodes at the next time step, and repeat the update and aggregation steps until the calculation of all time steps within the prediction period is completed, thereby obtaining the updated hidden graph node features at each time step.

[0083] In one embodiment, the processing module 40 is further configured to sequentially input the updated hidden graph node features of each time step into the fully connected layer of the graph decoder for dimension recovery and feature mapping in chronological order to obtain the physical domain features corresponding to each time step; extract displacement information from the physical domain features corresponding to each time step, the displacement information including the x-direction displacement, y-direction displacement and z-direction displacement of the structural degrees of freedom corresponding to the target building structure in the current time step; verify the validity of the displacement information of each time step to obtain the verification result; when the numerical range of the displacement information is within a preset displacement threshold, integrate the displacement information of all valid time steps in chronological order to obtain the seismic response displacement sequence of the target building structure.

[0084] In one embodiment, the construction module 20 is further configured to extract structural degrees of freedom from the node information of the finite element model, determine the physical coordinates and mechanical properties corresponding to each structural degree of freedom; map each structural degree of freedom to a graph node, and associate the graph node with the physical coordinates and mechanical properties of the corresponding structural degree of freedom; extract the structural element connection relationship from the element information of the finite element model, and determine the range of structural degrees of freedom connected by each structural element; set graph edges between the corresponding graph nodes according to the range of structural degrees of freedom connected by the structural elements, and associate the graph edges with the type and size of the corresponding structural element; generate an adjacency matrix based on the stiffness matrix of the finite element model, wherein the element value in the adjacency matrix is ​​a first preset value indicating that the corresponding graph node has a mechanical connection, and the element value is a second preset value indicating that there is no mechanical connection; and integrate the graph nodes, the graph edges, and the adjacency matrix to obtain initial graph data.

[0085] In one embodiment, the mapping module 30 is further configured to: extract a graph edge list from the initial graph data to determine the structural unit corresponding to each graph edge; map the sub-matrices in the mass matrix, the damping matrix, and the stiffness matrix corresponding to the structural unit to the first, second, and third edge feature components of each graph edge; combine the first, second, and third edge feature components to obtain graph edge features; extract a graph node list from the initial graph data to determine the structural degrees of freedom corresponding to each graph node; map the load time sequence corresponding to the structural degrees of freedom in the seismic motion time history data to the node feature components of each graph node; obtain the initial displacement data of the target building structure at the start of the prediction period; add the initial displacement data to the node feature components corresponding to the graph node to obtain graph node features; and integrate the graph edge features, the graph node features, and the adjacency matrix in the initial graph data to obtain graph input data.

[0086] This application provides a building structure seismic dynamic response calculation device based on a time integral graph network. The building structure seismic dynamic response calculation device based on a time integral graph network includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the building structure seismic dynamic response calculation method based on a time integral graph network in the above embodiment 1.

[0087] The following is for reference. Figure 7This document illustrates a structural schematic diagram of a building structure seismic dynamic response calculation device based on a time integral graph network, suitable for implementing embodiments of this application. The building structure seismic dynamic response calculation device based on a time integral graph network in this application embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 7 The illustrated seismic dynamic response calculation device for building structures based on time integral graph networks is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0088] like Figure 7 As shown, the seismic dynamic response calculation device for building structures based on time integral graph networks may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to programs stored in read-only memory (ROM) 1002 or programs loaded from storage device 1003 into random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the seismic dynamic response calculation device for building structures based on time integral graph networks. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the time integral graph network-based building structure seismic dynamic response calculation device to wirelessly or wiredly communicate with other devices to exchange data. Although various time integral graph network-based building structure seismic dynamic response calculation devices are shown in the figures, it should be understood that implementation or possession of all of them is not required. More or fewer may be implemented alternatively.

[0089] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0090] The seismic dynamic response calculation device for building structures based on time integral graph networks provided in this application employs the seismic dynamic response calculation method for building structures based on time integral graph networks described in the above embodiments. This addresses the technical problem of achieving high-precision, real-time dynamic response calculation of structures while minimizing the need for training data. Compared with existing technologies, the beneficial effects of the seismic dynamic response calculation device for building structures based on time integral graph networks provided in this application are the same as those of the seismic dynamic response calculation method for building structures based on time integral graph networks provided in the above embodiments. Furthermore, other technical features of this seismic dynamic response calculation device for building structures based on time integral graph networks are the same as those disclosed in the previous embodiment method, and will not be elaborated upon here.

[0091] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0092] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0093] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the method for calculating the seismic dynamic response of building structures based on time integral graph networks in the above embodiments.

[0094] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible storage medium containing or storing a program that can be executed by instructions, used by a device, or used in conjunction with it. The program code contained on the computer-readable storage medium may be transmitted using any suitable storage medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0095] The aforementioned computer-readable storage medium may be included in a time integral graph network-based seismic dynamic response calculation device for building structures; or it may exist independently and not be assembled into the time integral graph network-based seismic dynamic response calculation device for building structures.

[0096] The aforementioned computer-readable storage medium carries one or more programs that, when executed by a time-integral graph network-based building structure seismic dynamic response calculation device, enable the device to write computer program code for performing the operations of this application in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0097] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using dedicated hardware-based implementations that perform the specified functions or operations, or can be implemented using a combination of dedicated hardware and computer instructions.

[0098] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0099] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described method for calculating the seismic dynamic response of building structures based on time integral graph networks. This solves the technical problem of how to achieve high-precision, real-time dynamic response calculation of structures while having low requirements for training data. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the seismic dynamic response calculation method for building structures based on time integral graph networks provided in the above embodiments, and will not be repeated here.

[0100] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for calculating the seismic dynamic response of building structures based on time integral graph networks.

[0101] The computer program product provided in this application can solve the technical problem of how to achieve high-precision, real-time dynamic response calculation of structures while having low requirements for training data. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the seismic dynamic response calculation method for building structures based on time integral graph networks provided in the above embodiments, and will not be repeated here.

[0102] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for calculating seismic dynamic response of a building structure based on a time-integration graph network, characterized in that, include: Acquire the finite element model of the target building structure and the ground motion time history data during the prediction period, wherein the finite element model includes the node information, element information, mass matrix, damping matrix and stiffness matrix of the target structure; A graph structure is constructed based on the node and element information of the finite element model, and initial graph data is obtained; Based on the initial graph data, the mass matrix, damping matrix, stiffness matrix, and seismic motion time history data are mapped to obtain graph input data; The graph input data is input into a preset seismic response model for processing to obtain the seismic response displacement sequence of the target building structure. The preset seismic response model includes a graph encoder, a time integration calculation module, and a graph decoder. The graph encoder includes a residual network module and a convolutional residual network module. The residual network module includes a first preset number of residual blocks, and the convolutional residual network module includes a first preset number of convolutional residual blocks. The time integration module includes a parameter calculation unit, an edge update unit, and a node recursion unit. The graph decoder includes a first preset number of fully connected layers and an output layer. The seismic response displacement sequence is used as the seismic dynamic response result of the target building structure.

2. The method of claim 1, wherein, The step of inputting the graph input data into a preset seismic response model for processing to obtain the seismic response displacement sequence of the target building structure, wherein the preset seismic response model includes a graph encoder, a time integration calculation module, and a graph decoder, includes: The graph input data is reduced in dimensionality by the graph encoder to obtain a low-dimensional hidden graph. The time integration calculation module performs multi-step recursive calculations on the low-dimensional hidden graph to obtain the updated hidden graph node features at each time step. The target building structure's seismic response displacement sequence is obtained by decoding and restoring the target structure based on the updated hidden graph node features using the graph decoder.

3. The method as described in claim 2, characterized in that, The step of performing dimensionality reduction processing on the graph input data through the graph encoder to obtain a low-dimensional hidden graph includes: The graph node features in the graph input data are input into the residual network module. The residual block performs feature compression and dimensionality reduction on the graph node features to obtain low-dimensional node features. The dimension of the low-dimensional node features is a preset ratio of the dimension of the graph node features. The graph edge features in the graph input data are input into the convolutional residual network module to perform convolution operations and dimension reduction on the graph edge features to obtain low-dimensional edge features, wherein the dimension of the low-dimensional edge features is a preset ratio of the dimension of the graph edge features. The adjacency matrix in the graph input data is sparsified to obtain the processed adjacency matrix; The low-dimensional node features, low-dimensional edge features, and the processed adjacency matrix are combined to obtain a low-dimensional hidden graph.

4. The method as described in claim 2, characterized in that, The step of performing multi-step recursive calculations on the low-dimensional hidden graph through the time integration calculation module to obtain the updated hidden graph node features at each time step includes: Extract low-dimensional edge features and low-dimensional node features from the low-dimensional hidden graph; Set an edge update function and a node update function. The edge update function adopts a convolutional neural network, which includes a second preset number of convolutional layers and a third preset number of activation layers. The node update function adopts a fully connected neural network, which includes a second preset number of fully connected layers and a first preset number of activation layers. The low-dimensional edge features are input into the edge update function to optimize the low-dimensional edge features, resulting in updated edge features. The low-dimensional node features are recursively updated in the first step to obtain the hidden graph node features updated in the first time step. Based on the updated edge features, the low-dimensional node features of adjacent graph nodes in the low-dimensional hidden graph are aggregated through the node update function to obtain the graph nodes for the next time step. The updated hidden graph node features of the first time step are used as the initial node features of the graph nodes of the next time step. The update and aggregation steps are repeated until the calculation of all time steps in the prediction period is completed, and the updated hidden graph node features of each time step are obtained.

5. The method as described in claim 2, characterized in that, The step of obtaining the seismic response displacement sequence of the target building structure by decoding and reconstructing the target building structure based on the updated hidden graph node features using the graph decoder includes: The updated hidden graph node features at each time step are sequentially input into the fully connected layer of the graph decoder for dimension recovery and feature mapping to obtain the physical domain features corresponding to each time step. Displacement information is extracted from the physical domain features corresponding to each time step. The displacement information includes the x-direction displacement, y-direction displacement and z-direction displacement of the structural degrees of freedom corresponding to the target building structure at the current time step. The validity of the displacement information at each time step is verified, and the verification results are obtained. When the value of the displacement information is within the preset displacement threshold, the displacement information of all valid time steps is integrated in chronological order to obtain the seismic response displacement sequence of the target building structure.

6. The method as described in claim 1, characterized in that, The step of constructing a graph structure based on the node and element information of the finite element model and obtaining initial graph data includes: Structural degrees of freedom are extracted from the node information of the finite element model, and the physical coordinates and mechanical properties corresponding to each structural degree of freedom are determined. Each structural degree of freedom is mapped to a graph node, and the graph node is associated with the physical coordinates and mechanical properties of the corresponding structural degree of freedom; Extract the structural element connection relationships from the element information of the finite element model, and determine the range of structural degrees of freedom for each structural element connection; Based on the range of structural degrees of freedom connected by the structural units, graph edges are set between the corresponding graph nodes, and the graph edges are associated with the type and size of the corresponding structural units; An adjacency matrix is ​​generated based on the stiffness matrix of the finite element model, wherein the element value of the adjacency matrix is ​​a first preset value indicating that the corresponding graph node has a mechanical connection, and the element value is a second preset value indicating that there is no mechanical connection. By integrating the graph nodes, graph edges, and adjacency matrix, initial graph data is obtained.

7. The method as described in claim 1, characterized in that, The step of mapping the mass matrix, damping matrix, stiffness matrix, and seismic motion time history data based on the initial map data to obtain the map input data includes: Extract the graph edge list from the initial graph data and determine the structural unit corresponding to each graph edge; The sub-matrix in the mass matrix corresponding to the structural unit, the sub-matrix in the damping matrix corresponding to the structural unit, and the sub-matrix in the stiffness matrix corresponding to the structural unit are mapped to the first side feature component, the second side feature component, and the third side feature component of each edge of the graph. The graph edge features are obtained by combining the first edge feature components, the second edge feature components, and the third edge feature components. Extract a list of graph nodes from the initial graph data and determine the structural degrees of freedom corresponding to each graph node; The load time sequence corresponding to the structural degrees of freedom in the earthquake motion time history data is mapped to the nodal feature components of each graph node; Obtain the initial displacement data of the target building structure at the start of the prediction period; The initial displacement data is added to the node feature components corresponding to the graph node to obtain the graph node features; The graph edge features, the graph node features, and the adjacency matrix in the initial graph data are integrated to obtain the graph input data.

8. A device for calculating the seismic dynamic response of building structures based on a time integral graph network, characterized in that, The device includes: The acquisition module is used to acquire the finite element model of the target building structure and the ground motion time history data during the prediction period, wherein the finite element model includes the node information, element information, mass matrix, damping matrix and stiffness matrix of the target building structure. The construction module is used to construct the graph structure based on the node information and element information of the finite element model, and to obtain the initial graph data; The mapping module is used to map the mass matrix, damping matrix, stiffness matrix, and seismic motion time history data based on the initial map data to obtain map input data; The processing module is used to input the graph input data into a preset seismic response model for processing to obtain the seismic response displacement sequence of the target building structure. The preset seismic response model includes a graph encoder, a time integration calculation module, and a graph decoder. The graph encoder includes a residual network module and a convolutional residual network module. The residual network module includes a first preset number of residual blocks. The convolutional residual network module includes a first preset number of convolutional residual blocks. The time integration module includes a parameter calculation unit, an edge update unit, and a node recursion unit. The graph decoder includes a first preset number of fully connected layers and an output layer. The results module is used to use the seismic response displacement sequence as the seismic dynamic response result of the target building structure.

9. A device for calculating the seismic dynamic response of building structures based on time integral graph networks, characterized in that, The device includes: a memory, a processor, and a time integral graph network-based seismic dynamic response calculation program for building structures stored in the memory and running on the processor, wherein the time integral graph network-based seismic dynamic response calculation program for building structures is configured to implement the steps of the time integral graph network-based seismic dynamic response calculation method for building structures as described in any one of claims 1-7.

10. A storage medium, characterized in that, The storage medium stores a building structure seismic dynamic response calculation program based on a time integral graph network. When the time integral graph network-based building structure seismic dynamic response calculation program is executed by a processor, it implements the steps of the building structure seismic dynamic response calculation method based on a time integral graph network as described in any one of claims 1-7.

Images