Bridge time history prediction method, device, equipment and computer program product

Abstract

The invention belongs to the field of data processing, and provides a bridge time history prediction method, device and equipment and a computer program product. According to the embodiment of the invention, dynamic characteristic data such as wind speed, wave height and seismic oscillation of the target bridge and static characteristic data such as material strength and damping ratio of the target bridge are obtained and input into the pre-trained bridge response prediction model for processing, and the time-history response data of the target bridge is obtained and output, so that the method can adapt to bridge structure parameter changes, and the accuracy of bridge structure parameters is improved. The dynamic response of the bridge under the multi-source load coupling effect in a complex environment is predicted, and the time history prediction precision of the bridge is improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of data processing, and particularly relates to a bridge time history prediction method, device, equipment and computer program product. BACKGROUND

[0002] The bridge is a key component of important traffic and energy infrastructure, and is used to realize cross-domain traffic and energy transportation, and is an important traffic hub in the field of regional economic development.

[0003] In the use process of the bridge, it is usually disturbed by external environmental factors (such as wind, wave, flow and the like), which affects the structural safety and service life of the bridge to some extent.

[0004] Therefore, how to determine the dynamic response of the bridge under complex environment is an urgent problem to be solved. SUMMARY

[0005] Therefore, the embodiments of the present application provide a bridge time history prediction method, device, equipment and computer program product, which can predict the dynamic response of the bridge under complex environment, and improve the bridge time history prediction accuracy.

[0006] The first aspect of the embodiments of the present application provides a bridge time history prediction method, comprising: obtaining dynamic characteristic data and static characteristic data of a target bridge; wherein the dynamic characteristic data comprises at least one environmental parameter of wind speed, wave height, and seismic motion; and the static characteristic data comprises at least one structural parameter of material strength and damping ratio of the target bridge; inputting the dynamic characteristic data and the static characteristic data into a pre-trained bridge response prediction model for processing, to obtain time history response data of the target bridge and output.

[0007] The second aspect of the embodiments of the present application provides a bridge time history prediction device, comprising: an acquisition module configured to obtain dynamic characteristic data and static characteristic data of a target bridge; wherein the dynamic characteristic data comprises at least one environmental parameter of wind speed, wave height, and seismic motion; and the static characteristic data comprises at least one structural parameter of material strength and damping ratio of the target bridge; a prediction module configured to input the dynamic characteristic data and the static characteristic data into a pre-trained bridge response prediction model for processing, to obtain time history response data of the target bridge and output.

[0008] The third aspect of the embodiments of the present application provides a bridge time history prediction device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the method provided in the first aspect when executing the computer program.

[0009] The fourth aspect of the embodiments of the present application provides a computer program product, characterized in that when the computer program product is executed on a computer, the steps of the method provided in the first aspect are implemented.

[0010] The first aspect of the embodiments of the present application provides a bridge time history prediction device method, by acquiring dynamic characteristic data such as wind speed, wave height, and earthquake motion of a target bridge, and static characteristic data such as material strength and damping ratio of the target bridge, and inputting them into a pre-trained bridge response prediction model for processing, obtaining time history response data of the target bridge and outputting, which can adapt to changes in bridge structure parameters, predict the dynamic response of the bridge under complex environment (such as under the coupling action of wind, wave, earthquake and other multi-source loads), improve the bridge time history prediction accuracy, and provide data support for the safety evaluation and health monitoring of the bridge structure.

[0011] It can be understood that the beneficial effects of the above-mentioned second aspect to the fourth aspect can be referred to the related description in the first aspect, which will not be repeated here. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0013] Figure 1 is an application scenario diagram of a sea bridge provided by the embodiments of the present application; Figure 2 is one of the flowcharts of the bridge time history prediction method provided by the embodiments of the present application; Figure 3 is a data processing flowchart of the bridge time history prediction model provided by the embodiments of the present application; Figure 4 is a structural flowchart of the time sequence information encoding module provided by the embodiments of the present application; Figure 5 is the second flowchart of the bridge time history prediction method provided by the embodiments of the present application; Figure 6 is a structural flowchart of the bridge time history prediction device provided by the embodiments of the present application; Figure 7FIG. 1 is a structural schematic diagram of a bridge time prediction device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0014] In the following description, for purposes of explanation and not limitation, specific details are set forth, such as particular architectures, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known methods, devices, circuits, and

[0015] It is to be understood that the terminology "includes", "has", "holds", "contains" or "comprises", "comprising", or "including" when used in this specification and in the following claims, specifies the presence of stated features, integers, steps, operations, elements, or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or groups thereof.

[0016] It is also to be understood that the terminology "and / or" when used in this specification and in the following claims, refers to at least one of the items, or any combination of the items, or all of the items listed together.

[0017] As used in this specification and in the claims, the term "if" can be construed to mean "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [a described condition or event] is detected" can be construed to mean "upon determining" or "in response to determining" or "upon [the described condition or event] being detected" or "in response to [the described condition or event] being detected," depending on the context.

[0018] In addition, the terms "first", "second", "third", etc. in the description of the present application and the appended claims are only used to distinguish descriptions, and cannot be understood as indicating or implying relative importance.

[0019] Reference in the specification to "one embodiment" or "some embodiments" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application. The appearances of the phrase "in one embodiment" or "in some embodiments" in various places in the specification are not necessarily all referring to the same embodiment, although it can. The terms "including", "containing", "having" and variations thereof are meant to encompass the terms "including but not limited to", unless otherwise indicated. FIG. 1 is a structural schematic diagram of a bridge time prediction device provided by an embodiment of the present application.

[0020] Terminology Explanation: Long Short-Term Memory (LSTM) neural network: refers to a type of recurrent neural network, which is commonly used for time series modeling and can effectively capture long-term dependency features in time series.

[0021] Frequency-Attention Long Short-Term Memory (FA-LSTM) is an improved model that introduces a learnable frequency domain filter on the basis of the Long Short-Term Memory Neural Network (LSTM).

[0022] Gated Recurrent Unit (GRU): refers to a lightweight recurrent neural network.

[0023] Mean Squared Error (MSE): This refers to a type of data that can be used to measure the difference between predicted and actual values.

[0024] Coefficient of Determination (R²): This is a parameter used to measure the goodness of fit between predicted and actual results.

[0025] Correlation Index (CI): This is a metric used to measure the correlation between predicted and actual values.

[0026] Peak relative error (Epr): refers to a parameter used to evaluate the error between the peak values ​​of the predicted response and the actual response.

[0027] Fast Fourier Transform (FFT): This is a signal processing method that converts time-domain signals to the frequency domain.

[0028] The Inverse Fourier Transform (IFFT) is a signal processing method that converts frequency domain signals back to the time domain.

[0029] Fully Connected Layer (FC): This refers to a basic network layer of a neural network, which consists of multiple neurons operating in parallel.

[0030] Layer Normalization (LayerNorm) is a method for normalizing the inputs of neural network layers to speed up training and stabilize model performance.

[0031] Bridges are a key component of important transportation and energy infrastructure, used to facilitate cross-regional transportation and energy transport, and are important transportation hubs in the field of regional economic development.

[0032] During the use of bridges, they are often affected by external environmental factors (such as wind, waves, currents, etc.), which to some extent affect the structural safety and service life of bridges, making bridges face more severe challenges.

[0033] Therefore, determining the dynamic response of bridges in complex environments is an urgent problem to be solved. As is understood, a bridge is an architectural structure that stands in water (such as rivers, lakes, and seas) to facilitate cross-regional transportation and energy transport. It includes, but is not limited to, inland bridges and sea bridges. The following text mainly uses sea bridges as an example to illustrate the bridge time history prediction method provided in this application.

[0034] Figure 1 This is an application scenario diagram of a sea bridge.

[0035] See Figure 1 When a sea bridge stands in the sea, part of its structure is above the sea level, part is in the seawater, and part is connected to the seabed. This makes the structural safety of the sea bridge susceptible to external environmental factors such as wind, waves, and seismic waves.

[0036] To address the aforementioned issues, embodiments of this application provide a bridge time history prediction method, apparatus, device, and computer program product. This method can acquire at least one environmental parameter of the target bridge as dynamic feature data and at least one structural parameter as static feature data. The dynamic and static feature data are input into a pre-trained bridge response prediction model for processing, resulting in the time history response data of the target bridge, which is then output. This method can adapt to changes in bridge structural parameters and predict the dynamic response of the bridge under complex environments (such as the coupling of multiple loads from wind, waves, and earthquakes), improving the accuracy of bridge time history prediction and thus providing data support for the safety assessment and health monitoring of bridge structures.

[0037] Figure 2 This is a flowchart illustrating a bridge time history prediction method provided in an embodiment of this application.

[0038] like Figure 2 As shown, the bridge time history prediction method provided in this application includes the following steps S101 to S102: Step S101: Obtain dynamic and static characteristic data of the target bridge; wherein, the dynamic characteristic data includes at least one environmental parameter among wind speed, wave height, and ground motion; the static characteristic data includes at least one structural parameter among the target bridge's material strength and damping ratio, proceed to step S102.

[0039] In this application, dynamic characteristic data and static characteristic data of the target bridge are collected through sensing devices. Static characteristic data includes, but is not limited to, at least one time-independent structural parameter of the target bridge, such as material strength, damping ratio, and elastic modulus of steel reinforcement, used to determine the structural characteristics of the target bridge. Dynamic characteristic data includes, but is not limited to, at least one time-varying external environmental parameter, such as wind speed, wave height, and seismic motion, used to determine the influence of external environmental factors on the target bridge.

[0040] Correspondingly, the aforementioned sensing devices include, but are not limited to, at least one of the following: wind sensors, buoys, radar, pressure sensors, and satellite remote sensors.

[0041] Optionally, the bridge time history prediction method provided in this embodiment can be based on a deep learning framework (such as PyTorch or TensorFlow) and implemented by an electronic device equipped with a graphics processing unit (GPU). The aforementioned electronic device can be a workstation or a server or other terminal device, and this embodiment does not make any specific limitations.

[0042] Step S102: Input the dynamic feature data and the static feature data into the pre-trained bridge response prediction model for processing, and output the time history response data of the target bridge.

[0043] In the application, the dynamic and static feature data of the target bridge are input into a pre-trained bridge response prediction model for processing, thereby obtaining the time-history response data of the bridge structure determined based on the static feature data in a complex environment containing dynamic feature data, which is then output by the pre-trained bridge response prediction model. The time-history response data includes, but is not limited to, predicted time-history response values, peak time-history response values, and the frequency domain time-history response.

[0044] In applications, a pre-trained bridge response prediction model refers to a neural network model that has been pre-trained and is able to predict the future structural response of a target bridge in a complex environment (also known as time history response data, which includes dynamic responses such as displacement, acceleration, stress, or strain) based on the input dynamic and static feature data of the target bridge.

[0045] It is understandable that static characteristic data mainly includes structural parameters that play a key role in the structural response. These can be determined using parameter sensitivity analysis methods (such as single-factor perturbation analysis): among all candidate structural parameters, other structural parameters are kept constant, and only one structural parameter is perturbed and the change in response is quantified. If the change exceeds a corresponding preset threshold, then that structural parameter is selected as a sensitive structural parameter. The quantification index can be expressed as: (1); Where X and Y represent the displacement time history of the bridge, and This represents the average displacement of the bridge. and This represents the standard deviation of bridge displacement.

[0046] Understandably, a smaller ETHA indicates a stronger correlation between the two time histories. By setting a sensitivity threshold (e.g., setting the threshold to 0.05% or 0.06%), multiple sensitive structural parameters that play a key role in the structural response can be selected.

[0047] For example, 42 sensitive structural parameters, including concrete material strength, steel bar elastic modulus, and damping ratio, can be selected as static feature data for bridge time history prediction models through parameter sensitivity analysis.

[0048] Understandably, commonly used bridge structural response analysis methods primarily rely on finite element modeling and time history analysis. By establishing a high-precision finite element model and then performing nonlinear dynamic time history calculations under input loads, the displacement, stress, and other responses of the bridge structure are obtained. This method is widely used in engineering design and scientific research analysis. However, it suffers from low computational efficiency, inability to achieve real-time or near-real-time prediction, high model complexity, and time-consuming and laborious modeling and parameter calibration processes. While response analysis methods using deep learning surrogate models to replace finite element models can reduce computational costs and achieve advancements in areas such as seismic response prediction and vehicle load identification, they can only identify single load types (such as earthquakes) and have poor generalization ability, limiting response prediction to specific bridge structures.

[0049] The pre-trained bridge response prediction model provided in this application embodiment can fuse various input external environmental loads and variable bridge structural parameters, learn the relationship between various external environmental loads and bridge response, and is applicable to various bridge structures, thereby quickly completing the response prediction of the target bridge and improving the accuracy of bridge time history prediction.

[0050] like Figure 3As shown, in one embodiment, the pre-trained bridge response prediction model includes a temporal information encoding module, a static feature encoding module, a multi-branch attention fusion module, and a multi-task decoding prediction module.

[0051] In application, the pre-trained bridge response prediction model includes a temporal information encoding module, a static feature encoding module, a multi-branch attention fusion module, and a multi-task decoding prediction module. The temporal information encoding module and the static feature encoding module are connected to the multi-branch attention fusion module, and the multi-branch attention fusion module is connected to the multi-task decoding prediction module.

[0052] The system comprises several modules: a time-series information encoding module for deep feature extraction of dynamic feature data (also known as dynamic load time-series signals), and a static feature encoding module for determining the static feature information of static feature data. A multi-branch attention fusion module implements a multi-branch attention fusion method, dynamically assigning importance weights to different input branches at different times through an attention mechanism. A multi-task decoding and prediction module implements time-series prediction functions, including a main task (e.g., time-series response prediction) and auxiliary tasks (e.g., time-series response peak prediction, frequency domain feature prediction).

[0053] Correspondingly, step S102 includes the following steps: Step S1021: Input the dynamic feature data into the time-series information encoding module for processing to obtain dynamic feature information; Step S1022: Input the static feature data into the static feature encoding module for processing to obtain static feature information; Step S1023: Input the dynamic feature information and the static feature information into the multi-branch attention fusion module for processing to obtain fused feature information; Step S1024: Input the fused feature information into the multi-task decoding prediction module for processing to obtain the predicted time history response, peak time history response, and frequency domain time history response of the target bridge.

[0054] See Figure 3In application, when the dynamic and static feature data of the target bridge are input into the pre-trained bridge time history prediction model, the dynamic feature data is processed by the time-series information encoding module to obtain the dynamic feature information corresponding to the dynamic feature data. Similarly, the static feature data is processed by the static feature encoding module to obtain the static feature information. Thus, the dynamic feature information is input from the time-series information encoding module, and the static feature information is input from the static feature encoding module, to the multi-branch attention fusion module. The multi-branch attention fusion module performs fusion processing to obtain the fused feature information of the target bridge, which is then input into the multi-task decoding and prediction module for processing to obtain the predicted time history response value, peak time history response value, and frequency domain time history response of the target bridge.

[0055] In one embodiment, step S1021 includes the following steps: Each of the environmental parameters is input into the timing information encoding module for processing to obtain the frequency domain feature information corresponding to each of the environmental parameters. Determine the frequency components corresponding to each of the frequency domain feature information, and convert each of the frequency components into corresponding time domain feature information; Extract the dependent feature sequences of each of the time-domain feature information to obtain the dynamic feature information.

[0056] In the application, the dynamic feature data includes at least one environmental parameter. Each environmental parameter is input into the time-series information encoding module, and the environmental parameters are processed by time-frequency conversion to obtain the frequency domain feature information corresponding to each environmental parameter. The frequency domain components corresponding to each frequency domain feature information are determined, and the frequency components corresponding to each frequency domain feature information are converted into the corresponding time domain feature information. The dependent feature sequence of each time domain feature information is extracted, thereby obtaining the dynamic feature information corresponding to each environmental parameter.

[0057] Understandably, the input data for the time series information encoding module is a one-dimensional time series signal (e.g., wind speed time history, wave height time history, seismic wave acceleration time history, etc.). The shape of each time series signal is [batch size, time step, 1]. After processing by the time series information encoding module, it can output dynamic feature information with the shape [batch size, time step, hidden layer dimension]. Among them, parameters such as batch size, time step, and hidden layer dimension can be specifically set according to actual needs. For example, the batch size can be set to 4, the time step to 6000, and the hidden layer dimension to 60.

[0058] As an example, and not a limitation, the timing information encoding module can be implemented by cascading multiple frequency-enhanced Long Short-Term Memory Neural Network (FA-LSTM) modules. The number of frequency-enhanced LSTM modules can be determined based on the number of input environmental parameters.

[0059] For example, when inputting environmental parameters such as wind speed, wave height, and seismic waves, the timing information encoding module may include three frequency-domain enhanced LSTM modules.

[0060] Optionally, the kernel modeled by the temporal information encoding module is a long short-term memory neural network (LSTM), which can also be replaced by a one-dimensional convolutional neural network (1D-CNN), a gated recurrent unit (GRU), a deep learning model based on self-attention mechanism such as Transformer, or a temporal convolutional network (TCN) to achieve feature extraction of temporal signals.

[0061] Figure 4 This is a schematic diagram of the structure of a timing information encoding module provided in an embodiment of this application.

[0062] See Figure 4 The time-series information encoding module includes: a fully connected layer, a first time-frequency transformation unit, a frequency domain filter, a second time-frequency transformation unit, a Long Short-Term Memory (LSTM) neural network layer, and layer normalization. Taking wind speed as an example, the wind speed is input to the fully connected layer for processing. The first time-frequency transformation unit converts the time-series signal of the wind speed to the frequency domain, obtaining the corresponding frequency domain feature information (also called frequency domain components). The frequency domain filter determines the trainable parameter vectors that correspond one-to-one with the frequency domain components. Through training, the weighting weights of different frequency bands are automatically adjusted to determine the frequency components related to the structural response. Then, the second time-frequency transformation unit converts the frequency components into corresponding time-domain feature information. Time series modeling is performed through the LSTM neural network layer to extract long-term dependent features. Layer normalization is used to normalize the long-term dependent features to obtain the dynamic feature information of wind speed. In this way, the time-series information encoding module can capture the long-term dependence and key frequency features of the structural response, improving the prediction accuracy of strong nonlinear vibrations.

[0063] Optionally, the first time-frequency transformation unit can use the fast Fourier transform algorithm to realize time-frequency transformation, and the frequency domain filter can be a learnable frequency domain filter. The second time-frequency transformation unit can use the inverse Fourier transform algorithm to realize time-frequency transformation.

[0064] In applications, frequency domain enhancement can be achieved by combining Fast Fourier Transform (FFT) with learnable filters. Alternatively, frequency domain or time-frequency domain features can be extracted from time-series signals using methods such as wavelet transform, empirical mode decomposition (EMD), and variational mode decomposition (VMD).

[0065] In applications, the static feature encoding module can include, but is not limited to, a fully connected network. The input data for the static feature encoding module consists of multiple two-dimensional structural parameters (also called static feature vectors) with a shape of [batch size, number of parameters]. The static feature encoding module can map the static feature vectors to the same hidden layer dimension as the dynamic feature data and replicate and expand them in the time dimension, thereby converting the two-dimensional static feature vectors into three-dimensional vectors. The output is a tensor with a shape of [batch size, number of time steps, hidden layer dimension], achieving alignment with the dynamic features in the spatiotemporal dimension. This facilitates the concatenation and fusion of static and dynamic feature data at each time step, allowing the static feature data of the bridge structure to participate in the response prediction at each time step, improving prediction accuracy.

[0066] In the static feature encoding module, the batch size, time step number, and hidden layer dimension are consistent with those in the temporal information encoding module, and the number of parameters refers to the number of structural parameters of the target bridge.

[0067] In the application, the input data of the multi-branch attention fusion module includes dynamic feature information and static feature information, both of which have the shape of [batch size, time step, hidden layer dimension]. The importance weights of different input branches at different times are assigned by a dynamic weighting method based on the attention mechanism.

[0068] Alternatively, the fusion of features from different input branches can be achieved through simple concatenation, weighted averaging, or other methods.

[0069] In the application, the input data of the multi-branch attention fusion module includes multiple dynamic feature information branches (such as wind speed, wave height, and ground motion) and one static feature information. The multi-branch attention fusion module stacks the features of all branches to form a multi-dimensional tensor, resulting in data with the shape [batch size, time steps, 4 × hidden layer dimension]. Through the attention mechanism, the correlation weights of each branch at different times are calculated, generating a weight tensor with the shape [batch size, time steps, 4 × hidden layer dimension]. The features of each branch are weighted and summed according to the weights to obtain a unified fused feature information with the shape [batch size, time steps, hidden layer dimension].

[0070] Understandably, the aforementioned attention fusion mechanism can adaptively allocate the importance of each load branch based on the differences in input data. For example, in scenarios where seismic waves have a significant impact, the multi-branch attention fusion module automatically increases the weight of the seismic wave feature information branch; in scenarios where wind and wave coupling is significant, the multi-branch attention fusion module increases the contributions of the wind speed load feature information and wave load feature information branches, which can improve the prediction accuracy of bridge time history and vehicle models while enhancing the physical interpretability of the prediction results.

[0071] In the application, the input data of the multi-task decoding and prediction module is fused feature information with shape [batch size, time steps, hidden layer dimension], and the multi-task decoding and prediction module can adopt a multi-task learning strategy to add two auxiliary tasks, maximum response amplitude prediction and frequency domain feature prediction, while realizing the main task of time history prediction.

[0072] In the application, the multi-task decoding and prediction module includes, but is not limited to, a fully connected layer and three prediction heads. The multi-task decoding and prediction module can decode the input and output data through the shared fully connected layer, realize preliminary feature transformation and dimensionality reduction operations, output the corresponding shared feature information, thereby learning the general features that contribute to the three prediction tasks, and copy and distribute the shared feature information to the three independent prediction heads for prediction. The prediction heads include a time-history response prediction head, a time-history response peak prediction head, and a time-history response frequency domain prediction head.

[0073] The temporal response prediction head includes, but is not limited to, a fully connected layer. It can map shared feature information into a tensor of [batch size, time step, 1], and output the complete structural response temporal history.

[0074] In application, the time history response peak prediction head includes, but is not limited to, a global max pooling layer and a fully connected layer. The time history response peak prediction head can use a global max pooling layer to compress the shared feature information in the time dimension, extract the maximum value of the shared feature information over the entire time history, and output a scalar value through a fully connected layer to obtain the maximum response amplitude of the entire response time history. The output data shape is [batch size, prediction amplitude], which enables the bridge time history prediction model to accurately capture the extreme values ​​of the time history response, so as to assess the safety of the bridge structure based on the extreme values ​​of the time history response.

[0075] In applications, the time-history response frequency domain prediction head includes, but is not limited to, fully connected layers. This head performs a Fast Fourier Transform on shared feature information in the time dimension, converting it into corresponding frequency domain features. A fully connected layer then learns the mapping from these frequency domain features to the target spectrum, outputting a frequency domain representation of the response time-history signal. The output data has the shape [batch size, number of frequency points]. This enables bridge time-history prediction models to understand the characteristics of the time-history response from a frequency domain perspective, enhancing their ability to model nonlinear dynamic behavior.

[0076] Understandably, the encoder-decoder framework used in the bridge time history prediction model, combined with mechanisms such as frequency domain enhancement, attention fusion, and multi-task learning, can achieve rapid and accurate prediction of the dynamic response of bridges under multiple loads.

[0077] like Figure 5As shown, in one embodiment, the following steps are included before step S101: S201. Obtain the static simulation information of the simulated bridge, and construct the finite element simulation model of the simulated bridge based on the static simulation information; S202. Based on the static simulation information, simulate at least one set of dynamic simulation information.

[0078] In this application, the simulated static information includes at least one simulated bridge structural parameter; the simulated bridge refers to a virtual bridge structure. The simulated static information of the bridge is obtained, and a finite element simulation model of the simulated bridge is constructed based on this information.

[0079] As an example, and not a limitation, the structural response of offshore bridges can be analyzed using finite element numerical simulation. This involves establishing a three-dimensional finite element model of the offshore bridge, modeling the main girder, piers, foundation, and soil-structure interactions using solid or beam elements, and employing nonlinear material constitutive models to simulate the nonlinear behavior of the structure. Based on input external load data, nonlinear dynamic time history analysis is performed using finite element software (e.g., OpenSees, ANSYS, ABAQUS) through stepwise integration methods (e.g., the Newmark-b method), outputting displacement, stress, and / or acceleration responses for safety assessment and health monitoring of the offshore bridge.

[0080] As an example rather than a limitation, since the piers of a bridge structure are prone to nonlinear dynamic time histories under external loads, while other locations generally maintain linear dynamic time histories, in order to improve computational efficiency and accuracy, the main beams and piles can be modeled as linear features, while the piers can be modeled using fiber beam models to reflect the nonlinear characteristics of the piers.

[0081] For example, soil spring elements are introduced at the junction of the pile foundation and the soil to simulate the influence of different soil types (such as soft soil and hard soil) on the dynamic characteristics of the structure.

[0082] As an example rather than a limitation, the main environmental loads on bridges include wind loads, wave loads, and the effects of seismic motion and the resulting hydraulic shocks during extreme disasters.

[0083] In applications, at least one set of simulated dynamic information is generated based on the simulated static information. The simulated dynamic information refers to the parameters of the external environmental influences on the simulated bridge, including but not limited to simulated wind speed load, simulated wave height load, and simulated seismic load.

[0084] S203. Input the simulation static information and each of the simulation dynamic information into the finite element simulation model for simulation processing to obtain a training sample set; wherein, the training sample set includes multiple sample pairs, and each sample pair includes simulation static information, simulation dynamic information and simulation response information; S204. The bridge response prediction model is pre-trained using the training sample set to obtain the pre-trained bridge response prediction model, and then proceed to step S101.

[0085] In the application, the static and dynamic information input values ​​of the simulation are processed in the finite element simulation model to obtain a training sample set consisting of multiple sample pairs.

[0086] Understandably, by performing nonlinear dynamic time history analysis using a finite element simulation model, the dynamic response of a structure under coupled loads can be calculated. By systematically combining different load cases and structural parameter combinations, and performing finite element simulations in batches, multiple sets of sample pairs {simulation dynamic information (also known as input load), simulation static information (also known as structural parameters), and simulation response information} can be generated, thereby constructing a training sample set covering a variety of potential load cases.

[0087] In the application, each sample pair includes simulation static information, simulation dynamic information, and simulation response information corresponding to the aforementioned simulation static and dynamic information. The bridge response prediction model is pre-trained using the training sample set to obtain the pre-trained bridge response prediction model.

[0088] Alternatively, the training sample set can also be actual test data.

[0089] In one embodiment, the simulated dynamic information includes at least one of simulated wind speed load, simulated wave height load, and simulated seismic load. Step S202 includes: At least one seismic wave acceleration is selected from a preset earthquake database as the simulated seismic load.

[0090] In applications, multiple standard seismic wave accelerations can be selected from a preset seismic database (e.g., the general PEER strong earthquake record database) based on the design response spectrum, and used as simulated seismic loads.

[0091] Understandably, under the action of seismic motion, the dynamic interaction between the bridge and the surrounding water will generate a certain degree of hydrodynamic pressure, which will increase the inertial force of the structure and have an adverse effect on the stability of the bridge. The hydrodynamic pressure can be simplified to the additional mass attached to the underwater piers according to the method of added mass.

[0092] Taking a rectangular bridge pier with a length of 2ax and a width of 2ay as an example, the additional mass per unit length of the pier can be expressed as: (2); In the formula, k represents the additional mass coefficient of the rectangular cross-section of the bridge. This indicates the density of seawater.

[0093] Based on the preset wind spectrum and the simulated static information, a random wind speed is generated, and the random wind speed is converted into a simulated wind speed load.

[0094] In application, a preset wind spectrum can be selected in advance. The wind field can be characterized using the aforementioned wind spectrum and simulated static information, and can be expressed as follows: (3); In the formula, f represents the dimensionless frequency. Represents dimensionless frequency. This represents the dimensionless spectral density.

[0095] For example, a random wind speed time history generated by spectral representation and then transformed into a wind load time history can be expressed as: (4); In the formula, D represents the height of the main girder of the bridge. Indicates air density, U(z) represents the drag coefficient of the component, U(z) represents the average wind speed at height z, and u(x, z, t) represents the fluctuating wind speed time history. These are used to obtain the random wind speed time history and wind load time history data.

[0096] Based on the preset wave spectrum and the simulated static information, a random wave height is generated, and the random wave height is converted into a simulated wave height load.

[0097] In applications, a preset wave spectrum (e.g., the JONSWAP spectrum) is selected in advance. Based on the preset wave spectrum and the simulated static information, random waves are simulated, which can be expressed as: (5); In the formula, This represents the empirical correction factor. Indicates a high degree of meaning. Indicates frequency, Indicates the frequency of the spectral peak. Indicates the peak enhancement factor. Indicates the spectral width: .

[0098] Understandably, the three parameters of average wind speed, significant wave height, and wave period are used to determine wind time history and wave time history, and real monitoring data can be obtained from publicly available datasets of the target sea area.

[0099] Optionally, based on the selected wave spectrum, a random wave surface elevation time history H is generated using linear wave theory and harmonic superposition method.

[0100] Understandably, due to the size of the bridge piers Compared to wavelength Smaller, when meeting the pier size requirements / When the wave surface time history is less than 0.2, the Morrison equation can be used to determine the wave surface time history as the time history of the random wave forces acting on the underwater structure, which can be expressed as: (6); In the formula, Indicates the diameter of the bridge piles. and These represent the drag force coefficient and the inertia force coefficient, respectively, thus simulating and obtaining wave surface time history and wave load time history data.

[0101] in, and The values ​​can be referenced from the data in the specification document.

[0102] As an example, and not a limitation, to achieve alignment and compatibility of multi-source inputs in the time dimension, these dynamic time-history signals (wind speed, wave height, ground motion) need to be processed into sequence data of the same form. For example, sequence data with a duration of 60 seconds and a sampling frequency of 100 Hz. Signals with insufficient duration are padded with zeros, and those exceeding the duration are truncated, thus unifying the time reference of the dynamic time-history signal inputs and facilitating cross-modal correlation in the model.

[0103] Alternatively, the Latin hypercube sampling method can be used to randomly sample within the numerical variation range of each static feature data in order to efficiently generate parameter combinations with good representativeness and uniform distribution in the parameter space.

[0104] Understandably, by applying the simulated wind load and wave force time histories as nodal or distributed forces to the corresponding locations on the structure, using the seismic motion time histories as base acceleration input, and applying the seismic hydraulic forces as additional mass to the underwater pier nodes, it is possible to simulate the effects of multiple load sources. Furthermore, by substituting the various structural parameters from the static feature data into the finite element model, the dataset can cover multiple structural forms.

[0105] In one embodiment, step S204 includes the following steps: S2041. Normalize each of the sample pairs to obtain the training dataset; S2042. Input the training dataset into the bridge response prediction model, and pre-train the bridge response prediction model based on the total loss function to obtain the pre-trained bridge response prediction model.

[0106] In applications, dynamic feature data (also known as dynamic loads, such as wind speed, waves, and earthquakes) are independent time-series signals. To eliminate dimensional differences, all input signals and output data of the model can be normalized. This avoids gradient explosion or vanishing gradient problems during model training.

[0107] For example, the input simulation dynamic information is normalized to [-1, 1]. The input simulation static information is normalized to the [0, 1] interval within the range of each parameter. The input simulation response information [output response] is normalized to the [-1, 1] interval.

[0108] In one embodiment, the following steps are included before step S2042: Determine the weights corresponding to the time-domain loss function, peak loss function, and frequency-domain loss function respectively; The total loss function is determined based on the time-domain loss function, the peak loss function, the frequency-domain loss function, and each of the weights.

[0109] In applications, the multi-task output results based on the bridge time history prediction model consist of three parts: time domain loss, peak loss, and frequency domain loss.

[0110] Among them, time domain loss Used to measure the difference between the predicted time history and the actual time history; peak loss Used to constrain the extreme values ​​of the prediction results, ensuring that the predicted maximum response amplitude is consistent with the true value; frequency domain loss. It is used to measure the difference between the predicted signal and the true signal in the frequency domain. The formula for calculating the loss function can be expressed as: (7); (8); (9); in, and These represent the predicted and actual values ​​of the structural time history response, respectively. and Let k and n represent the frequency domain representations of the predicted and actual values, respectively, where k represents the number of frequency components and n represents the number of time steps.

[0111] Understandably, different loss terms have varying importance during training. Therefore, an uncertainty-weighted approach can be used to introduce learnable parameters for each subtask. And dynamically adjust during training to achieve an adaptive balance among the three prediction tasks, so that the total loss function is expressed as: (10); Alternatively, the model can be trained using the standard gradient descent algorithm, guided by the aforementioned total loss function.

[0112] In some embodiments, the difference between the model's output value and the simulated response information in each sample pair during training can be determined by a loss function. The model pre-training is considered complete when the difference for each sample pair is less than a preset threshold.

[0113] Understandably, the model's optimizer (Adam) can calculate the gradient based on the loss function and backpropagate it to all modules, such as the frequency domain attention LSTM and multi-branch attention fusion, to update the model parameters. This enables the model to learn to accurately predict structural responses from complex multi-source input data. Through multi-task joint training, the bridge time history prediction model can be optimized simultaneously in the time domain, extrema, and frequency domain, thereby achieving higher overall prediction accuracy and robustness.

[0114] As an example, and not a limitation, the bridge time history prediction model can replace or add other auxiliary tasks beyond time history prediction, peak prediction, and frequency domain prediction. For example, velocity response prediction, modal parameter prediction, and fatigue damage index prediction can be used to enhance the model's generalization ability.

[0115] Optionally, the aforementioned uncertainty-weighted multi-task loss function can be replaced with a fixed-weighted loss function to achieve multi-task balance.

[0116] Based on the experiment, a dataset of 2000 samples was divided into a training set, validation set, and test set in a 7:2:1 ratio. The bridge time history prediction model was tested 1500 times with an initial learning rate of 0.001. The test results are as follows: Under wind and wave load conditions, the prediction accuracy was R² = 0.9304; under extreme disaster (wind-wave-earthquake coupling) conditions, the prediction accuracy was R² = 0.9212. Furthermore, the bridge time history prediction model achieved high accuracy in several metrics, including mean squared error (MSE), mean absolute error (MAE), correlation index (CI), and peak prediction relative error (Epr), indicating high prediction accuracy and robustness. Model performance evaluation data are shown in Table 1.

[0117] Table 1 Model Performance Evaluation Data The bridge time-history prediction model provided in this application supports both multi-source dynamic loads (wind, waves, seismic waves, etc.) and static structural parameters, adapting to the coupling effect of complex multi-source loads on the structure and different structural parameters. By employing a frequency-domain enhanced long short-term memory neural network (FA-LSTM) combined with a learnable frequency-domain filter, the model enhances its ability to capture key frequency features. Through a multi-branch attention fusion mechanism, it achieves adaptive weighted fusion of different load branches and structural parameter features. The introduction of multi-task learning and uncertainty-weighted loss design improves the accuracy and robustness under multiple tasks such as time-history prediction, peak prediction, and frequency-domain prediction. Furthermore, the model supports dynamic expansion of input branches, flexibly adapting to different monitoring needs and operating conditions, and can combine the frequency-domain characteristics of time-domain signals to improve the model's prediction efficiency and accuracy. Moreover, when the bridge structure degrades or is damaged during long-term operation, time-history prediction can still be performed without retraining the model or adjusting the input structural parameters, thus improving the model's applicability.

[0118] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0119] This application also provides a bridge time history prediction device for executing the steps described in the bridge time history prediction method embodiments. The bridge time history prediction device can be a virtual device within a bridge time history prediction device, run by the processor of the bridge time history prediction device, or it can be the bridge time history prediction device itself. Figure 6 As shown, the bridge time history prediction device 100 provided in this application embodiment includes: The acquisition module 101 is used to acquire dynamic characteristic data and static characteristic data of the target bridge; wherein, the dynamic characteristic data includes at least one environmental parameter among wind speed, wave height, and seismic motion; and the static characteristic data includes at least one structural parameter among the material strength and damping ratio of the target bridge. The prediction module 102 is used to input the dynamic feature data and the static feature data into a pre-trained bridge response prediction model for processing, to obtain the time history response data of the target bridge and output it.

[0120] In one embodiment, the device 100 further includes: A construction module is used to acquire the static simulation information of the simulated bridge and construct the finite element simulation model of the simulated bridge based on the static simulation information. The simulation module is used to simulate at least one set of simulation dynamic information based on the static simulation information; The simulation processing module is used to input the simulation static information and each of the simulation dynamic information into the finite element simulation model for simulation processing to obtain a training sample set; wherein, the training sample set includes multiple sample pairs, and each sample pair includes simulation static information, simulation dynamic information and simulation response information; The pre-training module is used to pre-train the bridge response prediction model using the training sample set to obtain the pre-trained bridge response prediction model.

[0121] In one embodiment, the device 100 further includes: The first determining module is used to determine the weights corresponding to the time-domain loss function, the peak loss function, and the frequency-domain loss function, respectively. The second determining module is used to determine the total loss function based on the time-domain loss function, the peak loss function, the frequency-domain loss function, and each of the weights.

[0122] In applications, the modules in the bridge time history prediction device can be software program modules, or they can be implemented through different logic circuits integrated in the processor, or they can be implemented through multiple distributed processors.

[0123] like Figure 7 As shown, this application embodiment also provides a bridge time history prediction device 200, including: at least one processor 201 ( Figure 7 The diagram shows only one processor, memory 202, and computer program 203 stored in memory 202 and executable on at least one processor 201. When processor 201 executes computer program 203, it implements the steps in any of the above method embodiments.

[0124] In applications, bridge time history prediction devices may include, but are not limited to, processors and memory. Those skilled in the art will understand that... Figure 7 This is merely an example of a bridge time history prediction device and does not constitute a limitation on the bridge time history prediction device. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0125] In applications, the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0126] In applications, the memory may be an internal storage unit of the bridge time-history prediction device in some embodiments, such as the hard drive or RAM of the bridge time-history prediction device. In other embodiments, the memory may be an external storage device of the bridge time-history prediction device, such as a plug-in hard drive, Smart MediaCard (SMC), Secure Digital (SD) card, or Flash Card equipped on the bridge time-history prediction device. Furthermore, the memory may include both internal storage units and external storage devices of the bridge time-history prediction device. The memory is used to store the operating system, applications, boot loader, data, and other programs, such as program code for computer programs. The memory can also be used to temporarily store data that has been output or will be output.

[0127] It should be noted that the information interaction and execution process between the above-mentioned devices / modules are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0128] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The functional modules in the embodiments can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules can be implemented in hardware or as software functional modules. Furthermore, the specific names of the functional modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the modules in the above-described device can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0129] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps described in the various method embodiments above.

[0130] This application provides a computer program product that, when run on a bridge time history prediction device, enables the bridge time history prediction device to implement the steps described in the above-described method embodiments.

[0131] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable medium can include at least: any entity or device capable of carrying computer program code to the device / bridge time-history prediction device, a recording medium, a computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0132] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0133] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0134] In the embodiments provided in this application, it should be understood that the disclosed apparatus / device and method can be implemented in other ways. For example, the apparatus / device embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or modules may be electrical, mechanical, or other forms.

[0135] The modules described as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0136] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for predicting the time history of a bridge, characterized in that, include: Acquire dynamic and static characteristic data of the target bridge; wherein the dynamic characteristic data includes at least one environmental parameter among wind speed, wave height, and ground motion. Static characteristic data includes at least one structural parameter of the target bridge, such as material strength and damping ratio; The dynamic feature data and the static feature data are input into a pre-trained bridge response prediction model for processing to obtain and output the time history response data of the target bridge.

2. The bridge time history prediction method as described in claim 1, characterized in that, The pre-trained bridge response prediction model includes a temporal information encoding module, a static feature encoding module, a multi-branch attention fusion module, and a multi-task decoding prediction module. The step of inputting the dynamic feature data and the static feature data into a pre-trained bridge response prediction model for processing to obtain the time history response data of the target bridge includes: The dynamic feature data is input into the time-series information encoding module for processing to obtain dynamic feature information; The static feature data is input into the static feature encoding module for processing to obtain static feature information; The dynamic feature information and the static feature information are input into the multi-branch attention fusion module for processing to obtain fused feature information; The fused feature information is input into the multi-task decoding and prediction module for processing to obtain the predicted time history response, peak time history response, and frequency domain time history response of the target bridge.

3. The bridge time history prediction method as described in claim 2, characterized in that, The step of inputting the dynamic feature data into the time-series information encoding module for processing to obtain dynamic feature information includes: Each of the environmental parameters is input into the timing information encoding module for processing to obtain the frequency domain feature information corresponding to each of the environmental parameters. Determine the frequency components corresponding to each of the frequency domain feature information, and convert each of the frequency components into corresponding time domain feature information; Extract the dependent feature sequences of each of the time-domain feature information to obtain the dynamic feature information.

4. The bridge time history prediction method as described in any one of claims 1 to 3, characterized in that, Before acquiring the dynamic and static feature data of the target bridge, the process includes: Obtain the static simulation information of the simulated bridge, and construct the finite element simulation model of the simulated bridge based on the static simulation information; Based on the aforementioned static simulation information, at least one set of dynamic simulation information is simulated. The simulation static information and each of the simulation dynamic information are input into the finite element simulation model for simulation processing to obtain a training sample set; wherein, the training sample set includes multiple sample pairs, and each sample pair includes simulation static information, simulation dynamic information and simulation response information; The bridge response prediction model is pre-trained using the training sample set to obtain the pre-trained bridge response prediction model.

5. The bridge time history prediction method as described in claim 4, characterized in that, The simulated dynamic information includes at least one of simulated wind speed load, simulated wave height load, and simulated seismic load. Based on the simulated static information, at least one set of simulated dynamic information is simulated, including: At least one seismic wave acceleration is selected from a preset earthquake database as the simulated seismic load; Based on the preset wind spectrum and the simulated static information, a random wind speed is generated, and the random wind speed is converted into a simulated wind speed load. Based on the preset wave spectrum and the simulated static information, a random wave height is generated, and the random wave height is converted into a simulated wave height load.

6. The bridge time history prediction method as described in claim 4, characterized in that, The bridge response prediction model is pre-trained using the training sample set to obtain the pre-trained bridge response prediction model, which includes: The sample pairs are normalized to obtain the training dataset; The training dataset is input into the bridge response prediction model, and the bridge response prediction model is pre-trained based on the total loss function to obtain the pre-trained bridge response prediction model.

7. The bridge time history prediction method as described in claim 6, characterized in that, Before inputting the training dataset into the bridge response prediction model and pre-training the bridge response prediction model based on the total loss function to obtain the pre-trained bridge response prediction model, the process includes: Determine the weights corresponding to the time-domain loss function, peak loss function, and frequency-domain loss function respectively; The total loss function is determined based on the time-domain loss function, the peak loss function, the frequency-domain loss function, and each of the weights.

8. A bridge time history prediction device, characterized in that, include: The acquisition module is used to acquire dynamic and static characteristic data of the target bridge; wherein, the dynamic characteristic data includes at least one environmental parameter among wind speed, wave height, and ground motion; and the static characteristic data includes at least one structural parameter among the target bridge's material strength and damping ratio. The prediction module is used to input the dynamic feature data and the static feature data into a pre-trained bridge response prediction model for processing, to obtain the time history response data of the target bridge and output it.

9. A bridge time history prediction device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the bridge time history prediction method according to any one of claims 1 to 7.

10. A computer program product, characterized in that, When the computer program product is run on the bridge time history prediction device, it causes the bridge time history prediction device to implement the steps of the bridge time history prediction method according to any one of claims 1 to 7.

Images