Large-span bridge structure wind-induced response prediction method and device and electronic equipment

Abstract

The invention provides a wind-induced response prediction method and device for a large-span bridge structure and electronic equipment, and the method comprises the steps: obtaining the wind power response data of a bridge in a wind environment; decomposing the wind power response data by adopting a dynamic mode decomposition algorithm to obtain first mode information in the current wind environment; and inputting the mode information into a pre-trained target neural network model, and outputting a wind-induced response prediction result. According to the method, through the synergistic effect of the physical insight of the DMD and the data analysis capability of the neural network, the multi-scale flow mode can be accurately analyzed in real time, meanwhile, the dependence on a large number of computing resources is reduced, the DMD and the neural network model are fused, and the fusion strategy not only solves the limitation of the traditional DMD on nonlinear response prediction, but also improves the prediction accuracy of the DMD. And the defects of the neural network in physical law understanding are overcome, and high efficiency, accuracy and intelligence of wind-induced response prediction are realized.

Description

Technical Field

[0001] This invention belongs to the field of civil engineering technology, specifically relating to a method, device, and electronic equipment for predicting the wind-induced response of long-span bridge structures. Background Technology

[0002] In the field of wind-induced response analysis and prediction of long-span bridge structures, the current technical system covers computational fluid dynamics (CFD) simulation, statistical analysis based on historical monitoring data, application of empirical formulas, and preliminary attempts at data-driven methods such as artificial neural networks (ANN) and convolutional neural networks (CNN). However, these methods have increasingly prominent limitations when facing the complex and ever-changing wind field environment in reality.

[0003] While CFD simulations can meticulously depict flow field details, their high computational cost and lengthy simulation cycles make them insufficient for real-time predictions and emergency situations. Statistical analysis and empirical formulas, while summarizing the average response characteristics of bridges, neglect transient characteristics and multi-scale effects, leading to a significant decline in prediction accuracy under extreme conditions. Data-driven methods, excelling at handling massive amounts of data and pattern recognition, suffer from poor generalization capabilities in new environments when a deep understanding of the physical processes is lacking, especially when data is scarce or of low quality, drastically reducing the reliability of model predictions. Therefore, there is an urgent need for a method capable of rapidly and accurately predicting wind-induced responses. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method, apparatus and electronic device for predicting the wind-induced response of long-span bridge structures, so as to meet the need for rapid and accurate prediction of wind-induced response.

[0005] To achieve the above objectives, the present invention provides the following technical solution: According to a first aspect, embodiments of the present invention provide a method for predicting the wind-induced response of a long-span bridge structure, comprising: acquiring wind response data of the bridge in a wind environment; decomposing the wind response data using a dynamic mode decomposition algorithm to obtain first mode information in the current wind environment; inputting the mode information into a pre-trained target neural network model and outputting the wind-induced response prediction result.

[0006] Optionally, the target neural network model is a multi-scale convolutional neural network. The training process of the target neural network model includes: acquiring wind response data of the bridge under wind conditions and corresponding wind-induced response data; constructing a time series based on the wind response data, decomposing the time series using a dynamic mode decomposition algorithm to obtain multiple second mode information, using the second mode information as training samples, and the corresponding wind-induced response data as training labels; inputting the training samples and corresponding training labels into the target neural network model, and training with the goal of minimizing the loss function to obtain the trained target neural network model.

[0007] Optionally, a dynamic mode decomposition algorithm is used to decompose the wind response data to obtain the first mode information in the wind environment, including: dividing the wind response data of the bridge in the wind environment according to different times; performing singular value decomposition on the wind response data at any time and the next time; calculating the linear mapping approximation matrix between two adjacent times based on the singular value decomposition results of the wind response data at any time and the next time; solving the eigenvalue problem based on the linear mapping approximation matrix to obtain eigenvalues ​​and eigenvectors; and analyzing the eigenvalues ​​and eigenvectors to obtain the first mode information in the wind environment.

[0008] Optionally, the multi-scale convolutional neural network includes: an input layer for receiving input data; a multi-scale convolutional layer composed of multiple convolutional units, each with filters of different sizes, for capturing features at different scales of the image; a linear rectified layer for performing non-linear mapping; a pooling layer for reducing the spatial dimension of the feature map; a fully connected layer for classification or regression, mapping the features extracted by the convolutional layers to the predicted output; and an output layer for outputting the wind-induced response prediction result.

[0009] Optionally, the target neural network model includes prior knowledge of the bridge's structural characteristics and / or fundamental laws of fluid dynamics.

[0010] Optionally, the target neural network model is a model that has undergone transfer learning.

[0011] According to a second aspect, embodiments of the present invention provide a method for predicting the wind-induced response of a long-span bridge structure, comprising: a data acquisition module for acquiring wind response data of the bridge in a wind environment; a first mode information determination module for decomposing the wind response data using a dynamic mode decomposition algorithm to obtain the first mode information currently in the wind environment; and a wind-induced response prediction result output module for inputting the mode information into a pre-trained target neural network model and outputting the wind-induced response prediction result.

[0012] According to a third aspect, an embodiment of the present invention provides an electronic device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the steps of the wind-induced response prediction method for long-span bridge structures described in the first aspect or any embodiment of the first aspect.

[0013] According to a fourth aspect, embodiments of the present invention provide a computer storage medium storing computer instructions thereon, which, when executed by a processor, implement the steps of the wind-induced response prediction method for long-span bridge structures as described in the first aspect or any embodiment of the first aspect.

[0014] This method provides a wind-induced response prediction approach for long-span bridge structures. It utilizes Dynamic Mode Decomposition (DMD) to decompose dynamic modes, identify dominant dynamic modes and their corresponding responses, and then uses the mode information provided by DMD for feature extraction and learning through a target neural network to construct a predictive model for forecasting the bridge's wind-induced response. By combining the physical insights of DMD with the data analysis capabilities of neural networks, multi-scale flow patterns can be analyzed in real-time and accurately, while reducing reliance on large amounts of computational resources. This fusion strategy not only overcomes the limitations of traditional DMD in nonlinear response prediction but also compensates for the shortcomings of neural networks in understanding physical laws, achieving efficient, accurate, and intelligent wind-induced response prediction. This provides strong technical support for safety assessment and decision-making in bridge engineering.

[0015] This embodiment provides a method for predicting the wind-induced response of long-span bridge structures. By using a dynamic mode decomposition (DMD) algorithm to decompose the time series, the dynamic patterns of the bridge under wind conditions can be extracted. These patterns, used as training samples, can more accurately capture the response characteristics of the bridge under different wind conditions, thereby improving prediction accuracy.

[0016] Other advantages, objectives, and features of the invention will be set forth in the following description and will be apparent to those skilled in the art in some respects, or may be learned by practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0017] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration: Figure 1 This is a flowchart illustrating a specific example of a wind-induced response prediction method for long-span bridge structures proposed in this invention. Figure 2This is a schematic diagram of the multi-scale convolutional neural network results proposed in this invention; Figure 3 This is a schematic diagram illustrating the process of predicting the wind-induced response of long-span bridge structures by integrating DMD-MSCNN as proposed in this invention. Figure 4 This is a schematic block diagram of a specific example of an electronic device in an embodiment of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can also refer to the internal connection of two components; and they can refer to a wireless connection or a wired connection. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0020] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0021] This invention provides a method for predicting the wind-induced response of long-span bridge structures, such as... Figure 1 As shown, it includes: S101, Obtain wind response data of the bridge in windy conditions; S102, The dynamic mode decomposition algorithm is used to decompose the wind response data to obtain the first mode information in the current wind environment; S103 inputs the pattern information into the pre-trained target neural network model and outputs the wind-induced response prediction result.

[0022] For example, wind response data includes signals of bridge displacement, velocity, acceleration, stress, strain, etc., changing over time. This wind response data can be obtained through wind tunnel testing or field monitoring. The wind response data is then processed, including data cleaning to remove noise and outliers.

[0023] To enable the target neural network model to better understand wind response data, and considering that wind response data typically contains a large amount of noise and redundant information, this embodiment employs Dynamic Mode Decomposition (DMD) to extract the key dynamic characteristics of the system, i.e., mode information, from the complex data. This simplifies the data, removes unnecessary complexity, and makes the data processed by the neural network more refined. In this embodiment, DMD is used to process the wind response data, and after obtaining the first mode information, this first mode information is input into the target neural network model for prediction.

[0024] Dynamic Mode Decomposition (DMD) algorithms are applied in fluid analysis, complex system construction, and data mining. Developed from the Koopman algorithm, DMD innovatively combines the characteristics of power spectral analysis in the time dimension and principal component analysis in the spatial dimension. By decomposing time-series data into eigenvalues ​​and eigenvectors, it approximates and simplifies high-dimensional nonlinear systems into low-dimensional linear system representations, ultimately achieving current state reconstruction and short-term state prediction. Its specific principles are as follows: Extraction of the original flow field N Each snapshot contains m discrete physical quantities (such as velocity, pressure, temperature, etc.) within the sampling region space. The snapshot information at each time step is denoted as: ; (1) in and Let the time interval between them be t, and let the times t1, t2, ..., t be... N−1 The data is stored in a matrix V 1. Let t2, t3, ..., t N The data is stored in a matrix V 2 in: ; (2) ; (3) When the number of samples N If a sufficient number of such systems are considered to be linear, then a high-dimensional linear matrix exists. A This makes the physical quantity v at each moment... i+1 Both can be obtained through the following physical quantity v i Linear description: ; (4) Therefore, for a linear system, when the initial state v of the system is obtained... i and linear matrix A Then, the state of the system at any given time can be obtained. (Matrix) V 1 andV 2 can be written in Krylov's subspace form: ; (5) ; (6) ; (7) For discrete flow field matrix V 1 and V Both are m rows N A high-dimensional matrix with -1 columns, directly compared to a linear matrix. A Finding eigenvalues ​​is very difficult, so we first need to analyze the matrix. V 1. Perform singular value decomposition, where U is an m×m matrix and K is... N -1× N A -1 order matrix, and to achieve dimensionality reduction, singular values ​​that are approximately equal to 0 are usually removed, considering only the first r orders, thus a linear matrix. A It can be approximated as : ; (8) (9) At this point, the dimension-reduced linear matrix... Perform eigenvalue decomposition to obtain the corresponding eigenvalues. and eigenvectors have: (10) The modes of dynamic mode decomposition are as follows: (11) Unlike intrinsic orthogonal decomposition, in solving dynamic mode decomposition, the energy magnitude of each mode in DMD can be represented by the corresponding mode 2 norm. The relative magnitude of the energy can also be approximated by calculating the 2 norm of matrix E, based on the magnitude of each mode's contribution to the flow field.

[0025] (12) Due to linear matrix A It contains the time evolution characteristics of the entire dynamic system and reveals information such as the system's growth rate and frequency; therefore, the approximate matrix... A The real part of the eigenvalues ​​contains the modal-related growth rate, and the imaginary part contains the modal-related frequency information. Since eigenvalues ​​and eigenvectors are generally in complex form, the eigenvalues ​​are transformed into logarithmic form: (13) (14) (15) Where: t is the sampling time interval; The superscript Re represents the real part of the eigenvalue. The superscript Im represents the imaginary part of the eigenvalue, reflecting the frequency information of the i-th mode; It reflects the growth rate or decay rate information of the i-th mode; This represents the frequency corresponding to the i-th mode. When the growth rate is positive (outside the unit circle), the corresponding mode is an unstable mode; when the growth rate is negative (inside the unit circle), the corresponding mode is a stable mode; when the growth rate is 0 (on the unit circle), the corresponding mode is a periodic mode.

[0026] Based on the above principles, this embodiment divides the wind response data of the bridge in a wind environment according to different time periods; performs singular value decomposition on the wind response data at any time and the next time; calculates the linear mapping approximation matrix between two adjacent time periods based on the singular value decomposition results of the wind response data at any time and the next time; solves the eigenvalue problem based on the linear mapping approximation matrix to obtain eigenvalues ​​and eigenvectors; and analyzes the eigenvalues ​​and eigenvectors to obtain the first mode information in the current wind environment.

[0027] Specifically, the time series of wind response data is first converted into a matrix form suitable for DMD analysis. Typically, the data is organized into two snapshot matrices, X and X', where X contains data from time t=0 to t=T-1, and X' contains data from time t=1 to t=T. Then, the snapshot matrices are constructed. Assuming there is data from n sensors measured over m time steps, two snapshot matrices can be constructed: X = [x_1, x_2, ..., x_m], where x_i is the measurement vector at time t=i.

[0028] X' = ​​[x_2, x_3, ..., x_{m+1}], where x_{i+1} is the measurement vector at time t=i+1.

[0029] Then apply the DMD algorithm, specifically: Least square fitting of the data: Solve for the linear mapping A such that X' ≈ AX.

[0030] Singular Value Decomposition (SVD): For any matrix X ∈ Rm×n, its singular value decomposition can be expressed as: Where: U is an m×m orthogonal matrix whose column vectors are the left singular vectors of X. Σ is an m×n diagonal matrix whose diagonal elements are the singular values ​​of X, which are non-negative real numbers arranged in descending order. It is an n×n orthogonal matrix whose column vectors are the right singular vectors of X. This represents the transpose of the K matrix.

[0031] Approximation of the mapping operator: An approximation of the mapping operator A is calculated using the singular value decomposition (SVD) of X and X'. Specifically, the singular value decomposition of matrix X is performed to obtain... Perform singular value decomposition on matrix X' (the transpose of X) to obtain Retain the largest singular values, and based on the number of retained singular values, extract the corresponding singular values ​​from Σ and Σ' to construct an approximate diagonal matrix Σ'Σ. and The corresponding part in the matrix is ​​used to construct a linear mapping approximation matrix A'.

[0032] Eigenvalues ​​and eigenvectors: Solving the eigenvalue problem based on the linear mapping approximation matrix A' yields the eigenvalues ​​λ and eigenvectors. Where λ corresponds to the growth rate and frequency of the DMD mode, This corresponds to the DMD mode itself.

[0033] In DMD analysis, the real part of the eigenvalue λ represents the growth rate of the pattern, and the imaginary part represents the frequency of the pattern. Eigenvectors The DMD mode itself describes the dynamic shape of the system. These eigenvalues ​​and eigenvectors allow for the analysis of the system's dynamic behavior; each DMD mode represents a specific dynamic behavior, for example, a mode might correspond to vortex-induced vibration of a bridge.

[0034] To evaluate the accuracy and reliability of the model information output by DMD, the obtained DMD model and corresponding eigenvalues ​​can be used to predict future states. The prediction results can then be compared with actual observations or numerical simulations to determine their accuracy and reliability.

[0035] Since wind response data contains real-time vibration information of the bridge under wind load, and the first mode information extracted after DMD decomposition reflects the main vibration characteristics of the bridge, this embodiment performs dynamic mode decomposition on the wind response data to enable the target neural network model to better understand the dynamic features represented by the wind response data and to reduce the data processing dimensionality. The resulting eigenvalues ​​λ and eigenvectors are then analyzed. The first pattern information is used as the input to the pre-trained target neural network model. The target neural network model can be a multi-scale convolutional neural network (MSCNN) or other neural network models. This embodiment does not limit it. Those skilled in the art can determine it as needed. The target neural network model predicts the wind-induced response prediction result based on the input first pattern information.

[0036] This embodiment provides a method for predicting the wind-induced response of long-span bridge structures. This method utilizes Dynamic Mode Decomposition (DMD) to decompose dynamic modes, identify dominant dynamic modes and their corresponding responses, and then uses the mode information provided by DMD to extract and learn features through a target neural network to construct a predictive model for predicting the wind-induced response of the bridge. Through the synergistic effect of the physical insights of DMD and the data analysis capabilities of neural networks, multi-scale flow patterns can be analyzed in real time and accurately, while reducing reliance on large amounts of computing resources. This fusion strategy not only solves the limitations of traditional DMD in nonlinear response prediction but also compensates for the shortcomings of neural networks in understanding physical laws, achieving efficient, accurate, and intelligent wind-induced response prediction, and providing strong technical support for safety assessment and decision-making in the field of bridge engineering.

[0037] As an optional implementation, the target neural network model is a multi-scale convolutional neural network, and the training process of the target neural network model includes: Obtain wind force response data and corresponding wind-induced response data of the bridge under wind conditions; A time series was constructed based on wind response data. The time series was decomposed using a dynamic mode decomposition algorithm to obtain multiple second mode information. The second mode information was used as training samples, and the corresponding wind response data was used as training labels. The training samples and corresponding training labels are input into the target neural network model, and the model is trained with the goal of minimizing the loss function, resulting in a well-trained target neural network model.

[0038] For example, wind response data includes signals such as bridge displacement, velocity, acceleration, stress, and strain that change over time. Obtaining the bridge's wind response data and corresponding wind-induced response data under wind conditions can be achieved through wind tunnel testing or on-site monitoring. Wind-induced response data is the actual response data of the bridge under the same wind environment. Specifically, it may include: the bridge's vibration acceleration value at a specific moment, the bridge's displacement at a specific moment, the bridge's velocity value at a specific moment, or other response quantities. This embodiment does not limit the wind-induced response data; it can be set according to specific problems and needs, and may also include actual measured values ​​of other physical quantities such as bridge stress, strain, and deformation.

[0039] After obtaining the wind response data of the bridge under wind conditions and the corresponding wind-induced response data, data preprocessing can be performed, such as data cleaning, handling missing values, standardizing or normalizing the data, to ensure data quality and consistency.

[0040] A time series is constructed based on wind response data. The time series is then decomposed using a dynamic mode decomposition algorithm to obtain multiple second-mode information. The specific implementation process for obtaining the first-mode information is described above and will not be repeated here. The second-mode information is used as training samples, and the wind response data is used as training labels. The training samples and corresponding training labels are input into the target neural network model for supervised training, aiming to minimize the loss function. The trained target neural network model is then obtained. The loss function can be the mean squared error (MSE). This embodiment does not limit the type of loss function; those skilled in the art can determine it as needed.

[0041] Furthermore, the acquired wind response data of the bridge under wind conditions and the corresponding wind-induced response data can be allocated proportionally into training, validation, and test sets. The specific proportions can be 70%, 15%, 15%, or similar allocations. This embodiment does not limit the specific proportions, and those skilled in the art can determine them as needed. The trained target neural network model is evaluated on the validation set, and the model parameters are adjusted until satisfactory. Finally, the model's generalization ability is evaluated on the test set to ensure that the model performs well even on unseen data.

[0042] This embodiment provides a method for predicting the wind-induced response of long-span bridge structures. By using a dynamic mode decomposition (DMD) algorithm to decompose the time series, the dynamic patterns of the bridge under wind conditions can be extracted. These patterns, used as training samples, can more accurately capture the response characteristics of the bridge under different wind conditions, thereby improving prediction accuracy.

[0043] As an alternative implementation, multi-scale convolutional neural networks, such as Figure 2 As shown, it includes: The input layer receives input data; the multi-scale convolutional layer consists of multiple convolutional units, each with filters of different sizes, used to capture features at different scales of the image; the linear rectified layer performs non-linear mapping; the pooling layer reduces the spatial dimension of the feature map; the fully connected layer is used for classification or regression, mapping the features extracted by the convolutional layers to the predicted output; and the output layer outputs the wind-induced response prediction result.

[0044] For example, in the precise field of vibration signal analysis and response prediction, Multi-Scale Convolutional Neural Networks (MSCNN), as a cutting-edge deep learning architecture, have demonstrated unique advantages and powerful potential. Unlike traditional single-scale analysis methods, MSCNN cleverly incorporates the concept of multi-scale feature extraction. Through a series of carefully designed convolutional layers, it can simultaneously capture and analyze subtle features at different scales in both the frequency and time domains of vibration signals, thereby providing deep insights into the vibration response of complex systems. In practical applications, the implementation process of MSCNN encompasses the entire process from data preprocessing to feature extraction, multi-scale fusion, and finally, response prediction. First, the original vibration signal undergoes preprocessing to eliminate noise interference and standardize the signal amplitude, ensuring the accuracy and reliability of subsequent analysis. Then, MSCNN, through its multi-scale convolutional layers, extracts multi-level features from the vibration signal using different window sizes and strides. This process is equivalent to meticulously dissecting the signal at different frequency bandwidths and time resolutions, obtaining a comprehensive feature description from low to high frequencies and from short to long time windows. In particular, the multi-scale fusion stage of MSCNN is its essence. Through strategies such as cascading, weighted summation, or attention mechanisms, it integrates features extracted from different scales to construct a comprehensive signal representation. This representation not only preserves local transient information but also encompasses global periodic patterns, greatly enhancing the model's ability to understand the intrinsic laws of vibration response. Based on this, MSCNN can perform accurate response prediction. The key concepts are as follows: Convolutional layer: A convolutional layer consists of multiple convolutional units, the parameters of which are optimized through backpropagation. Convolutional operations are primarily used to extract image features; as the number of convolutional layers increases, multi-layer networks can extract more complex image features.

[0045] Linear rectified (Rectified Linear Units layer, ReLU layer): This mainly refers to the activation function operation using the linearly rectified ReLU function.

[0046] Pooling layer: After convolution, the image still has many dimensional features. The feature matrix is ​​divided into several individual blocks, and the maximum or average value is taken, which plays a role in dimensionality reduction.

[0047] Fully-connected layer: Combines all local features and the feature matrices of each channel into a vector representation, and calculates the score for each class.

[0048] The detailed process of using the MSCNN described above for wind-induced response prediction is as follows: First, data preparation is performed, including: data collection, obtaining bridge structure response data under different wind speeds and directions through DMD decomposition; data preprocessing, cleaning data, handling missing values, standardizing or normalizing data to ensure data quality and consistency; and data partitioning, dividing the dataset into training, validation, and test sets, typically in proportions of 70%, 15%, 15%, or similar. Second, feature extraction is performed, including: data transformation, converting time-series data into a format suitable for convolutional neural network input, such as the sliding window method, creating input sequences at multiple time steps to predict future responses; and multi-scale representation, considering that wind-induced responses may involve features at different time scales, preprocessing the original data using various sampling rates or window sizes so that MSCNN can learn features at different scales. Then, the MSCNN model is constructed, including: multi-scale convolutional layers, each with different filter sizes to capture features at different scales. For example, convolutional kernels of different sizes such as 1x1, 3x3, and 5x5 can be set; pooling layers are used to reduce the spatial dimension of the feature map, reducing computation while preserving important features; fully connected layers are used for the final classification or regression task, mapping the features extracted by the convolutional layers to the predicted output. Then, model training is performed, including: loss function selection, choosing an appropriate loss function based on the prediction task, such as mean squared error (MSE) suitable for regression problems; optimizer selection, choosing an appropriate optimization algorithm, such as Adam or SGD, to minimize the loss function; hyperparameter tuning, adjusting hyperparameters such as learning rate, batch size, and regularization parameters to optimize model performance; the training process: training the model using training set data and adjusting the model using validation set data to prevent overfitting. Finally, model evaluation and validation are performed, evaluating model performance on the validation set, adjusting model parameters until satisfactory, and evaluating the model's generalization ability on the test set to ensure good performance on unseen data.

[0049] The trained model is deployed to the actual system, receives real-time or historical wind load data, performs wind-induced response prediction, interprets the model prediction results, compares them with the actual situation, and evaluates the accuracy and reliability of the prediction.

[0050] Furthermore, continuous monitoring can be performed, regularly monitoring model performance to check for performance degradation, and retraining the model with newly collected data as needed to maintain the model's timeliness and accuracy.

[0051] As an alternative implementation, the target neural network model contains prior knowledge of the structural characteristics of the bridge and / or the basic laws of fluid dynamics, which helps the model make reasonable assumptions when data is scarce.

[0052] As an alternative implementation, the target neural network model is a model that has undergone transfer learning.

[0053] For example, MSCNN can be pre-trained using data from other related domains through transfer learning, which helps improve model performance when the target dataset is small.

[0054] Furthermore, when data volume is insufficient, models are prone to overfitting. Regularization techniques (such as L1 / L2 regularization, Dropout, etc.) and carefully designed optimization strategies can be used to control model complexity, avoid overfitting, and thus maintain good generalization ability on small datasets.

[0055] Based on the above description, the implementation process of wind-induced response prediction for long-span bridge structures fused with DMD-MSCNN provided in this embodiment is as follows: Figure 3 As shown, the first step is data preparation, including: data collection, obtaining bridge structure response data under different wind speeds and directions through DMD decomposition; data preprocessing, cleaning data, handling missing values, standardizing or normalizing data to ensure data quality and consistency; and data partitioning, dividing the dataset into training, validation, and test sets. The training set is used to train a wind-induced response prediction model for long-span bridge structures that integrates DMD and MSCNN. Specifically, DMD identifies the dominant dynamic patterns and their corresponding responses, and then the pattern information provided by DMD is used for feature extraction and learning through a multi-scale convolutional neural network to obtain the prediction model and the final prediction result. The test set is used for model evaluation and validation. The model performance is evaluated on the validation set, and the model parameters are updated until satisfactory.

[0056] This embodiment provides a method for predicting the wind-induced response of long-span bridge structures, including: The data acquisition module is used to acquire wind response data of the bridge in windy conditions; for details, please refer to the corresponding description in the method embodiment, which will not be repeated here.

[0057] The first mode information determination module is used to decompose the wind response data using a dynamic mode decomposition algorithm to obtain the first mode information in the current wind environment; for details, please refer to the corresponding description in the method embodiment, which will not be repeated here.

[0058] The wind-induced response prediction output module is used to input pattern information into a pre-trained target neural network model and output the wind-induced response prediction result. For details, please refer to the corresponding description in the method embodiment; it will not be repeated here.

[0059] This application also provides an electronic device, such as... Figure 4As shown, processor 501 and memory 502 are connected via a bus or other means.

[0060] Processor 501 can be a central processing unit (CPU). Processor 501 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, or combinations of the above types of chips.

[0061] The memory 502, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the wind-induced response prediction method for long-span bridge structures in this embodiment of the invention. The processor executes various functional applications and data processing by running the non-transitory software programs, instructions, and modules stored in the memory.

[0062] Memory 502 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory 502 may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0063] The one or more modules are stored in the memory 502, and when executed by the processor 501, they perform actions such as... Figure 1 The wind-induced response prediction method for long-span bridge structures in the illustrated embodiment.

[0064] For specific details regarding the aforementioned electronic devices, please refer to the relevant documentation. Figure 1 The relevant descriptions and effects in the illustrated embodiments are for understanding purposes only and will not be repeated here.

[0065] This embodiment also provides a computer storage medium storing computer-executable instructions that can execute the wind-induced response prediction method for long-span bridge structures in any of the above method embodiments. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium may also include combinations of the above types of memory.

[0066] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A method for predicting the wind-induced response of long-span bridge structures, characterized in that, include: Obtain wind response data of the bridge under wind conditions; The wind response data is decomposed using a dynamic mode decomposition algorithm to obtain the first mode information in the current wind environment; The pattern information is input into a pre-trained target neural network model, and the wind-induced response prediction result is output.

2. The method for predicting the wind-induced response of a long-span bridge structure according to claim 1, characterized in that, The target neural network model is a multi-scale convolutional neural network. The training process of the target neural network model includes: Obtain wind force response data and corresponding wind-induced response data of the bridge under wind conditions; A time series was constructed based on wind response data. The time series was decomposed using a dynamic mode decomposition algorithm to obtain multiple second mode information. The second mode information was used as training samples, and the corresponding wind response data was used as training labels. The training samples and corresponding training labels are input into the target neural network model, and the model is trained with the goal of minimizing the loss function, resulting in a well-trained target neural network model.

3. A method for predicting the wind-induced response of a long-span bridge structure according to claim 1 or 2, characterized in that, The wind response data is decomposed using a dynamic mode decomposition algorithm to obtain the first mode information in the current wind environment, including: The wind response data of the bridge under wind conditions were divided according to different time periods; Perform singular value decomposition on the wind response data at any given time and at the next time. Based on the singular value decomposition results of the wind response data at any time and the next time, calculate the linear mapping approximation matrix between two adjacent time points; Based on the linear mapping approximation matrix, solve the eigenvalue problem to obtain eigenvalues ​​and eigenvectors; By analyzing the eigenvalues ​​and eigenvectors, we can obtain the first mode information in the current wind environment.

4. The method for predicting the wind-induced response of a long-span bridge structure according to claim 2, characterized in that, Multi-scale convolutional neural networks include: The input layer is used to receive input data; Multi-scale convolutional layers consist of multiple convolutional units, each with filters of different sizes, used to capture features at different scales of the image; Linear rectifier layer, used for nonlinear mapping; Pooling layers are used to reduce the spatial dimensionality of feature maps; Fully connected layers, used for classification or regression, map the features extracted by the convolutional layers to the predicted output; The output layer is used to output the wind-induced response prediction results.

5. The method for predicting the wind-induced response of a long-span bridge structure according to claim 1, characterized in that, The target neural network model contains prior knowledge of the structural characteristics of the bridge and / or the basic laws of fluid dynamics.

6. The method for predicting the wind-induced response of a long-span bridge structure according to claim 2, characterized in that, The target neural network model is a model that has undergone transfer learning.

7. A method for predicting the wind-induced response of long-span bridge structures, characterized in that, include: The data acquisition module is used to acquire wind response data of the bridge in windy conditions; The first mode information determination module is used to decompose the wind response data using a dynamic mode decomposition algorithm to obtain the first mode information in the current wind environment. The wind response prediction result output module is used to input pattern information into a pre-trained target neural network model and output the wind response prediction result.

8. An electronic device, the device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor performs the steps of the wind-induced response prediction method for long-span bridge structures according to any one of claims 1-6.

9. A computer storage medium storing computer instructions thereon, characterized in that, When executed by the processor, this instruction implements the steps of the wind-induced response prediction method for long-span bridge structures as described in any one of claims 1-6.

Images