Method and device for identifying wind load spectrum of high-rise structures based on transfer deep learning

Abstract

The present disclosure provides a method and device for identifying the wind load spectrum of a high-rise structure based on transfer deep learning. The method includes: calculating the dynamic response time-history data of the target structure under the target wind load to obtain the corresponding observation response power spectrum matrix; performing discrete cosine transform decomposition on the wind speed spectrum of the target wind load to obtain the DCT coefficients after dimensionality reduction; constructing a deep learning model based on a physical guidance mechanism, using the observation response power spectrum matrix as the input of the deep learning model and the DCT coefficients as the output, and training the deep learning model to learn the mapping relationship between the two; using the trained deep learning model to output the corresponding DCT coefficients based on the input observation response power spectrum matrix, and reconstructing the power spectrum matrix of the wind load based on the DCT coefficients. The technical solution of the embodiment of the present disclosure can effectively establish the correlation between the response power spectrum and the load power spectrum, and ensure the accuracy of wind load identification.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, and in particular to a method and device for identifying wind load spectra of high-rise structures based on transfer deep learning. Background Art

[0002] In civil engineering, analyzing the dynamic response of large structures such as high-rise buildings and bridges under natural wind loads is crucial for structural health monitoring, safety assessment, and design optimization. Because wind loads are characterized by significant randomness and spatial distribution, accurately identifying their power spectrum matrix presents a key challenge.

[0003] Current technical solutions for wind load identification mostly rely on solving an inverse problem—inferring the load from the structural response. However, this process is susceptible to noise, leading to unstable solutions and requiring the use of regularization techniques, which increases computational complexity. Furthermore, identifying the full-degree-of-freedom load spectrum requires high-dimensional observation data, making it difficult to implement in practical projects and lacking noise immunity. Therefore, effectively establishing a correlation between the response power spectrum and the load power spectrum to ensure accurate wind load identification has become a pressing technical challenge. Summary of the Invention

[0004] The present disclosure provides a method and device for identifying wind load spectra of high-rise structures based on transfer deep learning, which can effectively establish a correlation between the response power spectrum and the load power spectrum to at least a certain extent, thereby ensuring the accuracy of wind load identification.

[0005] According to one aspect of the present disclosure, a method for identifying wind load spectra of high-rise structures based on transfer deep learning is provided, comprising:

[0006] Calculate the dynamic response time history data of the target structure under the target wind load and obtain the corresponding observation response power spectrum matrix;

[0007] Performing discrete cosine transform decomposition on the wind speed spectrum of the target wind load to obtain DCT coefficients after dimensionality reduction;

[0008] A deep learning model for wind load identification based on a physical guidance mechanism is constructed, the observed response power spectrum matrix is used as the input of the deep learning model, and the DCT coefficients are used as the output. The deep learning model is trained to learn the mapping relationship between the two. The physical guidance mechanism embeds the frequency domain transfer equation of wind load and structural response into the loss function to constrain the model training process.

[0009] When performing wind load identification, the trained deep learning model is used to output corresponding DCT coefficients based on the input observation response power spectrum matrix, and the power spectrum matrix of the wind load is reconstructed according to the DCT coefficients;

[0010] After the deep learning model is trained, the method further includes:

[0011] Based on the transfer learning strategy, for wind loads at discrete points of different frequencies, the model of the target frequency is initialized using the model parameters trained based on the wind load of the source frequency, and the model parameters are fine-tuned using the target frequency data to obtain a deep learning model corresponding to the wind load of the target frequency.

[0012] According to one aspect of the present disclosure, a device for identifying wind load spectra of high-rise structures based on transfer deep learning is provided, comprising:

[0013] A calculation module is used to calculate the dynamic response time history data of the target structure under the target wind load to obtain the corresponding observation response power spectrum matrix;

[0014] a decomposition module, configured to perform discrete cosine transform decomposition on the wind speed spectrum of the target wind load to obtain DCT coefficients after dimensionality reduction;

[0015] A training module is configured to construct a deep learning model for wind load identification based on a physical guidance mechanism, using the observed response power spectrum matrix as input to the deep learning model and the DCT coefficients as output, and enabling the deep learning model to learn the mapping relationship between the two through training. The physical guidance mechanism embeds a frequency domain transfer equation between wind load and structural response into a loss function to constrain the model training process.

[0016] A processing module is used to use the trained deep learning model to output corresponding DCT coefficients based on the input observation response power spectrum matrix when performing wind load identification, and reconstruct the power spectrum matrix of the wind load based on the DCT coefficients;

[0017] After the deep learning model is trained, the processing module is further configured to:

[0018] Based on the transfer learning strategy, for wind loads at discrete points of different frequencies, the model of the target frequency is initialized using the model parameters trained based on the wind load of the source frequency, and the model parameters are fine-tuned using the target frequency data to obtain a deep learning model corresponding to the wind load of the target frequency.

[0019] According to another aspect of the present disclosure, an electronic device is provided, comprising: a memory storing execution instructions; and a processor executing the execution instructions stored in the memory, so that the processor executes the high-rise structure wind load spectrum identification method based on transfer deep learning according to any embodiment of the present disclosure.

[0020] According to another aspect of the present disclosure, a readable storage medium is provided, in which execution instructions are stored. When the execution instructions are executed by a processor, they are used to implement the high-rise structure wind load spectrum identification method based on transfer deep learning according to any embodiment of the present disclosure.

[0021] According to another aspect of the present disclosure, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the method for identifying wind load spectra of high-rise structures based on transfer deep learning according to any embodiment of the present disclosure.

[0022] In the technical solutions provided in some embodiments of the present application, the dynamic response time-history data of the target structure under the target wind load is calculated to obtain the corresponding observation response power spectrum matrix, and the wind speed spectrum of the target wind load is discrete cosine transform decomposition to obtain the DCT coefficients after dimensionality reduction. Then, a deep learning model for wind load identification based on a physical guidance mechanism is constructed, and the observation response power spectrum matrix is used as the input of the deep learning model, and the DCT coefficients are used as its output. Through training, the wind load model is enabled to learn the mapping relationship between the two. The physical guidance mechanism embeds the frequency domain transfer method of wind load and structural response into the loss function to constrain the model training process. When performing wind load identification, the trained deep learning model is used to output the corresponding DCT coefficients based on the input observation response power spectrum matrix, and the power spectrum matrix of the wind load is reconstructed according to the output DCT coefficients to achieve wind load identification.

[0023] In this way, by establishing a deep learning model to directly find the connection between the response power spectrum and the load power spectrum, the ill-posed problem of the inverse process of the traditional method under noise is avoided, and the use of regularization technology is avoided, making the recognition process more stable and accurate, and improving its noise resistance performance.

[0024] Furthermore, by introducing discrete cosine transforms for dimensionality reduction, it is possible to identify the wind load spectrum for all degrees of freedom using partial structural response observations. Furthermore, embedding a physical guidance mechanism within the deep learning model significantly reduces the amount of data required for model training, improving the training efficiency while ensuring the accuracy of the deep learning model's recognition results.

[0025] Furthermore, by leveraging a transfer learning strategy, reusing the physics-guided feature extraction capabilities of the source frequency model and fine-tuning with a small amount of target frequency data, efficient and high-precision identification of cross-frequency wind loads is achieved. This approach significantly reduces data requirements and computational costs, providing a scalable technical path for full-band load identification in complex wind environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings illustrate exemplary embodiments of the present disclosure and together with the description serve to explain the principles of the present disclosure. These drawings are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification.

[0027] Figure 1 A schematic diagram of a process for identifying wind load spectra of high-rise structures based on transfer deep learning according to an embodiment of the present application is shown;

[0028] Figure 2 A schematic flow chart of a method for identifying wind load spectra of high-rise structures based on transfer deep learning according to another embodiment of the present application is shown;

[0029] Figure 3 and Figure 4 shows the network identification results for a smooth and uniform wind speed spectrum at different frequencies according to one embodiment of the present application;

[0030] Figure 5 A schematic diagram showing a comparison of loss functions when a network training process is constrained with and without a physical guidance mechanism according to an embodiment of the present application is shown;

[0031] Figure 6 and Figure 7 shows the network identification results for the stationary non-uniform wind speed spectrum at different frequencies according to one embodiment of the present application;

[0032] Figure 8 A schematic block diagram of a high-rise structure wind load spectrum identification device based on transfer deep learning according to an embodiment of the present application is shown;

[0033] Figure 9 A schematic diagram of the structure of a computer system suitable for implementing an electronic device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0034] The present disclosure is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are intended only to illustrate the relevant content and are not intended to limit the present disclosure. It should also be noted that, for ease of description, only the portions relevant to the present disclosure are shown in the accompanying drawings.

[0035] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in the present disclosure can be combined with each other. The technical solution of the present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0036] In practical engineering, structures are constantly subjected to various forms of dynamic loads. Therefore, dynamic loads on structures play a crucial role in structural health monitoring, safety and reliability analysis, and dynamic optimization design. These loads are often influenced by various factors and exhibit a degree of randomness, making explicit expressions difficult to identify and significantly increasing the difficulty of identification.

[0037] Dynamic loads can be divided into concentrated dynamic loads and distributed dynamic loads according to their nature. Distributed loads are more challenging to identify because they are not only dynamic in time, but the time history and amplitude of the loads also change with the change of spatial position. The most common distributed load in nature is the wind load. In recent years, scholars have proposed some methods. For example, some related technologies use continuous and discrete Kalman filtering and Taylor polynomial expansion to identify fluctuating wind loads, but the fluctuating wind needs to be regarded as a random walk process or zero-mean white noise. Other related technologies construct a stable input estimator based on Kalman filtering to estimate the structural state and unknown inputs, and use the measured data of the Guangzhou Tower to verify it and extend it to complex structures. However, the above methods all identify the determined wind load.

[0038] Research has also been conducted on randomly distributed loads, with frequency domain methods, probabilistic estimation, and interval estimation being common. For example, some techniques consider frequency domain identification models that account for frequency response function errors and response measurement errors, analyzing the error propagation and amplification effects of load identification, thereby theoretically identifying and controlling these errors. Another group of researchers uses B-spline functions to fit spatial distribution functions and establishes the transfer relationship between excitation and response using classic frequency response function methods in the frequency domain. This method can provide good results for annularly distributed dynamic loads and can be used to identify dynamic loads at satellite-rocket interfaces. Other researchers have studied methods such as Monte Carlo and KL expansion to identify statistical properties such as the mean and variance of loads. Alternatively, interval estimation methods have been employed, combining genetic algorithms with Latin hypercube sampling and improved L-curve methods to identify upper and lower bounds on distributed loads. However, these methods do not account for the spatial correlation of distributed loads, which is present in many natural distributed dynamic loads, such as wind loads.

[0039] While some methods, based on most existing methods, consider the spatial correlation of loads—that is, the load power spectrum can obey an expression that varies with distance, fit this expression with orthogonal polynomials, establish a relationship between the polynomial coefficients and the response power spectrum, and reconstruct the load power spectrum by observing the identification coefficients—these methods, when calculating the load power spectrum from the response power spectrum, often suffer from ill-posedness due to the influence of noise. Regularization techniques are generally required, making the calculation more complex.

[0040] To this end, the present application provides a high-rise structure wind load spectrum identification method based on transfer deep learning to at least partially solve the above-mentioned technical problems.

[0041] For ease of description and to make the technical solutions of the specific embodiments of the present disclosure easier to understand, before describing the method for identifying wind load spectra of high-rise structures based on transfer deep learning implemented in the present disclosure, the physical mechanisms involved in the specific embodiments of the present disclosure are explained as follows:

[0042] For a high-rise building, wind load acts on it in the form of distributed load. When the height-to-width ratio of the structure is large, the influence of the width direction can be ignored, and it can be regarded as a one-dimensional distributed dynamic load along the height direction. Considering the complexity of the actual structure, the finite element method is used to N The high-rise building structure is discretized along the height; the wind load is also discretized. The number of discrete wind loads along the height direction is N .

[0043] When the wind load is a random load, the relationship between the wind load power spectrum and the structural response power spectrum can be described in the frequency domain:

[0044] (1)

[0045] in, ω is the circular frequency after structural frequency conversion, The power spectrum matrix representing the observed response of the structure; The power spectrum matrix representing the wind load, is the structural frequency response function, represents the conjugate transpose of the structural frequency response function, A is the observation matrix, B is the wind load influence matrix, with the superscript T Indicates transpose.

[0046] In one example, the structural frequency response function includes a frequency response function of the structural displacement response and a frequency response function of the structural acceleration response. It should be noted that the above structural frequency response function can be determined based on the dynamic response time-history data of the selected structure. For example, if the dynamic response time-history data of the selected structure is displacement response, the corresponding structural frequency response function is the frequency response function of the structural displacement response; alternatively, if the dynamic response time-history data of the selected structure is acceleration response, the corresponding structural frequency response function is the frequency response function of the structural acceleration response.

[0047] The frequency response function of the structural displacement response is as follows:

[0048] ,in, ω is the circular frequency after structural frequency conversion; M is the mass matrix of the structure; Q is the damping matrix of the structure; i is the imaginary unit, the value is -1; K is the stiffness matrix of the structure.

[0049] The frequency response function of the structural acceleration response is shown below:

[0050] .

[0051] Then, the relationship between wind load and wind speed in the time domain along the height direction is as follows:

[0052] (2)

[0053] in, fi ( t ) indicates that the structure is t Moment i The wind load value of discrete points along the height direction, represents the air density, Indicates the i The height of the point, for The aerodynamic coefficient, for The wind action area; represent i The wind speed at the point height.

[0054] wind speed By average wind speed and pulsating wind speed composition:

[0055] , (3)

[0056] in, It represents the average wind speed at 10m. Indicates a height of 10m. is a constant between 0.15 and 0.5. Substituting formula (3) into formula (2) yields:

[0057] (4)

[0058] It should be noted that this application only considers the fluctuating wind load and assumes that the turbulence is small, that is, the quadratic term of the fluctuating wind is ignored. Combined with formula (3), formula (4) can be re-expressed as:

[0059] (5)

[0060] in .

[0061] The relationship between the fluctuating wind load power spectrum and the wind speed power spectrum is expressed as:

[0062] (6)

[0063] in, C is the relationship matrix between the fluctuating wind load spectrum and the fluctuating wind speed spectrum, which is a positive definite matrix; The power spectrum matrix representing the fluctuating wind speed can be specifically expressed as:

[0064] (7)

[0065] Among them, the diagonal elements are the self-spectral elements of the wind speed spectrum, and the rest are the cross-spectral elements of the wind speed spectrum at two different heights.

[0066] Substituting formula (6) into formula (1) yields:

[0067] (8)

[0068] This establishes the relationship between the observed response spectrum of the structure and the wind load spectrum.

[0069] Different treatment methods are used for steady uniform wind load and non-uniform wind load, as described below:

[0070] 1. Identification of uniform wind load spectrum:

[0071] Since the uniform pulsating wind speed spectrum is only related to the height difference between two points, , the wind speed spectrum matrix only has N There are independent self-spectrum and cross-spectrum elements, so the independent elements in the fluctuating wind speed spectrum matrix can be extracted to form a uniform fluctuating wind speed spectrum vector :

[0072] (9)

[0073] The present application embodiment uses Shitoni's coherence function to illustrate:

[0074] (10)

[0075] in, and Indicates the i The height of point j and point j. This coherence function is only related to the height difference. L z is a constant value of 60, and exp(•) is an exponential function with e as the base.

[0076] A notable feature of discrete cosine transform (DCT) is that it can concentrate most of the signal's energy on a few transform coefficients. By converting the signal into the frequency domain, it can effectively reflect this energy concentration.

[0077] Then the DCT decomposition process of the uniformly fluctuating wind speed spectrum vector element is:

[0078] (11)

[0079] in, n Indicates the sequence number of the fluctuating wind speed spectrum vector element, N indivual; k is the DCT order, Indicates the k DCT coefficients of order; is the Kronecker delta function.

[0080] The reconstruction of the uniform fluctuating wind speed spectrum vector elements can be expressed as follows:

[0081] (12)

[0082] In short:

[0083] (13)

[0084] Then the power spectrum matrix of fluctuating wind speed (Formula (7)) can be expressed as:

[0085] (14)

[0086] According to the characteristics of DCT energy concentration, this application only considers the front p DCT coefficients can be reconstructed to obtain the approximate value of the original wind load spectrum signal. The truncated DCT coefficient vector is From formulas (14) and (8), we can get:

[0087] (15)

[0088] in, Represents the basis function matrix after truncation, that is, p A matrix of cosine basis function vectors.

[0089] Thus, the mapping relationship between the observed response power spectrum matrix of the structure and the DCT coefficients after truncated decomposition of the pulsating steady uniform wind speed spectrum was found.

[0090] 2. Identification of non-uniform wind load spectrum

[0091] The embodiment of the present application adopts Davenport's exponential coherence function:

[0092] (16)

[0093] (17)

[0094] in, and Respectively represent i Point and j The average wind speed at the point, is the attenuation coefficient, and this coherence function is frequency dependent.

[0095] The two-dimensional DCT decomposition of the non-uniform fluctuating wind speed spectrum matrix elements is as follows:

[0096] (18)

[0097] in, k 、 l Respectively represent the order of rows and columns of the DCT coefficient matrix, and Same expression .

[0098] The corresponding reconstruction formula is as follows:

[0099] (19)

[0100] It can be expressed in matrix form:

[0101] (20)

[0102] Similarly, the complete wind load spectrum can be reconstructed using the partially concentrated front DCT coefficients, which can be expressed as:

[0103] (twenty one)

[0104] in, Before use The fluctuating wind speed spectrum matrix reconstructed by the coefficients is express Before the rows and columns of the matrix coefficients, Represents an orthogonal matrix D Before q OK, Represents an orthogonal matrix D Before p OK.

[0105] Substituting formula (21) into formula (7) yields:

[0106] (twenty two)

[0107] Because the observation response power spectrum matrix and the DCT coefficient matrix are both symmetric matrices, when the number of independent elements in the observation response power spectrum matrix is greater than the number of independent DCT coefficients, that is, , the coefficient matrix can be solved.

[0108] Thus, the mapping relationship between the response power spectrum of the structure and the DCT coefficients after truncated decomposition of the pulsating non-stationary uniform wind speed spectrum was found.

[0109] Based on the above description, Figure 1 A flow chart of a method for identifying wind load spectra of high-rise structures based on transfer deep learning according to an embodiment of the present application is shown.

[0110] It should be noted that the method can be applied to a terminal device or a server, wherein the terminal device may include but is not limited to one or more of a smart phone, a tablet computer, a portable computer, and a desktop computer; the server may be a physical server or a cloud server.

[0111] The following is an example of how this method is applied to a terminal device. Figure 1 As shown, the high-rise structure wind load spectrum identification method based on transfer deep learning includes at least steps S110 to S140, which are described in detail as follows.

[0112] In step S110 , calculation is performed based on the dynamic response time history data of the target structure under the target wind load to obtain a corresponding observed response power spectrum matrix.

[0113] The dynamic response time history data may be parameter information about the structure changing over time under wind loads, including but not limited to acceleration, velocity, and displacement responses. In one example, the dynamic response time history data may be obtained through sensor measurement or finite element simulation, such as sampling an acceleration signal 100 times per second.

[0114] The observed response power spectrum matrix can be a matrix that characterizes the energy distribution of the structural dynamic response in the frequency domain. Its diagonal elements are the power spectrum density of each degree of freedom response, and the non-diagonal elements are the cross-power spectrum density between different degrees of freedom, reflecting the correlation of the response in the frequency domain.

[0115] In this embodiment, the terminal can obtain the dynamic response time history data of the target structure under the target wind load through sensor measurement or finite element simulation, and perform calculations based on this to obtain the corresponding observation response power spectrum matrix.

[0116] In one example, after acquiring dynamic response time-course data, the terminal may preprocess it. This preprocessing may include, but is not limited to, denoising and filtering, segmentation, and windowing. Specifically, the terminal may low-pass filter the raw dynamic response time-course data to remove high-frequency noise and interference signals. The long-term data may then be segmented into several subsegments, for example, each with 1024 sampling points, and a Hanning window applied to each segment to reduce spectral leakage.

[0117] The terminal then performs a fast Fourier transform on each segment of time-course data, converting the time-domain signal into a complex frequency-domain spectrum. The results from multiple subsegments are then averaged to reduce the impact of random noise. After this processing, the terminal calculates the auto-power spectral density (APD) based on the frequency-domain signal for each degree of freedom, as well as the cross-power spectral density (CPD) of the frequency-domain signals for different degrees of freedom. The APD and the off-diagonal CPD are then arranged according to the degrees of freedom to form a symmetrical observation response power spectrum matrix.

[0118] In step S120, discrete cosine transform (DCT) decomposition is performed on the wind speed spectrum of the target wind load to obtain DCT coefficients after dimensionality reduction.

[0119] The wind speed spectrum is a statistical property that describes how wind speed varies with frequency, representing the distribution of the fluctuating energy of wind speed at different frequencies. It is obtained by converting the time history of wind speed into the frequency domain through Fourier transform and is commonly used to analyze wind turbulence.

[0120] Discrete cosine transform (DCT) is a signal processing technology that achieves energy concentration by decomposing the signal into cosine function components of different frequencies. It is suitable for compression and feature extraction of high-dimensional data.

[0121] The DCT coefficient is the coefficient obtained after DCT transformation, which is used to represent the projection weight of the original signal on the cosine basis functions of different frequencies. The front coefficient usually contains the main energy of the signal.

[0122] In this embodiment, the terminal can obtain the wind speed spectrum (vector or matrix) corresponding to the target wind load and then decompose it through discrete cosine transform to obtain reduced-dimensional DCT coefficients. In this way, the DCT transform compresses the high-dimensional wind speed spectrum of the wind load into low-dimensional coefficients, significantly reducing data complexity while retaining core physical information, providing efficient and noise-resistant input features for machine learning models.

[0123] In some embodiments of the present application, performing discrete cosine transform decomposition on the wind speed spectrum of the target wind load to obtain a dimensionally reduced DCT coefficient includes:

[0124] For the wind speed spectrum vector of uniform wind load, one-dimensional DCT coefficient is used to decompose it, and the front p main coefficients;

[0125] For the wind speed spectrum matrix of non-uniform wind load, the two-dimensional DCT coefficient is used to decompose it and the front q × p principal coefficients, among which q and p are the truncation orders for rows and columns, respectively.

[0126] In this embodiment, for the wind speed spectrum of uniform wind load, since the elements in its vector are only related to the height difference of the position, it can be decomposed using a one-dimensional DCT coefficient and the front is truncated and retained. p For the wind speed spectrum matrix of non-uniform wind load, since the elements in the matrix are related to both frequency and spatial position, it can be decomposed using two-dimensional DCT coefficients, and the front q × p principal coefficients, among which q and p It should be understood that the energy concentration characteristic of DCT coefficients makes the main information concentrated in the front coefficients, and the noise is dispersed in the high-frequency coefficients. Therefore, after truncating them, the influence of noise can be effectively suppressed.

[0127] Please continue to refer to Figure 1In step S130, a deep learning model for wind load identification based on a physical guidance mechanism is constructed, the observed response power spectrum matrix is used as the input of the deep learning model, and the DCT coefficients are used as the output. The deep learning model is trained to learn the mapping relationship between the two. The physical guidance mechanism embeds the frequency domain transfer equation of wind load and structural response into the loss function to constrain the model training process.

[0128] Among them, the physical guidance mechanism is to embed physical laws (such as frequency domain transfer equations) into the machine learning model training process, and enhance its physical rationality and generalization ability by constraining the model output.

[0129] It should be noted that the frequency domain transfer equation is used to describe the relationship between the structural response power spectrum and the wind load power spectrum.

[0130] The deep learning model may be a model pre-built by a person skilled in the art, which is used to output corresponding DCT coefficients based on the input observation response power spectrum matrix, thereby realizing wind load identification.

[0131] In this embodiment, the terminal can train a pre-built deep learning model for wind load identification based on the observed response power spectrum matrix and its corresponding DCT coefficients obtained in the aforementioned steps. Specifically, the observed response power spectrum matrix can be used as input to the deep learning model, which is trained to output corresponding DCT coefficients. The parameters of the deep learning model can be optimized by minimizing the loss function, thereby improving the accuracy of the deep learning model's output results.

[0132] In one example, the loss function of a deep learning model may include a data error term (such as the difference between the predicted DCT coefficients and the true values) and a physical constraint term (such as the residual of the frequency domain transfer equation).

[0133] By forcing the model's output to conform to the frequency-domain transfer equation through physical constraints, the accuracy of the model's output can be effectively improved. Furthermore, the physical guidance mechanism can reduce the deep learning model's reliance on large-scale data, reducing the amount of training data required while ensuring output accuracy.

[0134] Please continue to refer to Figure 1 In step S140, when performing wind load identification, the trained deep learning model is used to output corresponding DCT coefficients based on the input observation response power spectrum matrix, and the power spectrum matrix of the wind load is reconstructed according to the DCT coefficients.

[0135] In this embodiment, after the deep learning model is trained, the terminal can use it to identify wind loads. Specifically, the terminal can input the observed response power spectrum matrix obtained through actual measurement or simulation into the deep learning model, which can then perform inference based on the observed response power spectrum matrix and output corresponding DCT coefficients. The terminal can then reconstruct the output DCT coefficients to obtain the corresponding wind load power spectrum matrix, thereby achieving wind load identification.

[0136] In some embodiments of the present application, reconstructing the power spectrum matrix of the wind load according to the DCT coefficients includes:

[0137] The DCT coefficients output by the deep learning model are substituted into the DCT inverse transform formula to restore the complete power spectrum matrix of uniform or non-uniform wind loads.

[0138] In this embodiment, the terminal can substitute the DCT coefficients output by the deep learning model into the corresponding DCT inverse transform formula, thereby restoring it to the original signal or matrix through inverse operation, and then calculating the corresponding power spectrum matrix of the wind load.

[0139] In this way, the trained model achieves end-to-end recognition from the response power spectrum to the wind load spectrum. Combined with inverse DCT transform and error control, it significantly improves computational efficiency while ensuring high accuracy.

[0140] In some embodiments of the present application, after completing the training of the deep learning model, the method further includes:

[0141] Based on the transfer learning strategy, for wind loads at discrete points of different frequencies, the model of the target frequency is initialized using the model parameters trained based on the wind load of the source frequency, and the model parameters are fine-tuned using the target frequency data to obtain a deep learning model corresponding to the wind load of the target frequency.

[0142] In this embodiment, the transfer learning strategy can be to transfer model parameters obtained from training in a source domain (e.g., wind loads at a certain frequency) to a target domain (e.g., wind loads at another frequency), leveraging existing knowledge to accelerate the learning process for the new task. Specifically, the terminal can fully train the deep learning model based on a training dataset obtained based on the wind loads at the source frequency. It should be understood that this training dataset includes several sets of observed response power spectrum matrices and their corresponding DCT coefficients.

[0143] When it is necessary to identify wind loads of a target frequency (different from the source frequency), the terminal can reuse the parameters of the model trained for the source frequency wind load and fine-tune the model parameters based on a small amount of training data sets corresponding to the target frequency wind load, thereby obtaining a deep learning model corresponding to the target frequency wind load.

[0144] Thus, the present embodiment utilizes a transfer learning strategy, reuses the source frequency model's physical guidance feature extraction capabilities, and fine-tunes the target frequency with a small amount of data, achieving efficient and high-precision identification of cross-frequency wind loads. This method significantly reduces data requirements and computational costs, providing a scalable technical path for full-band load identification in complex wind environments.

[0145] Based on the technical solutions of the above embodiments, a specific application scenario of the embodiments of the present application is introduced below:

[0146] Figure 2 A schematic flow chart of a method for identifying wind load spectra of high-rise structures based on transfer deep learning according to another embodiment of the present application is shown.

[0147] The following is an explanation of the technical terms in the accompanying drawings:

[0148] Wind load spectra: It represents the distribution of the force exerted by wind on the structure in the frequency domain and describes how the pulsating energy of wind load changes with frequency.

[0149] Wind velocity spectra: describes the pulsation characteristics of wind speed, that is, the distribution of wind speed pulsation energy at different frequencies.

[0150] Wind load: The pressure or suction exerted by wind on a building or structure.

[0151] Finite Element Method (FEM): The finite element method is used to simulate the response of the structure under wind load and obtain the structural response time history data.

[0152] Structural response: The time-history data of the structure’s response to wind loads, such as displacement and acceleration, obtained through finite element method simulation.

[0153] Discrete cosine transform (DCT): Decomposes the wind speed spectrum or wind load spectrum using the DCT method to extract independent DCT coefficients. DCT concentrates most of the signal's energy into a few coefficients, thereby achieving dimensionality reduction.

[0154] Observed response spectra: The frequency domain representation of the response data of a structure under wind load, obtained through actual measurement or simulation. These response data usually include the acceleration, velocity, displacement, etc. of the structure.

[0155] Training / Testing: Train a physics-guided convolutional neural network (CNN) using the observed response spectrum as input and the DCT coefficients as output. A physical loss function is embedded in the training process to reduce the number of training sets and improve training efficiency.

[0156] Data Loss: It represents the difference between the predicted wind load spectrum and the true value, and is used to evaluate the prediction accuracy of the model.

[0157] Physically guided CNN: A CNN network embedded in a physical model uses a physical loss function to guide network training, improving the accuracy and generalization ability of the model.

[0158] Physical Loss: Embedding physical models (such as the relationship between wind load and structural response) into the loss function allows the network to consider physical laws during training, improving the stability and accuracy of the model.

[0159] Uniform spectrum: The wind speed spectrum or wind load spectrum has a uniform distribution characteristic in space.

[0160] Non-uniform spectrum: The wind speed spectrum or wind load spectrum is not uniformly distributed in space, that is, the statistical characteristics of wind speed or wind load vary with different locations.

[0161] In one example, for a uniform spectrum, the corresponding physical loss is:

[0162] .

[0163] in, β is the weight coefficient; Indicates that the structure has a frequency of ω The observed response power spectrum matrix at time ; W ( ω ) coefficient vector at the corresponding frequency, is a generalized orthogonal basis function;

[0164] For non-uniform spectrum, the corresponding Physical Loss is:

[0165] .

[0166] It should be understood that at this time In matrix form, 、 They represent the front of the orthogonal basis function matrix p Line and front q OK.

[0167] Transfer Learning: Utilizes a model trained at one frequency and transfers it to other frequencies for identification. With transfer learning, only a small amount of training data is needed at other frequencies to complete wind load spectrum identification across all frequencies, significantly improving training efficiency.

[0168] like Figure 2 As shown in Figure 1, this method for identifying wind load spectra for high-rise structures based on transfer deep learning begins by simulating the structural response under wind loads using the finite element method (FEM) based on the wind velocity spectrum to obtain time-history data. Frequency domain analysis is then performed on the structural response data to obtain the observed response spectrum. Simultaneously, a discrete cosine transform (DCT) is performed on the wind velocity spectrum to extract DCT coefficients for dimensionality reduction.

[0169] The observed response spectra are then used as input and the DCT coefficients as output for training using a physics-guided convolutional neural network (CNN). During training, a physical loss function is introduced, which is based on the physical relationship between wind loads and structural responses. This reduces the number of required training sets and improves the stability and accuracy of the model.

[0170] Once training is complete, new observed response spectrum data can be fed into the trained network, which will then output the corresponding DCT coefficients. By reconstructing these coefficients, an estimate of the wind load spectrum can be obtained, enabling wind load identification.

[0171] Taking a height of 10m as an example, by changing the wind speed at different 10m points, a pulsating wind speed spectrum at a certain frequency is generated, and then the observed response spectrum of the structure is calculated. The corresponding fluctuating wind speed spectrum vector (matrix) is decomposed and truncated using the one-dimensional (two-dimensional) DCT to obtain the DCT coefficient vector (matrix). Then, given a generalized cosine basis function matrix, a physics-guided neural network for uniform (non-uniform) random wind speed spectrum recognition is trained using an embedded physics loss function. The network input and output are the observed structural response spectrum at a certain frequency from the previous step and the DCT coefficient vector (matrix), respectively.

[0172] Next, the structural response time history under wind loads generated by other forms of fluctuating wind speed spectra at a specific wind speed of 10 m is observed, and the corresponding observed response power spectrum at that frequency is calculated to form a test set. The calculated observed structural response power spectrum at that frequency is input into the trained network for testing, causing it to output a DCT coefficient vector (matrix). The uniform (or non-uniform) fluctuating wind speed spectrum at that frequency is then reconstructed based on this DCT coefficient vector (matrix).

[0173] When it is necessary to expand to other frequencies, the transfer learning method is used, the parameters trained by the above process network are used as the initial values, small sample data sets are made at other frequencies and the above steps are repeated until all frequency discrete points are traversed, and the random stationary uniform (non-uniform) wind speed spectrum recognition of the entire frequency band is completed.

[0174] As in Figure 2 As shown in the lower half of the figure, after fully training a model at a specific frequency, it can be transferred to other frequencies for identification. Only a small training set is required for fine-tuning at these other frequencies to complete wind load spectrum identification across the entire frequency range, significantly improving training efficiency and saving significant time and computing resources. Specifically, after fully training a model at a specific frequency (such as ω1), this model can be transferred to other frequencies (such as ω2, ω3, etc.), requiring only a small training set for fine-tuning at these frequencies. Thus, through transfer learning, wind load spectrum identification can be efficiently completed across the entire frequency range, saving significant training time and computing resources.

[0175] To verify the effectiveness and accuracy of the methods provided in the aforementioned embodiments, a 76-story benchmark model was used in the calculations. The 76-story wind-induced vibration benchmark building established by the International Association for Structural Control and the American Society of Civil Engineers (IASC-ASCE) is not described in detail here.

[0176] For the recognition network of uniform wind speed spectrum, it is trained with Davenport spectrum and tested with Harris spectrum. There are 5 test sets, and the recognition accuracy index uses L2 norm. First, the network is trained with 1Hz training set (recognition results are shown in Figure 2). Figure 3 As shown in the figure, Identification Value is the estimated value of the wind load spectrum, and True Value is the true value of the wind load spectrum. The horizontal axis in the figure represents the self-spectrum or cross-spectrum between any two points on the structure. For example, S 1−1 Represents the autospectrum between the first point and the first point on the structure, S 1−10 Represents the cross spectrum between the first point and the tenth point on the structure, and so on. The vertical axis is the amplitude of the wind load spectrum, in m 2 / Hz, reflecting the energy distribution of wind load at the corresponding frequency), and then migrated to 4Hz (the identification results are as follows Figure 4 The average recognition accuracy is above 90%. The specific recognition accuracy of each test set is shown in Table 1 below.

[0177] Table 1: Identification accuracy of stable and uniform wind speed spectrum at different frequencies

[0178]

[0179] To further verify the superiority of physical guidance, that is, embedding the objective function into the loss function, two sets of training sets can be used for training respectively. The number of samples in training set 1 is 20,000, and the physical loss function is not embedded; the number of samples in training set 2 is 5,000, and the physical loss function is embedded. The most intuitive difference lies in the rate of decrease of the loss function and the training time. Among them, the 100 rounds of training time for training set 1 is 737.91 seconds, and the 100 rounds of training time for training set 2 is 226.13 seconds. The trend of the loss function change during the training process is as follows Figure 5 (Training dataset 1 is training set 1, and Training dataset 2 is training set 2. The horizontal axis is the number of training epochs, and the vertical axis is the value of the loss function.) The blue dotted line is the loss function curve for training set 1, and the red solid line is the loss function curve for training set 2. It can be seen that the network loss curve with the loss function embedded in it decreases significantly, and the correct gradient can be found more quickly. In addition, the loss function value is smaller in the first 100 rounds of training.

[0180] For the recognition network for non-uniform wind speed spectrum, Von-Karman spectrum is used for training and Kaimal spectrum is used for testing. The recognition results of non-uniform wind speed spectrum at different frequencies are as follows: Figure 6 and Figure 7 As shown, Figure 6 is the recognition result at a frequency of 1 Hz (where Identification is the recognition value, True Value is the true value, and Relative Error is the relative error). Figure 7 This is the recognition result when the frequency is 4 Hz. It can be seen that the error between the recognition value and the true value is small.

[0181] To further verify the advantages of transfer learning in the application of this method, Table 2 compares the original pre-training time with the transfer learning time of other frequency discrete points. The pre-training set uses 5000 groups of samples, the number of samples in the transfer learning training set is 500, and the number of training rounds is 1000. It can be seen that after the pre-training process is completed, the time required for transfer training is greatly reduced, the training efficiency is greatly improved, and a lot of computing resources and time are saved.

[0182] Table 2: Transfer learning time of stationary non-uniform wind speed spectrum at different frequencies

[0183]

[0184] Based on any of the above embodiments, the present disclosure also provides a high-rise structure wind load spectrum identification device based on transfer deep learning.

[0185] Figure 8 It is a structural schematic block diagram of a high-rise structure wind load spectrum identification device based on transfer deep learning according to an embodiment of the present invention.

[0186] like Figure 8 As shown, the high-rise structure wind load spectrum identification device based on transfer deep learning includes:

[0187] A calculation module is used to calculate the dynamic response time history data of the target structure under the target wind load to obtain the corresponding observation response power spectrum matrix;

[0188] a decomposition module, configured to perform discrete cosine transform decomposition on the wind speed spectrum of the target wind load to obtain DCT coefficients after dimensionality reduction;

[0189] A training module is configured to construct a deep learning model for wind load identification based on a physical guidance mechanism, using the observed response power spectrum matrix as input to the deep learning model and the DCT coefficients as output, and enabling the deep learning model to learn the mapping relationship between the two through training. The physical guidance mechanism embeds a frequency domain transfer equation between wind load and structural response into a loss function to constrain the model training process.

[0190] The processing module is used to use the trained deep learning model to output corresponding DCT coefficients based on the input observation response power spectrum matrix when performing wind load identification, and reconstruct the power spectrum matrix of the wind load according to the DCT coefficients.

[0191] It should be noted that the above-mentioned high-rise structure wind load spectrum identification device based on transfer deep learning can be in the form of computer software, and each module of the above-mentioned high-rise structure wind load spectrum identification device based on transfer deep learning can be implemented by computer software modules.

[0192] In some embodiments of the present disclosure, performing discrete cosine transform decomposition on the wind speed spectrum of the target wind load to obtain a dimensionally reduced DCT coefficient includes:

[0193] For the wind speed spectrum vector of uniform wind load, one-dimensional DCT is used to decompose it and the front p main coefficients;

[0194] For the wind speed spectrum matrix of non-uniform wind load, two-dimensional DCT is used to decompose it and the front q×p principal coefficients, among which q and p are the truncation orders for rows and columns, respectively.

[0195] In some embodiments of the present disclosure, after the deep learning model is trained, the processing module is further configured to:

[0196] Based on the transfer learning strategy, for wind loads at discrete points of different frequencies, the model of the target frequency is initialized using the model parameters trained based on the wind load of the source frequency, and the model parameters are fine-tuned using the target frequency data to obtain a deep learning model corresponding to the wind load of the target frequency.

[0197] In some embodiments of the present disclosure, the frequency domain transfer equation is as follows:

[0198]

[0199] in, is the observed response power spectrum matrix of the structure; The power spectrum matrix representing the wind load, is the frequency response function of the structure, represents the conjugate transpose of the structure's frequency response function, A is the observation matrix, B is the wind load influence matrix, with the superscript T Indicates transpose.

[0200] In some embodiments of the present disclosure, reconstructing the power spectrum matrix of the wind load according to the DCT coefficients includes:

[0201] The DCT coefficients output by the deep learning model are substituted into the DCT inverse transform formula to restore the complete power spectrum matrix of uniform or non-uniform wind loads.

[0202] In some embodiments of the present disclosure, the dynamic response time history data includes acceleration, velocity or displacement response.

[0203] The implementation process of the functions and effects of each module in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0204] Figure 9 A schematic diagram of the structure of a computer system suitable for implementing an electronic device according to an embodiment of the present application is shown.

[0205] It should be noted that Figure 9 The computer system of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0206] like Figure 9As shown, the computer system includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes, such as the methods described in the above embodiments, based on programs stored in a read-only memory (ROM) 302 or programs loaded from a storage unit 308 into a random access memory (RAM) 303. RAM 303 also stores various programs and data required for system operation. CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to bus 304.

[0207] The following components are connected to the I / O interface 305: an input section 306 including a keyboard, mouse, and the like; an output section 307 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and speakers; a storage section 308 including devices such as a hard disk; and a communication section 309 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the I / O interface 305 as needed. Removable media 311, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 310 as needed, so that computer programs read from the media can be installed in the storage section 308 as needed.

[0208] In particular, according to embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for executing the methods illustrated in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via the communication section 309 and / or installed from removable media 311. When executed by the central processing unit (CPU) 301, the computer program performs the various functions defined in the system of the present application.

[0209] It should be noted that the computer-readable medium described in the embodiments of the present application may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present application, a computer-readable signal medium may include a data signal transmitted in baseband or as part of a carrier wave, which carries a computer-readable computer program. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. A computer program embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0210] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. Among them, each box in the flowchart or block diagram can represent a module, program segment, or part of the code, and the above-mentioned module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0211] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. In some cases, the names of these units do not constitute limitations on the units themselves.

[0212] As another aspect, the present application further provides a computer-readable medium, which may be included in the electronic device described in the above embodiments, or may exist independently without being incorporated into the electronic device. The computer-readable medium carries one or more programs, and when the one or more programs are executed by the electronic device, the electronic device implements the method described in the above embodiments.

[0213] The present disclosure also provides a computer program product. The method of the present disclosure can be implemented in whole or in part using software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed, the process or function of the present disclosure is performed in whole or in part.

[0214] A computer program or instruction can be stored in a readable storage medium or transferred from one readable storage medium to another. For example, the computer program or instruction can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The readable storage medium can be any accessible medium or a data storage device such as a server or data center that integrates one or more accessible media. The accessible medium can be a magnetic medium such as a floppy disk, hard disk, or magnetic tape; an optical medium such as a digital video disk; or a semiconductor medium such as a solid-state drive. The computer-readable storage medium can be a volatile or non-volatile storage medium, or can include both volatile and non-volatile types of storage media.

[0215] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0216] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0217] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0218] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0219] In the description of this specification, the description with reference to the terms "one embodiment / method", "some embodiments / methods", "example", "specific example", or "some examples" means that the specific features, structures, or characteristics described in conjunction with the embodiment / method or example are included in at least one embodiment / method or example of the present disclosure. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment / method or example. Moreover, the specific features, structures, or characteristics described may be combined in a suitable manner in any one or more embodiments / methods or examples. In addition, those skilled in the art may combine and combine different embodiments / methods or examples described in this specification and the features of different embodiments / methods or examples, unless they are contradictory.

[0220] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. Throughout the present disclosure, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0221] Those skilled in the art will appreciate that the above embodiments are merely intended to clearly illustrate the present disclosure and are not intended to limit the scope of the present disclosure. Other changes or modifications may be made based on the above disclosure, and such changes or modifications are still within the scope of the present disclosure.

Claims

1. A method for identifying wind load spectra of high-rise structures based on transfer deep learning, characterized in that: include: Calculate the dynamic response time history data of the target structure under the target wind load and obtain the corresponding observation response power spectrum matrix; Performing discrete cosine transform decomposition on the wind speed spectrum of the target wind load to obtain DCT coefficients after dimensionality reduction; A deep learning model for wind load identification based on a physical guidance mechanism is constructed, the observed response power spectrum matrix is used as the input of the deep learning model, and the DCT coefficients are used as the output. The deep learning model is trained to learn the mapping relationship between the two. The physical guidance mechanism embeds the frequency domain transfer equation of wind load and structural response into the loss function to constrain the model training process. When performing wind load identification, the trained deep learning model is used to output corresponding DCT coefficients based on the input observation response power spectrum matrix, and the power spectrum matrix of the wind load is reconstructed according to the DCT coefficients; After the deep learning model is trained, the method further includes: Based on the transfer learning strategy, for wind loads at discrete points of different frequencies, the model of the target frequency is initialized using the model parameters trained based on the wind load of the source frequency, and the model parameters are fine-tuned using the target frequency data to obtain a deep learning model corresponding to the wind load of the target frequency.

2. The method according to claim 1, characterized in that The wind speed spectrum of the target wind load is subjected to discrete cosine transform decomposition to obtain DCT coefficients after dimensionality reduction, including: For the wind speed spectrum vector of uniform wind load, one-dimensional DCT is used to decompose it and the front p main coefficients; For the wind speed spectrum matrix of non-uniform wind load, two-dimensional DCT is used to decompose it and the front q×p principal coefficients, among which q and p are the truncation orders for rows and columns, respectively.

3. The method according to claim 1, characterized in that The frequency domain transfer equation is shown as follows: in, is the observed response power spectrum matrix of the structure; The power spectrum matrix representing the wind load, is the frequency response function of the structure, represents the conjugate transpose of the structure's frequency response function, A is the observation matrix, B is the wind load influence matrix, with the superscript T Indicates transpose.

4. The method according to claim 1, wherein Reconstructing the power spectrum matrix of the wind load according to the DCT coefficients includes: The DCT coefficients output by the deep learning model are substituted into the DCT inverse transform formula to restore the complete power spectrum matrix of uniform or non-uniform wind loads.

5. The method according to claim 1, wherein The dynamic response time history data includes acceleration, velocity or displacement response.

6. A high-rise structure wind load spectrum identification device based on transfer deep learning, characterized in that: include: A calculation module is used to calculate the dynamic response time history data of the target structure under the target wind load to obtain the corresponding observation response power spectrum matrix; a decomposition module, configured to perform discrete cosine transform decomposition on the wind speed spectrum of the target wind load to obtain DCT coefficients after dimensionality reduction; A training module is configured to construct a deep learning model for wind load identification based on a physical guidance mechanism, using the observed response power spectrum matrix as input to the deep learning model and the DCT coefficients as output, and enabling the deep learning model to learn the mapping relationship between the two through training. The physical guidance mechanism embeds a frequency domain transfer equation between wind load and structural response into a loss function to constrain the model training process. A processing module is used to use the trained deep learning model to output corresponding DCT coefficients based on the input observation response power spectrum matrix when performing wind load identification, and reconstruct the power spectrum matrix of the wind load based on the DCT coefficients; After the deep learning model is trained, the processing module is further configured to: Based on the transfer learning strategy, for wind loads at discrete points of different frequencies, the model of the target frequency is initialized using the model parameters trained based on the wind load of the source frequency, and the model parameters are fine-tuned using the target frequency data to obtain a deep learning model corresponding to the wind load of the target frequency.

7. An electronic device, characterized in that: include: a memory storing execution instructions; as well as A processor, wherein the processor executes the execution instructions stored in the memory, so that the processor executes the method according to any one of claims 1 to 5.

8. A readable storage medium, characterized in that: The readable storage medium stores execution instructions, which are used to implement the method according to any one of claims 1 to 5 when executed by a processor.

9. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images