Steel structure building long-term load metal fatigue prediction method based on machine learning

Abstract

The invention provides a steel structure building long-term load metal fatigue prediction method based on machine learning, and relates to the field of data prediction. The DWN prediction model is applied to a steel structure building long-term load metal fatigue prediction scene and comprises a frequency domain feature extraction module, a time domain feature extraction module and a learning prediction module, specifically, the frequency domain feature extraction module can extract frequency domain information and model frequency domain features, and the time domain feature extraction module can extract time domain information and model time domain features; and the time domain feature modeling and learning prediction module is used for combining the frequency domain features and the time domain features and converting the frequency domain features and the time domain features into an available long-term load metal fatigue prediction result of the steel structure building through comparative learning, and all the modules cooperate with one another to achieve accurate prediction of the long-term load metal fatigue of the steel structure building.

Description

Technical Field

[0001] The present invention belongs to the field of data prediction, and in particular relates to a method for predicting long-term load metal fatigue of steel structure buildings based on machine learning. Background Art

[0002] Steel structures are widely used in modern urban construction and are subjected to long-term loads, including self-weight, service loads, wind, earthquakes and other external forces. The impact of long-term loads on steel structures is gradually accumulated. With the passage of time, the microstructure of steel changes, leading to the emergence of metal fatigue. The fatigue damage of steel structures is manifested as the gradual appearance of cracks in the material after being subjected to repeated loads, which eventually leads to structural failure. Especially under complex loads and non-uniform environments, the prediction of long-term load metal fatigue of steel structures is an important issue in structural health monitoring.

[0003] Machine learning has made significant progress in the field of engineering structure health monitoring in recent years. In terms of steel structure fatigue prediction, machine learning methods can automatically learn and extract potential patterns from large amounts of data. The fatigue damage process of steel structures is highly nonlinear and complex. Traditional analysis methods often cannot fully consider the influence of various loads and environmental factors. Machine learning algorithms can identify key patterns in the fatigue process of steel structures by analyzing large amounts of historical data. They can process and analyze data in real time in a constantly changing actual environment, identify early signs of fatigue damage, and provide predictions for future fatigue development trends.

[0004] At present, the problem faced by fatigue prediction of steel structures is how to build a model that can cope with complex nonlinear relationships and has high prediction accuracy in the presence of noise, sparse data and variable working conditions. In the face of these challenges, multi-scale feature extraction can be used to effectively identify fatigue change patterns at different time scales. By combining time domain and frequency domain features, multi-scale information can be effectively captured when processing long time series data, thereby enhancing the model's robustness to noise and sparse data. Summary of the invention

[0005] The present invention provides a method for predicting long-term load metal fatigue of steel structure buildings based on machine learning. Aiming at long-term load time series data of steel structure buildings with strong noise, nonlinearity and complex dependencies, a DWN (Dynamic WaveNet) model based on machine learning is proposed. The model consists of a frequency domain feature extraction module, a time domain feature extraction module and a learning prediction module. The frequency domain feature extraction module captures frequency domain features through wavelet transform, and introduces a variational local adaptive wavelet transform method to extract the most representative frequency components. The time domain feature extraction module extracts effective time domain features from the long-term load metal fatigue data of steel structure buildings through a designed learnable convolution kernel. The learning prediction module combines the above features, designs a learnable contrast loss weight training model and realizes long-term load metal fatigue prediction.

[0006] The technical solution adopted by the present invention to achieve the above-mentioned purpose specifically includes the following steps: S1. Collect long-term load metal fatigue data of steel structure buildings, including physical parameters and use environment parameters, and pre-process the collected data; S2, the preprocessed metal fatigue data is standardized using the mean normalization method, and mapped through dilated convolution to divide the data into training set and test set; S3, build a frequency feature extraction module, introduce variational local adaptive wavelet transform, and build frequency domain features. The specific steps are as follows: S31, selecting an adaptive scale by dynamically calculating the time series characteristics of the metal fatigue data of the steel structure through variational optimization; S32, calculating the local frequency domain energy of the adaptive scale, and optimizing the adaptive scale by path integration; S33, using the inverse graph Laplace weight calculation to optimize the contribution of different frequency components, and using the inverse wavelet transform to reconstruct the weighted wavelet features of the steel structure building metal fatigue data back to the time domain signal; S4. Build a time domain feature extraction module, design a learnable convolution kernel, and build time domain features through residual connections. The specific steps are as follows: S41. Design of learnable convolution kernels Perform convolution operation on the time series features of metal fatigue data of steel structure buildings to obtain the output after convolution operation; S42. Apply convolution results Activation function and layer normalization; S43, performing residual connection on the normalized output of the layer and the input, concatenating the features of all time steps, and obtaining the final time domain feature representation; S5. Build a learning prediction module, integrate the metal fatigue data characteristics of steel structure buildings in the time domain and frequency domain, and design learnable weights , adjust the contrast loss weight, train the model through contrast learning, and input the processed metal fatigue data into the model to obtain the metal fatigue prediction results.

[0007] Preferably, in said S1, long-term load metal fatigue data of steel structure buildings is collected, including physical parameter data of steel structure materials and use environment parameter data, wherein the physical parameter data includes tensile strength data, compressive strength data, yield strength data, density data, elastic modulus data, fatigue life data and hardness data of steel structure materials, and the installation environment parameter data includes long-term load data, ambient temperature data, ambient humidity data, ambient vibration data and light condition data, and the original sequence is calculated. The mean and standard deviation , the specific formula is: ; ; In the formula, is the total number of time points, for Metal fatigue data of steel structure buildings at all times; Generate enhanced metal fatigue data for steel buildings through scaling and offset operations , the specific formula is: ; In the formula, is the scaling factor, is the offset factor, To control the scaling hyperparameters, A hyperparameter that controls the bias.

[0008] Preferably, in S2, the pre-processed metal fatigue data of the steel structure building is standardized using a mean normalization method, and the specific formula is: ; In the formula, This is the metal fatigue data of steel structure buildings after data enhancement. is the mean of the series, is the standard deviation of the series; Use the dilated convolution module to extract the time series features of the steel structure building metal fatigue data as the input signal , the specific formula is: ; In the formula, is a linear layer, is the dilated convolution, The projection layer.

[0009] Preferably, in S3 and S31, the input signal Dynamically compute adaptive scale selection via variational optimization , the specific formula is: ; In the formula, is the wavelet scale, is the time offset, is the smoothing factor, for Gradients with respect to wavelet scale; is the wavelet transform, the specific formula is: ; In the formula, is about the wavelet scale The Morlet basis function of is: ; In the formula, is the wavelet scale, is the time offset.

[0010] Preferably, a variational optimization algorithm is introduced to dynamically select the optimal adaptive scale according to the time series characteristics of the extracted metal fatigue data of steel structures. The adaptive scale is automatically optimized according to the changes in the scale, the flexibility and accuracy of feature extraction are improved, and the frequency domain characteristics of metal fatigue data of steel structure buildings at different scales are accurately extracted.

[0011] Preferably, the adaptive scale obtained in S3 and S32 Calculate the local frequency domain energy and optimize the adaptive scale through path integration. The specific formula is: ; In the formula, is the optimal path obtained by variational optimization. The specific formula is: ; In the formula, is the local signal energy, and the specific formula is: , In the formula, For adaptive scale Morlet basis functions.

[0012] Preferably, by combining the path integral optimization technology, the weights of different frequency components of metal fatigue data of steel structure buildings can be dynamically adjusted according to the local characteristics of the input. According to the changes in the input data, the optimal path is selected to calculate the energy of each frequency component, thereby improving the accuracy and efficiency of frequency domain feature extraction. Combined with the multi-scale characteristics of wavelet transform, path integral further enhances the model's ability to capture detailed features, so that effective frequency domain features can be extracted under all scale conditions.

[0013] Preferably, in S3 and S33, the inverse graph Laplace weight calculation is used to optimize the contribution of different frequency components. The specific formula is: ; In the formula, is the optimized adaptive scale, is the time offset, The time series characteristics of the input steel structure metal fatigue data, is the Morlet basis function, is the smoothing factor that controls the smoothness of the neighborhood, is the graph Laplace matrix, and the specific formula is: ; In the formula, is the degree matrix, is the adjacency matrix, which represents the similarity between wavelet scales; Degree Matrix The specific formula is: ; In the formula, is the number of nodes, For Node The specific formula is: ; Adjacency Matrix The specific formula is: ; in, is the number of nodes, the adjacency matrix The Line List Representation Node and nodes Is there an edge between them? The specific formula is: ; The wavelet features of weighted steel structure building metal fatigue data are converted into time domain signals using inverse wavelet transform. The specific formula is: ; In the formula, It is the time domain signal after the conversion of the metal fatigue data of the steel structure building. The specific formula is: ; In the formula, is the optimized adaptive scale, is the time offset.

[0014] Preferably, the graph Laplacian matrix and back propagation method are used to eliminate the noise in the frequency domain data of the metal fatigue data of the steel structure building, and to optimize the accuracy of the frequency domain signal during reconstruction, and to perform weighted optimization on different frequency components to more accurately capture the frequency domain information of the signal.

[0015] Preferably, the time series characteristics of metal fatigue data of steel structure buildings in S4 and S41 are At each time step , design learnable convolution kernels For time step Input Perform convolution operation, the specific formula is: ; In the formula, is the output after the convolution operation, is a 1D convolution operation, Designed learnable convolution kernels; The specific formula for the learnable convolution kernel is: ; In the formula, is the local curvature, which indicates the time series characteristics of the metal fatigue data of the steel structure building at the time step The speed of change at is a learnable parameter, is the maximum convolution kernel, is the minimum convolution kernel, is the Sigmoid function, is the rounding function; The specific formula for local curvature is: ; In the formula, is the second-order derivative; Learnable parameters It is obtained through training optimization. The specific formula is: ; ; In the formula, is the cross entropy loss function, is the convolution kernel weight, is the learning rate; The specific formula for updating the convolution kernel weight is: ; In the formula, is the cross entropy loss function, is the learning rate; The specific formula of the cross entropy loss function is: ; In the formula, is the total number of samples of metal fatigue data of steel structure buildings, is the true value, is the predicted value.

[0016] Preferably, by designing a learnable convolution kernel, the model can automatically adjust the convolution kernel according to the characteristics of the metal fatigue data of the steel structure building, thereby enhancing the ability to learn the time series characteristics of the metal fatigue data of the steel structure building in detail, and enabling the model to flexibly adjust the feature extraction process according to the characteristics of each time step, thereby optimizing the expression of time domain features, further improving the ability to recognize detailed changes in steel structure fatigue prediction, and ensuring efficient feature extraction and prediction results.

[0017] Preferably, in S4 and S42, for different convolution results ,application The activation function gets the processed data , the specific formula is: ; In the formula, is the time step The output after dynamic convolution is is the activation function, the specific formula is: ; In the formula, and are predefined hyperparameters, ; right Perform layer normalization, the specific formula is: ; In the formula, is the layer normalization operation, the specific formula is: ; In the formula, For each time step The mean of For each time step The standard deviation of Small constant set to prevent division by zero errors.

[0018] Preferably, the model applies an activation function to each convolution result for nonlinear transformation, and then performs layer normalization on the output result, ensuring that the nonlinear characteristics of the data after passing through the activation function can be effectively normalized and maintaining stability during training. Layer normalization reduces internal covariate shift, improves training stability and convergence speed, and further enhances the model's ability to process features of different time steps, thereby improving prediction accuracy.

[0019] Preferably, in S4 and S43, the output after layer normalization is With input signal Perform residual connection, the specific formula is: ; In the formula, It is the feature representation after adding residual connection; Use feature splicing to merge the features of all time steps to obtain the final time domain feature representation of the metal fatigue data of the steel structure building. The specific formula is: ; In the formula, For splicing operation, is the time step Feature representation after residual connection.

[0020] Preferably, the output after layer normalization is residually connected with the original steel structure building metal fatigue data, and then the features of all time steps are merged through feature splicing to obtain the final time domain feature representation. The introduction of residual connection helps to avoid the gradient vanishing problem and improves the model's learning ability for complex time series data. By splicing the features of multiple time steps together, the model can more comprehensively understand and capture the changing trends in the time series, enhance the prediction ability of long-term load metal fatigue of steel structures, and further improve the prediction accuracy and stability of the model.

[0021] Preferably, in S5, the metal fatigue data features of steel structure buildings extracted from the time domain and the frequency domain are fused to form a unified feature representation. The specific formula is ; In the formula, For splicing operation, is a linear layer, is the time domain feature representation, It is the frequency domain feature representation; Designing learnable weights , adjust the weights of the time domain contrast loss and the frequency domain contrast loss, use contrastive learning to learn the contrast features of the time domain and frequency domain, and minimize the total loss To train the model, the specific formula is: ; In the formula, is the contrast loss in the time domain, is the contrast loss in the frequency domain; Learnable weights The specific formula is: ; In the formula, is the rate of change of time domain loss, is the rate of change of frequency domain loss, It is a similarity measure of time domain and frequency domain features, controlling the weights of time domain and frequency domain. is the learning rate; Change rate of time domain loss The specific formula is: ; Frequency domain loss change rate The specific formula is: ; Similarity measures between time domain and frequency domain features The specific formula is: ; In the formula, is the L2 norm of the time domain feature, is the L2 norm of the frequency domain feature; Contrast loss in the time domain The specific formula is: ; In the formula, is the feature obtained through residual connection, is the final time domain feature representation; Frequency Domain Contrast Loss The specific formula is: ; In the formula, is the dynamic weight, is the amplitude loss, the specific formula is: ; In the formula, is the total number of samples of metal fatigue data of steel structure buildings, is the predicted frequency domain amplitude, is the real frequency domain amplitude; is the phase loss, and the specific formula is: ; In the formula, is the predicted frequency domain phase, is the actual frequency domain phase.

[0022] Preferably, by fusing the metal fatigue data features of steel structure buildings extracted from the time domain and frequency domain to form a unified feature representation, and further optimizing the feature learning in the time domain and frequency domain through contrastive learning, and designing learnable loss weights, the model can effectively contrast and learn the time domain and frequency domain features by minimizing the total loss, thereby improving the quality of feature representation. Contrastive learning not only improves the feature resolution ability, but also enhances the efficiency of the model in processing multi-source data, so that the prediction model can make more accurate predictions in complex and changeable working environments, thereby improving the overall prediction effect.

[0023] In summary, due to the adoption of the technical solution, the beneficial effects of the present invention are as follows: the present invention proposes a DWN prediction model, which is applied to the long-term load metal fatigue prediction scenario of steel structure buildings, including a frequency domain feature extraction module, a time domain feature extraction module and a learning prediction module. Specifically, the frequency domain feature extraction module can extract frequency domain information and model frequency domain features. The time domain feature extraction module can extract time domain information and model time domain features. The learning prediction module is used to combine frequency domain features and time domain features, and convert them into available long-term load metal fatigue prediction results for steel structure buildings through comparative learning. The modules cooperate with each other to achieve accurate prediction of long-term load metal fatigue of steel structure buildings. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a step-by-step diagram for the long-term load metal fatigue prediction method for steel structure buildings.

[0025] Figure 2 This is the structural diagram of the DWN prediction model.

[0026] Figure 3 This is the structure diagram of the frequency domain feature extraction module.

[0027] Figure 4 This is the structural diagram of the time domain feature extraction module.

[0028] Figure 5 Fitting effect diagram for DWN prediction model to realize long-term load metal fatigue prediction of steel structure buildings. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0030] See also Figure 1-Figure 5 The present invention provides a technical solution: a method for predicting long-term load metal fatigue of steel structure buildings based on machine learning, which can extract frequency domain information and model frequency domain features by constructing a frequency domain feature extraction module, and can extract time domain information and model time domain features by constructing a time domain feature extraction module. The learning prediction module is used to combine frequency domain features and time domain features, and convert them into available long-term load metal fatigue prediction results of steel structure buildings through comparative learning. The specific steps are as follows: Figure 1 shown.

[0031] Construct a DWN prediction model, whose structure is as follows Figure 2 As shown, the specific steps are as follows.

[0032] S1. Collect long-term load metal fatigue data of steel structure buildings, including physical parameters and use environment parameters, and pre-process the collected data.

[0033] Furthermore, long-term load metal fatigue data of steel structure buildings are collected, including physical parameter data of steel structure materials and use environment parameter data, where physical parameter data include tensile strength data, compressive strength data, yield strength data, density data, elastic modulus data, fatigue life data and hardness data of steel structure materials, and installation environment parameter data include long-term load data, ambient temperature data, ambient humidity data, ambient vibration data and light condition data. A total of 140 days of long-term load metal fatigue data of steel structure buildings are collected, and the original sequence is calculated. The mean and standard deviation , the specific formula is: ; ; In the formula, is the total number of time points, which is 140. for Metal fatigue data of steel structures at each moment, generated by scaling and offsetting operations to enhance the time series , the specific formula is: ; In the formula, is the scaling factor, is the offset factor, To control the hyperparameters of scaling , the initial value is set to 0.01, The hyperparameters that control the bias, The initial value is set to 0.01.

[0034] S2. The preprocessed steel structure building metal fatigue data is standardized using the mean normalization method, and mapped through dilated convolution, and the data is divided into training set and test set in a ratio of 7:3.

[0035] Furthermore, the pre-processed metal fatigue data of steel structure buildings is standardized using the mean normalization method. The specific formula is: ; In the formula, This is the metal fatigue data of steel structure buildings after data enhancement. is the mean of the series, is the standard deviation of the series; Use the dilated convolution module to extract the time series features of the steel structure building metal fatigue data as the input signal , the specific formula is: ; In the formula, is a linear layer, is the dilated convolution, The projection layer is then used to divide the steel structure building data into training and test sets in proportion.

[0036] S31, input signal Dynamically compute adaptive scale selection via variational optimization .

[0037] Furthermore, in S31, the input signal Dynamically compute adaptive scale selection via variational optimization , the specific formula is: ; In the formula, is the wavelet scale, , the initial value is set to 1, is the time offset, , is the time step, the value is 1, is the smoothing factor to control the scale change, , the initial value is set to 0.001, Extracted time series characteristics of metal fatigue data of steel structure buildings Gradients about scale; is the wavelet transform, the specific formula is: ; In the formula, is about the wavelet scale Morlet basis function, so that the input data at different wavelet scales and time offset The following is decomposed into the following formula: ; In the formula, is the wavelet scale, the smaller Focus on high-frequency features, larger Focus on low-frequency features. is the time offset.

[0038] S32, the adaptive scale obtained Calculate local frequency domain energy , optimizing the adaptive scale through path integral .

[0039] Furthermore, in S32, the adaptive scale obtained Calculate the local frequency domain energy and optimize the adaptive scale through path integration. The specific formula is: ; In the formula, is the optimal path obtained by variational optimization. The specific formula is: ; In the formula, is the local signal energy, and the specific formula is: , In the formula, For adaptive scale Morlet basis functions.

[0040] S33, using the inverse graph Laplacian weight calculation to optimize the contribution of different frequency components , use the inverse wavelet transform to reconstruct the weighted wavelet features back to the time domain signal .

[0041] Furthermore, in S33, the inverse graph Laplace weight calculation is used to optimize the contribution of different frequency components. The specific formula is: ; In the formula, is the optimized adaptive scale, is the time offset, is the input signal, is the Morlet basis function, is the smoothing factor that controls the smoothness of the neighborhood, , The initial value is set to 0.05. is the graph Laplace matrix, and the specific formula is: ; In the formula, is the degree matrix, is the adjacency matrix, which represents the similarity between wavelet scales; Degree Matrix The specific formula is: ; In the formula, is the number of nodes, For Node The specific formula is: ; Adjacency Matrix The specific formula is: ; in, is the number of nodes, the adjacency matrix The Line List Representation Node and nodes Is there an edge between them? The specific formula is: ; The weighted wavelet features of the steel structure metal fatigue data are converted into time domain signals using the inverse wavelet transform. The specific formula is: ; In the formula, is the converted time domain signal of the metal fatigue data of the steel structure building. The specific formula is: ; In the formula, is the optimized adaptive scale, is the time offset, is the Morlet basis function.

[0042] S41. Metal fatigue data for steel structures At each time step , design learnable convolution kernels For time step Input signal Perform convolution operation to obtain the output after convolution operation .

[0043] Furthermore, S41 is used to describe the time series characteristics of metal fatigue data for steel structure buildings. At each time step , using learnable convolution kernels For time step Input Perform convolution operation, the specific formula is: ; In the formula, is the output after the convolution operation, is a 1D convolution operation, is a learnable convolution kernel; The specific formula for the learnable convolution kernel is: ; In the formula, is the local curvature, which indicates the time series characteristics of the metal fatigue data of the steel structure building at the time step The speed of change at is a learnable parameter, is the maximum convolution kernel, set to 9, is the minimum convolution kernel, set to 3, is the Sigmoid function, is the rounding function; The specific formula for local curvature is: ; In the formula, is the second-order derivative; Learnable parameters It is obtained through training optimization. The specific formula is: ; ; In the formula, is the cross entropy loss function, is the convolution kernel weight, is the learning rate, set to 0.01; The specific formula for updating the convolution kernel weight is: ; In the formula, is the cross entropy loss function, is the learning rate, set to 0.01; The specific formula of the cross entropy loss function is: ; In the formula, The total number of samples for the steel structure building metal fatigue data is set to 140. is the true value, is the predicted value.

[0044] S42. For each convolution result ,application The activation function gets the processed data ,right Perform layer normalization to obtain normalized data .

[0045] Furthermore, for different convolution results in S42 ,application The activation function gets the processed data , the specific formula is: ; In the formula, is the time step The output after dynamic convolution is is the activation function, the specific formula is: ; In the formula, and are predefined hyperparameters, , the initial value is set to 0.1, , the initial value is set to 1.6733; right Perform layer normalization, the specific formula is: ; In the formula, is the layer normalization operation, the specific formula is: ; In the formula, For each time step The mean of For each time step The standard deviation of To prevent division by zero errors set a small constant, .

[0046] S43, normalize the output of the layer Time series characteristics of metal fatigue data of steel structure buildings Perform residual connection and use feature concatenation to merge the features of all time steps to obtain the final time domain feature representation .

[0047] Furthermore, the output of the layer normalized in S43 is Time series characteristics of metal fatigue data of steel structure buildings Perform residual connection, the specific formula is: ; In the formula, It is the feature representation after adding residual connection; Use feature concatenation to merge the features of all time steps to obtain the final time domain feature representation. The specific formula is: ; In the formula, For splicing operation, is the time step Feature representation after residual connection.

[0048] S5. Build a learning prediction module to fuse the data features of steel structure metal fatigue data extracted from the time domain and frequency domain to form a unified feature representation , design learnable weights , adjust the weights of the time domain contrast loss and the frequency domain contrast loss, train the model through contrast learning and input the preprocessed data into the model for prediction.

[0049] Furthermore, in S5, the features extracted from the time domain and frequency domain are fused to form a unified feature representation The specific formula is ; In the formula, For splicing operation, is a linear layer, is the time domain feature representation, It is the frequency domain feature representation; Designing learnable weights , adjust the weights of the time domain contrast loss and the frequency domain contrast loss, use contrastive learning to learn the contrast features of the time domain and frequency domain, and minimize the total loss To train the model, the specific formula is: ; In the formula, is the contrast loss in the time domain, is the contrast loss in the frequency domain; Learnable weights The specific formula is: ; In the formula, is the rate of change of time domain loss, is the rate of change of frequency domain loss, It is a similarity measure of time domain and frequency domain features, controlling the weights of time domain and frequency domain. is the learning rate, set to 0.01; Change rate of time domain loss The specific formula is: ; Frequency domain loss change rate The specific formula is: ; Similarity measures between time domain and frequency domain features The specific formula is: ; In the formula, is the L2 norm of the time domain feature, is the L2 norm of the frequency domain feature; Temporal contrast loss The specific formula is: ; In the formula, is the feature obtained through residual connection, is the final time domain feature representation; Frequency Domain Contrast Loss The specific formula is: ; In the formula, is the dynamic weight, is the amplitude loss, the specific formula is: ; In the formula, is the total number of samples of the metal fatigue data of steel structure buildings, which is 140. is the predicted frequency domain amplitude, is the real frequency domain amplitude; is the phase loss, and the specific formula is: ; In the formula, is the predicted frequency domain phase, is the actual frequency domain phase.

[0050] Furthermore, the DWN prediction model is written in Python, the experiment runs on Windows operating system, Pytorch is selected as the framework in the CUDA11.27 environment, and training is performed on GeForceRTX3090. The optimizer is selected, the initial learning rate is set to 0.001, the training batch is set to 64, and the data set is 140 days of long-term load metal fatigue data of steel structure buildings, which is input into the DWN prediction model after preprocessing.

[0051] Furthermore, the DWN model can achieve the fitting effect of long-term load metal fatigue prediction of steel structure buildings. Figure 5 As shown in the figure, the horizontal axis is the load cycle (days), the vertical axis is the metal fatigue degree (%), the gray dotted line and dots are real data, and the black solid line and squares are predicted data. It can be seen from the figure that the overall trend of the predicted curve is highly consistent with the real curve, especially in the early stage of the load cycle, the two curves basically overlap, indicating that the model can accurately capture the changing characteristics of short-term load metal fatigue and accurately model the initial fatigue condition. With the increase of the load cycle, the metal fatigue degree gradually increases, which indicates that the steel structure gradually bears a greater fatigue load. In summary, the overall prediction effect of the model is good, which can effectively capture the long-term load metal fatigue characteristics of steel structure buildings and has good long-term prediction capabilities.

Claims

1. A method for predicting long-term load metal fatigue of steel structure buildings based on machine learning, characterized in that: The following steps are involved: S1. Collect long-term load metal fatigue data of steel structure buildings, including physical parameters and use environment parameters, and pre-process the collected data; S2, the preprocessed metal fatigue data is standardized using the mean normalization method, and mapped through dilated convolution to divide the data into training set and test set; S3, build a frequency feature extraction module, introduce variational local adaptive wavelet transform, and build frequency domain features. The specific steps are as follows: S31, dynamically calculating the input sequence through variational optimization to select an adaptive scale; S32, calculating the local frequency domain energy of the adaptive scale, and optimizing the adaptive scale by path integration; S33, using the inverse graph Laplace weight calculation to optimize the contribution of different frequency components, and using the inverse wavelet transform to reconstruct the weighted wavelet features back to the time domain signal; S4. Build a time domain feature extraction module, design a learnable convolution kernel, and build time domain features through residual connections. The specific steps are as follows: S41. Design of learnable convolution kernels Perform convolution operation on the input signal; S42. Apply convolution results Activation function and layer normalization; S43, performing residual connection on the normalized output of the layer and the input signal, concatenating the features of all time steps, and obtaining the final time domain feature representation; S5. Build a learning prediction module and design learnable weights , adjust the contrast loss weight, train the model through contrast learning, and input the preprocessed data into the model to obtain the metal fatigue prediction results.

2. The method for predicting long-term load metal fatigue of steel structure buildings based on machine learning according to claim 1 is characterized in that: In step S31, the input signal Dynamically compute adaptive scale selection via variational optimization , the specific formula is: ； In the formula, is the wavelet scale, is the time offset, is the smoothing factor, For input signal Gradients with respect to wavelet scale; is the wavelet transform, the specific formula is: ； In the formula, is about the wavelet scale The Morlet basis function of is: ； In the formula, is the wavelet scale, is the time offset.

3. The method for predicting long-term load metal fatigue of steel structure buildings based on machine learning according to claim 2 is characterized in that: In step S32, the adaptive scale Calculate the local frequency domain energy and optimize the adaptive scale through path integration. The specific formula is: ； In the formula, is the optimal path obtained by variational optimization. The specific formula is: ； In the formula, is the local signal energy, and the specific formula is: ， In the formula, For adaptive scale Morlet basis functions.

4. The method for predicting long-term load metal fatigue of steel structure buildings based on machine learning according to claim 3 is characterized in that: In step S33, the inverse graph Laplace weight calculation is used to optimize the contribution of different frequency components. The specific formula is: ； In the formula, is the optimized adaptive scale, is the time offset, is the input signal, is the smoothing factor that controls the smoothness of the neighborhood, is the graph Laplace matrix, and the specific formula is: ； In the formula, is the degree matrix, is the adjacency matrix, which represents the similarity between wavelet scales; Degree Matrix The specific formula is: ； In the formula, is the number of nodes, For Node The specific formula is: ; Adjacency Matrix The specific formula is: ； in, is the number of nodes, the adjacency matrix The Line List Representation Node and nodes Is there an edge between them? The specific formula is: ； The weighted wavelet features are converted into time domain signals using the inverse wavelet transform. The specific formula is: ; In the formula, is the converted time domain signal, the specific formula is: ； In the formula, is the optimized adaptive scale, is the time offset.

5. The method for predicting long-term load metal fatigue of steel structure buildings based on machine learning according to claim 4 is characterized in that: In step S41, for the input sequence At each time step , design learnable convolution kernels For time step Input signal Perform convolution operation, the specific formula is: ； In the formula, is the output after the convolution operation, is a 1D convolution operation, Designed learnable convolution kernels; The specific formula for the learnable convolution kernel is: ； In the formula, is the local curvature, is a learnable parameter, is the maximum convolution kernel, is the minimum convolution kernel, is the Sigmoid function, is the rounding function; The specific formula for local curvature is: ； In the formula, is the second-order derivative; Learnable parameters It is obtained through training optimization. The specific formula is: ； ； In the formula, is the cross entropy loss function, is the convolution kernel weight, is the learning rate; The specific formula for updating the convolution kernel weight is: ； In the formula, is the cross entropy loss function, is the learning rate; The specific formula of the cross entropy loss function is: ； In the formula, is the total number of samples, is the true value, is the predicted value.

6. The method for predicting long-term load metal fatigue of steel structure buildings based on machine learning according to claim 5 is characterized in that: The convolution results at different times in step S42 are ,application The activation function gets the processed data , the specific formula is: ； In the formula, is the time step The output after dynamic convolution is is the activation function, the specific formula is: ； In the formula, and are predefined hyperparameters; right Perform layer normalization, the specific formula is: ； In the formula, is the layer normalization operation, the specific formula is: ； In the formula, For each time step The mean of For each time step The standard deviation of Small constant set to prevent division by zero errors.

7. The method for predicting long-term load metal fatigue of steel structure buildings based on machine learning according to claim 6 is characterized in that: In step S43, the output after layer normalization is With input signal Perform residual connection, the specific formula is: ； In the formula, It is the feature representation after adding residual connection; Use feature concatenation to merge the features of all time steps to obtain the final time domain feature representation. The specific formula is: ； In the formula, For splicing operation, is the time step Feature representation after residual connection.

8. The method for predicting long-term load metal fatigue of steel structure buildings based on machine learning according to claim 7 is characterized in that: In step S5, the features extracted from the time domain and the frequency domain are fused to form a unified feature representation The specific formula is ； In the formula, For splicing operation, is a linear layer, is the time domain feature representation, It is the frequency domain feature representation; Designing learnable weights , adjust the weights of the time domain contrast loss and the frequency domain contrast loss, use contrastive learning to learn the contrast features of the time domain and frequency domain, and minimize the total loss To train the model, the specific formula is: ； In the formula, is the contrast loss in the time domain, is the contrast loss in the frequency domain; Learnable weights The specific formula is: ； In the formula, is the rate of change of time domain loss, is the rate of change of frequency domain loss, It is a similarity measure of time domain and frequency domain features, controlling the weights of time domain and frequency domain. is the learning rate; Change rate of time domain loss The specific formula is: ； Frequency domain loss change rate The specific formula is: ； Similarity measures between time domain and frequency domain features The specific formula is: ； In the formula, is the L2 norm of the time domain feature, is the L2 norm of the frequency domain feature; Contrast loss in the time domain The specific formula is: ； In the formula, is the feature obtained through residual connection, is the final time domain feature representation; Frequency Domain Contrast Loss The specific formula is: ； In the formula, is the dynamic weight, is the amplitude loss, the specific formula is: ； In the formula, Sample size, is the predicted frequency domain amplitude, is the true frequency domain amplitude; is the phase loss, and the specific formula is: ； In the formula, is the predicted frequency domain phase, is the actual frequency domain phase.

9. The method for predicting long-term load metal fatigue of steel structure buildings based on machine learning according to claim 1, characterized in that: Aiming at the problem of long-term load metal fatigue prediction of steel structure buildings, the collected data for long-term load metal fatigue prediction of steel structure buildings include physical parameter data and usage environment parameter data of steel structure materials, among which the physical parameter data include tensile strength data, compressive strength data, yield strength data, density data, elastic modulus data, fatigue life data and hardness data of steel structure materials, and the installation environment parameter data include long-term load data, ambient temperature data, ambient humidity data, ambient vibration data and lighting condition data. The collected relevant data are preprocessed to ensure that there are no outliers in the data, and then the processed data are divided into training set and test set for training and evaluating the performance of the long-term load metal fatigue prediction model of steel structure buildings.

Images