Structural abnormal state diagnosis method and device considering complex environment influence

Abstract

The invention discloses a structure abnormal state diagnosis method and device considering complex environment influence, and the method comprises the steps: A, collecting structure reference state data under the influence of a time-varying environment, and constructing a training data set; b, designing and training a convolution auto-encoder, and converting high-dimensional original data into low-dimensional implicit features; c, designing and training a conditional diffusion probability model, introducing an attention mechanism to fuse and analyze multi-source data, and establishing a conditional probability model for reference state monitoring data under different environmental influences; d, calculating a reference state statistical index, and constructing an outlier recognition model SVDD by using the reference state statistical index; and E, inputting a test state data set, and diagnosing whether the test state is an abnormal state or not. According to the method, high-precision abnormal state diagnosis of the structure under the influence of the time-varying environment is realized by only needing a structure reference state data set and not depending on structural physical models such as finite elements and the like.

Description

Technical Field

[0001] The present invention relates to a method and device for diagnosing abnormal state of a structure taking into account the influence of complex environments. The method adopts a conditional probability deep learning model ATT-CLDPM to identify the benchmark state monitoring data pattern under different environmental influences, and constructs statistical indicators to measure the deviation between the test state and the benchmark state data pattern. The outlier recognition model SVDD is used to identify the statistical indicators that deviate from the benchmark state, so as to effectively diagnose the abnormal state of the structure under the influence of time-varying environment, and belongs to the field of civil structure health monitoring. Background Art

[0002] Civil engineering structures are subject to multiple influences such as loads, natural disasters, material aging, and environmental effects for a long time, and their structural performance gradually degrades, leading to potential safety hazards. In order to effectively monitor and evaluate the health of structures, data-driven structural anomaly diagnosis methods have received extensive attention. Such methods can accurately identify structural abnormalities by extracting and analyzing statistical patterns in monitoring data, thereby providing support for early warning of structural damage and destruction.

[0003] Deep learning methods are commonly used data-driven structural abnormal state diagnosis methods, which are divided into supervised learning methods and unsupervised learning methods. This type of method requires data on the normal and abnormal states of the structure as training samples, and extracts signal features of the normal and abnormal states of the structure from them to classify and identify the test state. However, structural abnormalities usually do not exist in the early stages of the structure and are quite rare in the long-term service life of the structure. Therefore, it may not be possible to obtain monitoring data related to the abnormal state of the structure. In order to overcome this difficulty, unsupervised learning methods that do not require labeled structural abnormal state data are applied to abnormal state diagnosis.

[0004] In recent years, unsupervised learning methods have gradually become a hot topic of research. These methods do not rely on abnormal state labeled data, but learn the data pattern under the baseline state, and diagnose abnormal behavior by obtaining the test state data pattern and the baseline state data pattern. Therefore, they have important applications in structural health monitoring. However, most existing methods fail to effectively consider the changes in environmental conditions, which plays a vital role in actual monitoring. Changes in environmental conditions, such as temperature, humidity, wind speed, etc., often affect the distribution of structural responses, which in turn leads to fluctuations in monitoring data. If these environmental factors are ignored, the abnormal diagnosis method may misjudge the changes in data patterns caused by environmental changes and fail to accurately diagnose the abnormal state of the structure. In order to overcome this difficulty, unsupervised learning methods that only require normal state data are applied to abnormal state diagnosis. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention proposes a method and device for diagnosing abnormal conditions of structures taking into account the influence of complex environments, so as to achieve high-precision abnormal condition diagnosis of structures under the influence of time-varying environments.

[0006] A first aspect of the present invention relates to a method for diagnosing structural abnormality considering the influence of complex environment, comprising the following steps:

[0007] A. Collect the structural baseline state data under the influence of time-varying environment, perform preprocessing, and construct a data set for deep learning model training;

[0008] B. Design and train a convolutional auto-encoder (CAE) to convert high-dimensional raw data into low-dimensional latent features, use convolution kernels and deconvolution kernels to extract key timing features and correlation features of multi-point signals, and select the optimal latent variable dimension and CAE hidden layer dimension according to the minimum convergence value of the loss function to optimize the CAE network, effectively using a small number of key features to characterize the raw data;

[0009] C. Introduce the U-net (Attention U-net, ATT-Unet) network with attention mechanism, design and train the Attention-guided Conditional Latent Diffusion Probabilistic Model (ATT-CLDPM), integrate and analyze multi-source data, establish a conditional probability model for the benchmark state monitoring data under different environmental influences, and select the optimal ATT-Unet hidden layer dimension according to the minimum convergence value of the loss function to achieve data pattern representation under different environmental conditions;

[0010] D. construct a support vector data representation model (SVDD) using the statistical indicator data point set under the baseline state, and construct a baseline state domain that represents the baseline state;

[0011] E. Input the test status data set and determine whether the test status is an abnormal state based on the relationship between the statistical indicator points and the benchmark status domain.

[0012] Further, step A specifically includes:

[0013] A1. Use sensing technology to collect long-term structural benchmark state data, including multi-point response data reflecting structural mechanical behavior and environmental condition data;

[0014] A2. Preprocess the structural benchmark state data and use the maximum-minimum normalization method to scale the data to the range of [0,1] to enhance the generalization ability of the deep learning model;

[0015] A3. Detrend the data collected by the sensor with drift fault to ensure the validity of the data;

[0016] A4. Use a fixed-step time window to slice the preprocessed data, and use the sliced ​​samples to construct a structural benchmark state dataset, which is divided into a training set and a validation set.

[0017] Further, step B specifically includes:

[0018] B1.CAE neural network is divided into two parts: encoder and decoder;

[0019] B2. The encoder gradually reduces the dimension of the original data through multiple convolutional layers and converts it into low-dimensional latent features;

[0020] B3. The decoder uses multiple deconvolution layers to up-scale low-dimensional latent features and reconstruct them into high-dimensional original data.

[0021] B4. When designing the CAE network framework, the latent variable dimension and hidden layer dimension are used as important design parameters to optimize the CAE network, so that the network framework can be adjusted according to the loss function later;

[0022] B5. The mean square error between the input original data and the output reconstructed data is used as the loss function of CAE. The training objective of CAE can be expressed as:

[0023]

[0024] in, represents the CAE loss function, x is the original input data, is the reconstructed data for output, ε represents the decoder and encoder respectively;

[0025] B6. For different latent variable dimensions c latent The CAE network is trained with the optimal number of iterations, and the optimal latent variable dimension c corresponding to the minimum convergence value of the damage function of the minimum validation set is selected. latent ;

[0026] B7. For different hidden layer dimensions c CAE The CAE network is trained with the optimal number of iterations, and the optimal hidden layer dimension c corresponding to the minimum convergence value of the damage function of the minimum validation set is selected. CAE ;

[0027] B8. Using the optimal latent variable dimension c latent and the optimal hidden layer dimension c CAE Design the optimal CAE network and use the training data set to train the optimal CAE network to convert the original data into low-dimensional latent features.

[0028] Further, step C specifically includes:

[0029] C1. Establish ATT-CLDPM to identify monitoring data patterns under different environmental conditions. The model is mainly divided into forward process and reverse process;

[0030] C2. The forward process is based on the Markov chain hypothesis. In the T-step Markov chain, the variance β is gradually added to the implicit features. t The zero-mean Gaussian noise of the original hidden feature is converted into standard Gaussian noise. The noisy hidden feature z of the tth step t It can be expressed as

[0031]

[0032] Among them, z0 is the original implicit feature z0=ε(x), The hyperparameter can be expressed as

[0033]

[0034] C3. The reverse process is to predict the noise component through ATT-Unet, and use the environmental condition variable ec and the initial implicit feature z0 as conditional constraints to gradually remove the noise component in the standard Gaussian noise and gradually restore it to the initial implicit feature. This process can be expressed as

[0035]

[0036] Among them, α t =1-β t , represents the noise predicted by ATT-Unet;

[0037] C4. When designing the ATT-Unet network framework, a multi-head attention mechanism and attention gate are introduced to perform fusion analysis on data of different dimensions;

[0038] C5. The input of multi-head attention includes environmental condition variables and the predicted noisy features of the t-th step structural response to form a fusion feature. The output of the multi-head attention feature is;

[0039]

[0040] Among them, MA represents the multi-head attention mechanism, W A , is a learnable linear transformation matrix, d represents the dimension of the environmental condition variable, and softmax(·) is the softmax activation function softmax(x i ) = exp(x i) / ∑ i exp(x i );

[0041] C6. Resize the output of multi-head attention to form fusion feature z ec , and compare it with the predicted noisy feature of the tth step The conditional feature z0 is combined and input into the subsequent steps;

[0042] The subsequent steps of C7.ATT-Unet are to first convolve the input features to obtain deep features, and then use upsampling to extract the predicted noise parameters from the deep features. In order to retain both deep and shallow features, jump connections with attention mechanisms are added between different layers, that is, the features in the previous convolutional layer and the upsampled features are connected through the attention gate, which can be expressed as

[0043]

[0044] in, is the output of the attention gate, α i is the attention weight, ψ,W v ,W g is a learnable linear transformation matrix;

[0045] C8. When designing the ATT-Unet network framework, the hidden layer dimension is used as an important design parameter so that the network framework can be adjusted according to the loss function later;

[0046] C9. Add the noise predicted by ATT-Unet and the actual noise ε added in the tth step t The mean square error is used as the loss function of ATT-Unet, and the training objective can be expressed as:

[0047]

[0048] in, Represents the loss function of ATT-Unet;

[0049] C10. For different hidden layer dimensions c ATT-Unet The ATT-Unet network uses the optimal number of iterations during training, and selects the optimal hidden layer dimension c corresponding to the minimum convergence value of the minimum validation set damage function. ATT-Unet .

[0050] Further, step D specifically includes:

[0051] D1. Obtain the abnormal statistical indicators MSE and ORSR based on the input and output data of ATT-CLDPM deep learning. The two abnormal statistical indicators can be expressed as:

[0052]

[0053] where x i is the i-th input sample of ATT-CLDPM, is the i-th output sample of ATT-CLDPM, and m is the total number of samples;

[0054] D2. Construct anomaly statistical indicator point set S = {s i |s i =(MSE i ,ORSR i )}, substitute the SVDD model to construct a minimum volume hypersphere plane as the boundary of the reference state domain. The reference state domain can surround all points corresponding to the reference state and exclude the points corresponding to the abnormal state. The minimum volume hypersphere plane can be expressed as:

[0055]

[0056] Where R is the radius of the hypersphere, C is the penalty coefficient, ξ i represents the i-th slack variable, l represents the number of slack variables, c represents the coordinates of the center of the hypersphere, Φ(s i ) means to convert s i The feature vector obtained after mapping to high-dimensional space;

[0057] D3. Use different combinations of penalty coefficients and Gaussian kernel coefficients to construct the SVDD model and select the appropriate reference state domain boundary.

[0058] Further, step E specifically includes:

[0059] E1. Substitute the data set collected under the structural test state into the trained ATT-CLDPM to obtain two abnormal statistical indicators MSE and ORSR;

[0060] E2. Use the decision function to determine whether the abnormal statistical indicator point corresponding to the test data set is within the decision boundary. The decision function can be expressed as:

[0061] f(s)=R 2 -||Φ(s)-c|| (9)

[0062] Where S represents the abnormal statistical indicator point of the test state. If f(s) is less than 0, the indicator point is outside the decision boundary, indicating that the structure is in an abnormal state at this time.

[0063] A second aspect of the present invention relates to a structural abnormal state diagnosis device that takes into account complex environmental influences, including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement a structural abnormal state diagnosis method that takes into account complex environmental influences of the present invention.

[0064] A third aspect of the present invention relates to a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements a structural abnormality diagnosis method of the present invention that takes into account complex environmental influences.

[0065] The innovations of the present invention are: First, based on the Markov chain hypothesis, CLDPM is used to model the conditional probability distribution of monitoring data, which helps to effectively identify data patterns under different environmental conditions. And introduce potential representation to map the original data to a low-dimensional feature space, thereby improving the accuracy and efficiency of abnormal diagnosis. Second, in order to conduct a comprehensive analysis of structural responses and environmental variables, an attention mechanism is introduced to form ATT-Unet, and integrated into CLDPM, realizing the fusion analysis of multi-source data such as structural response and environment. Third, two abnormality-sensitive statistics are selected to measure the difference between the distribution of monitoring data and the distribution of ATT-CLDPM modeling, and the outliers in the statistical data are identified through the SVDD method, thereby realizing the diagnosis of structural abnormality under the influence of complex environments.

[0066] The working principle of the present invention is: using the probability-based deep learning model CLDPM to perform probabilistic characterization of data distribution can better capture the variability of complex data patterns. By integrating conditional probability constraints, the formed conditional probability model can establish the conditional dependency relationship between structural response data and environmental variables, thereby accurately diagnosing structural anomalies in complex and changeable structural service environments.

[0067] The advantages of the present invention are: it can fuse and analyze structural response data and environmental data under complex environmental conditions, establish a conditional probability distribution model to identify monitoring data that characterizes structural abnormal conditions, without relying on accurate structural modeling and without obtaining various types of structural abnormal state data, has the advantage of simple operation, and can achieve high-precision and high-efficiency structural abnormal state diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 is a flow chart of the method of the present invention;

[0069] Figure 2 It is a schematic diagram of the dimensions of the monitoring structure of the present invention;

[0070] Figure 3 It is a schematic diagram of the ATT-LCDPM framework of the present invention;

[0071] Figure 4 It is a result diagram of the normal state statistical index and the reference state domain of the monitoring structure of the present invention;

[0072] Figure 5 It is a schematic diagram of an abnormal state of the monitoring structure of the present invention;

[0073] Figure 6 Diagram of the diagnostic results of the monitoring structure status of the present invention.

[0074] Figure 7 It is a schematic diagram of the device of the present invention. DETAILED DESCRIPTION

[0075] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0076] Example 1

[0077] like Figure 1 As shown, this embodiment relates to a structural abnormal state diagnosis method considering the influence of complex environment, which specifically includes the following steps:

[0078] A. Collect the structural baseline state data under the influence of time-varying environmental conditions, perform preprocessing, and construct a data set for deep learning model training; specifically include:

[0079] A1. Monitor the truss bridge structure, such as Figure 2 As shown, 8 acceleration sensors are arranged on the lower chord of the truss bridge, and the acceleration of the truss bridge structure under the influence of time-varying environmental conditions is sampled multiple times with a sampling frequency of 256 Hz. The duration of a single sampling is 32 seconds, and the environmental conditions at each sampling are recorded. Since the structure is sensitive to temperature, the temperature value reflecting the environmental conditions is recorded synchronously with each sampling, and the temperature variation range is 10°C to 30°C.

[0080] A2. Preprocess the collected baseline state data and use the maximum-minimum normalization method to scale the data to the range of [0,1] to enhance the generalization ability of the deep learning model

[0081] A3. Detrend the data collected by the sensor with drift fault to ensure the validity of the data;

[0082] A4. Use a time window with a fixed step size of 1024 to slice the preprocessed data, use the sliced ​​samples to construct a structural benchmark state dataset, and divide it into a training set and a validation set in a ratio of 19:1.

[0083] B. Design and train a convolutional auto-encoder (CAE) to convert high-dimensional raw data into low-dimensional latent features, use convolution kernels and deconvolution kernels to extract key timing features and correlation features of structural multi-point signals, and select the optimal latent variable dimension and CAE hidden layer dimension according to the minimum convergence value of the loss function to optimize the CAE network, and effectively use a small number of key features to characterize the raw data; specifically, including:

[0084] B1.CAE neural network is divided into two parts: encoder and decoder;

[0085] B2. The encoder gradually reduces the dimension of the original data through multiple convolutional layers and converts it into low-dimensional latent features. The encoder parameters are shown in Table 1:

[0086] Table 1 Encoder design parameters

[0087]

[0088] Note: Conv = convolution; BN = batch normalization; LReLU = LeakyReLU activation function;

[0089] B3. The decoder uses multiple deconvolution layers to up-scale the low-dimensional latent features and reconstruct them into high-dimensional original data. The decoder parameters are shown in Table 2:

[0090] Table 2 Decoder design parameters

[0091]

[0092] Note: DeConv = Deconvolution;

[0093] B4. When designing the CAE network framework, the latent variable dimension and hidden layer dimension are used as important design parameters to optimize the CAE network, so that the network framework can be adjusted according to the loss function later;

[0094] B5. The mean square error between the input original data and the output reconstructed data is used as the loss function of CAE, and the Adam algorithm is used to optimize the neural network training to minimize the loss function and realize network training;

[0095] B6. For different latent variable dimensions c latent The CAE network is trained with the optimal number of iterations, and the optimal latent variable dimension c corresponding to the minimum convergence value of the damage function of the minimum validation set is selected. latent ;

[0096] B7. For different hidden layer dimensions c CAEThe CAE network is trained with the optimal number of iterations, and the optimal hidden layer dimension c corresponding to the minimum convergence value of the damage function of the minimum validation set is selected. CAE ;

[0097] B8. Using the optimal latent variable dimension c latent and the optimal hidden layer dimension c CAE Design the optimal CAE network and use the training data set to train the optimal CAE network to convert the original data into low-dimensional latent features.

[0098] C. Introduce the U-net (Attention U-net, ATT-Unet) network with attention mechanism, design and train the Attention-guided Conditional Latent Diffusion Probabilistic Model (ATT-CLDPM), integrate and analyze multi-source data, establish a conditional probability model for the benchmark state monitoring data under different environmental influences, and select the optimal ATT-Unet hidden layer dimension according to the minimum convergence value of the loss function to achieve data pattern representation under different environmental conditions;

[0099] C1. Establish ATT-CLDPM to identify monitoring data patterns under different environmental conditions. The model is mainly divided into forward process and reverse process;

[0100] C2. The forward process is based on the Markov chain hypothesis. In the T-step Markov chain, the variance β is gradually added to the implicit features. t The zero-mean Gaussian noise transforms the original latent features into standard Gaussian noise;

[0101] C3. The reverse process is to predict the noise component through ATT-Unet, and use the environmental condition variable ec and the initial implicit feature z0 as conditional constraints to gradually remove the noise component in the standard Gaussian noise and gradually restore it to the initial implicit feature;

[0102] C4. When designing the ATT-Unet network framework, a multi-head attention mechanism and an attention gate are introduced to perform fusion analysis on data of different dimensions. The ATT-Unet network parameters are shown in Table 3:

[0103] Table 3 Decoder design parameters

[0104]

[0105]

[0106]

[0107]

[0108] Note: Concat = concatenation; ATT-Gate = attention gate;

[0109] C5. The input of multi-head attention includes environmental condition variables and the predicted noisy features of the t-th step structural response to form a fusion feature;

[0110] C6. Resize the output of multi-head attention to form fusion feature z ec , and compare it with the predicted noisy feature of the tth step The conditional feature z0 is combined and input into the subsequent steps;

[0111] The subsequent steps of C7.ATT-Unet are to first convolve the input features to obtain deep features, and then use upsampling to extract the predicted noise parameters from the deep features. In order to retain both deep and shallow features, jump connections with attention mechanisms are added between different layers, that is, the features in the previous convolutional layer and the upsampled features are passed through the attention gate;

[0112] C8. When designing the ATT-Unet network framework, the hidden layer dimension is used as an important design parameter so that the network framework can be adjusted according to the loss function later;

[0113] C9. Add the noise predicted by ATT-Unet and the actual noise ε added in the tth step t The mean square error is used as the loss function of ATT-Unet, and the Adam algorithm is used to optimize the neural network training to minimize the loss function and realize network training;

[0114] C10. For different hidden layer dimensions c ATT-Unet The ATT-Unet network uses the optimal number of iterations during training, and selects the optimal hidden layer dimension c corresponding to the minimum convergence value of the minimum validation set damage function. ATT-Unet .

[0115] D. Use the statistical indicator data point set under the baseline state to construct a support vector data representation model (SVDD) and construct a baseline state domain that represents the baseline state; specifically, it includes:

[0116] D1. Obtain the abnormal statistical indicators mean square error (MSE) and original-reconstructed signal ratio (ORSR) based on the input and output data of ATT-CLDPM deep learning;

[0117] D2. Construct an abnormal statistical indicator point set, substitute it into the SVDD model to construct a minimum volume hypersphere plane as the boundary of the reference state domain, which can surround all points corresponding to the reference state and exclude points corresponding to the abnormal state;

[0118] D3. Use different combinations of penalty coefficients and Gaussian kernel coefficients to build the SVDD model and select the appropriate abnormal state decision boundary.

[0119] E. Input the test status data set and determine whether the test status is abnormal based on the relationship between the statistical indicator points and the reference status domain; specifically including:

[0120] E1. The test state includes 4 test states: in test state 1, the structure has no abnormality, and the temperature range is 10℃ to 30℃; in test state 2, the structure has no abnormality, and the temperature range is 5℃ to 35℃; in test state 3, the cross section of a single lower chord of the structure is reduced by 20%, and the temperature range is 5℃ to 35℃; in test state 4, the cross section of two lower chords of the structure is reduced by 20%, and the temperature range is 5℃ to 35℃. Test state 1 and test state 2 are normal states, test state 3 and test state 4 are abnormal states, and the cross section of the reduced member is Figure 5 The data set collected under the structural test state is substituted into the trained ATT-CLDPM to obtain two abnormal statistical indicators MSE and ORSR;

[0121] E2. The decision function is used to determine whether the abnormal statistical indicator point corresponding to the test data set is within the decision boundary. The abnormal state diagnosis result is displayed in Figure 6 The diagnostic accuracy is summarized in Table 4.

[0122] Table 4 Structural status diagnosis accuracy

[0123]

[0124] Example 2

[0125] like Figure 7 This embodiment relates to a structural abnormal state diagnosis device that takes into account the influence of complex environments, including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, it is used to implement a structural abnormal state diagnosis method that takes into account the influence of complex environments in Example 1.

[0126] Example 3

[0127] This embodiment relates to a computer-readable storage medium on which a program is stored. When the program is executed by a processor, a structural abnormal state diagnosis method considering the influence of complex environments in embodiment 1 is implemented.

[0128] The contents described in the implementation cases of this specification are merely an enumeration of the implementation forms of the inventive concept. The protection scope of the present invention should not be deemed to be limited to the specific forms described in the implementation cases. The protection scope of the present invention also extends to the equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for diagnosing abnormal structural conditions taking into account the influence of complex environments, comprising the following steps: A. Collect the structural baseline state data under the influence of time-varying environment, perform preprocessing, and construct a data set for deep learning model training; B. Design and train a convolutional auto-encoder (CAE) to convert high-dimensional raw data into low-dimensional latent features, use convolution kernels and deconvolution kernels to extract key timing features and correlation features of multi-point signals, and select the optimal latent variable dimension and CAE hidden layer dimension according to the minimum convergence value of the loss function to optimize the CAE network, effectively using a small number of key features to characterize the raw data; C. Introduce the U-net (Attention U-net, ATT-Unet) network with attention mechanism, design and train the attention-guided conditional implicit diffusion probability model (Attention-guided Conditional Latent Diffusion Probabilistic Model (ATT-CLDPM), which integrates and analyzes multi-source data, establishes a conditional probability model for baseline condition monitoring data under different environmental influences, and selects the optimal ATT-Unet hidden layer dimension according to the minimum convergence value of the loss function to achieve data pattern representation under different environmental conditions; D. Using the statistical indicator data point set under the baseline state to construct a support vector data representation model (SVDD) and construct a baseline state domain that represents the baseline state; E. Input the test status data set and determine whether the test status is an abnormal state based on the relationship between the statistical indicator points and the benchmark status domain.

2. A structural abnormality diagnosis method considering the influence of complex environment as claimed in claim 1, characterized in that: Step A specifically includes: A1. Use sensing technology to collect long-term structural benchmark state data, including multi-point response data reflecting structural mechanical behavior and environmental condition data; A2. Preprocess the structural benchmark state data and use the maximum-minimum normalization method to scale the data to In the range of [0,1], the generalization ability of deep learning models is enhanced; A3. Detrend the data collected by the sensor with drift fault to ensure the validity of the data; A4. Use a fixed-step time window to slice the preprocessed data, and use the sliced ​​samples to construct a structural benchmark state dataset, which is divided into a training set and a validation set.

3. A structural abnormality diagnosis method considering the influence of complex environment as claimed in claim 1, characterized in that: Step B specifically includes: B1.CAE neural network is divided into two parts: encoder and decoder; B2. The encoder gradually reduces the dimension of the original data through multiple convolutional layers and converts it into low-dimensional latent features; B3. The decoder uses multiple deconvolution layers to increase the dimension of low-dimensional latent features and reconstruct them into high-dimensional original data; B4. When designing the CAE network framework, the latent variable dimension and hidden layer dimension are used as important design parameters to optimize the CAE network, so that the network framework can be adjusted according to the loss function later; B5. The mean square error between the input original data and the output reconstructed data is used as the loss function of CAE. The training objective of CAE can be expressed as: in, represents the CAE loss function, x is the original input data, is the output reconstructed data, ε represents the decoder and encoder respectively; B6. For different latent variable dimensions c latent The CAE network is trained with the optimal number of iterations, and the optimal latent variable dimension c corresponding to the minimum convergence value of the damage function of the minimum validation set is selected. latent ; B7. For different hidden layer dimensions c CAE The CAE network is trained with the optimal number of iterations, and the optimal hidden layer dimension c corresponding to the minimum convergence value of the damage function of the minimum validation set is selected. CAE ; B8. Using the optimal latent variable dimension c latent and the optimal hidden layer dimension c CAE Design the optimal CAE network and use the training data set to train the optimal CAE network to convert the original data into low-dimensional latent features.

4. A structural abnormality diagnosis method considering the influence of complex environment as claimed in claim 1, characterized in that: Step C specifically includes: C1. Establish ATT-CLDPM to identify monitoring data patterns under different environmental conditions. The model is mainly divided into forward process and reverse process; C2. The forward process is based on the Markov chain hypothesis. In the T-step Markov chain, the variance β is gradually added to the implicit features. t The zero-mean Gaussian noise of the original hidden feature is converted into standard Gaussian noise. The noisy hidden feature z of the tth step t It can be expressed as Among them, z0 is the original implicit feature z0=ε(x), The hyperparameter can be expressed as C3. The reverse process is to predict the noise component through ATT-Unet, and use the environmental condition variable ec and the initial implicit feature z0 as conditional constraints to gradually remove the noise component in the standard Gaussian noise and gradually restore it to the initial implicit feature. This process can be expressed as Among them, α t =1-β t , represents the noise predicted by ATT-Unet; C4. When designing the ATT-Unet network framework, a multi-head attention mechanism and attention gate are introduced to perform fusion analysis on data of different dimensions; C5. The input of multi-head attention includes environmental condition variables and the predicted noisy features of the t-th step structural response to form a fusion feature. The output of the multi-head attention feature is; Among them, MA represents the multi-head attention mechanism, W A , is a learnable linear transformation matrix, d represents the dimension of the environmental condition variable, and softmax(·) is the softmax activation function softmax(x i ) = exp(x i ) / ∑ i exp(x i ); C6. Resize the output of multi-head attention to form fusion feature z ec , and compare it with the predicted noisy feature of the tth step The conditional feature z0 is combined and input into the subsequent steps; The subsequent steps of C7.ATT-Unet are to first convolve the input features to obtain deep features, and then use upsampling to extract the predicted noise parameters from the deep features. In order to retain both deep and shallow features, jump connections with attention mechanisms are added between different layers, that is, the features in the previous convolutional layer and the upsampled features are connected through the attention gate, which can be expressed as in, is the output of the attention gate, α i is the attention weight, ψ,W v ,W g is a learnable linear transformation matrix; C8. When designing the ATT-Unet network framework, the hidden layer dimension is used as an important design parameter so that the network framework can be adjusted according to the loss function later; C9. Add the noise predicted by ATT-Unet and the actual noise ε added in the tth step t The mean square error is used as the loss function of ATT-Unet, and the training objective can be expressed as: in, Represents the loss function of ATT-Unet; C10. For different hidden layer dimensions c ATT-Unet The ATT-Unet network uses the optimal number of iterations during training, and selects the optimal hidden layer dimension c corresponding to the minimum convergence value of the minimum validation set damage function. ATT-Unet .

5. The structural abnormality diagnosis method considering the influence of complex environment as claimed in claim 1, characterized in that: Step D specifically includes: D1. Obtain the abnormal statistical indicators MSE and ORSR based on the input and output data of ATT-CLDPM deep learning. The two abnormal statistical indicators can be expressed as: where x i is the i-th input sample of ATT-CLDPM, is the i-th output sample of ATT-CLDPM, and m is the total number of samples; D2. Constructing abnormal statistical indicator point set S = {s i |s i =(MSE i ,ORSR i )}, substitute the SVDD model to construct a minimum volume hypersphere plane as the boundary of the reference state domain. The reference state domain can surround all points corresponding to the reference state and exclude the points corresponding to the abnormal state. The minimum volume hypersphere plane can be expressed as: Where R is the radius of the hypersphere, C is the penalty coefficient, ξ i represents the i-th slack variable, l represents the number of slack variables, c represents the coordinates of the center of the hypersphere, Φ(s i ) means to convert s i The feature vector obtained after mapping to high-dimensional space; D3. Use different combinations of penalty coefficients and Gaussian kernel coefficients to construct the SVDD model and select the appropriate reference state domain boundary.

6. A structural abnormality diagnosis method considering the influence of complex environment as claimed in claim 1, characterized in that: Step E specifically includes: E1. Substitute the data set collected under the structural test state into the trained ATT-CLDPM to obtain two abnormal statistical indicators MSE and ORSR; E2. Use the decision function to determine whether the abnormal statistical indicator point corresponding to the test data set is within the decision boundary. The decision function can be expressed as: f(s)=R 2 -||Φ(s)-c|| (9) Where S represents the abnormal statistical indicator point of the test state. If f(s) is less than 0, the indicator point is outside the decision boundary, indicating that the structure is in an abnormal state at this time.

7. A structural abnormality diagnosis device considering the influence of complex environment, characterized in that: It comprises a memory and one or more processors, wherein the memory stores executable codes, and when the one or more processors execute the executable codes, they are used to implement a structural abnormal state diagnosis method considering the influence of complex environments as described in any one of claims 1-6.

8. A computer-readable storage medium, characterized in that: A program is stored thereon, and when the program is executed by a processor, a structural abnormal state diagnosis method considering the influence of complex environment as described in any one of claims 1 to 6 is implemented.

Images