A structural damage identification method and device combining convolutional autoencoder and gradient boosting decision tree and a medium

Abstract

This invention discloses a method, device, and medium for structural damage identification that combines a convolutional autoencoder and a gradient boosting decision tree, relating to the fields of structural health monitoring and machine learning. The invention first uses a convolutional autoencoder to simultaneously extract features from structural response data from multiple structural response sensors. Then, the extracted deep-encoded features and corresponding damage labels are used as input to train a gradient boosting decision tree model, uncovering the relationship between features and damage, thereby achieving structural damage identification and improving the accuracy and speed of structural damage identification.

Description

Technical Field

[0001] This invention relates to the fields of structural health monitoring and machine learning, and in particular to a method, device and medium for structural damage identification that combines a convolutional autoencoder and a gradient boosting decision tree. Background Technology

[0002] Structural damage identification is of significant necessity and urgency. First, civil engineering structures or components in service are exposed to a variety of changing environments, such as temperature variations, rain or acid rain corrosion, and are subjected to various loads. Second, after operation, as the years of service increase, vulnerable components will exhibit different forms of damage, such as cracking, corrosion, broken wires, and material aging.

[0003] Traditional damage detection methods based on structural dynamic characteristics require deriving a complex relationship between damage and dynamic characteristics, which varies and exhibits complex nonlinear relationships. While current mainstream structural damage identification methods based on deep learning and other artificial intelligence algorithms have overcome the drawbacks of traditional methods, issues such as reliance on large amounts of training data, high equipment requirements, and long algorithm processing times still need improvement. Existing structural damage identification methods suffer from high manpower and material costs, low accuracy, and poor robustness. Therefore, improving the performance of intelligent structural damage identification methods is crucial and necessary.

[0004] In recent years, artificial intelligence technology has developed rapidly and has been applied to the field of structural damage recognition, with deep learning methods, primarily convolutional neural networks, being widely studied. While these techniques offer higher accuracy compared to traditional damage recognition methods, they still suffer from drawbacks such as high requirements for training samples, long training times, and demanding hardware requirements. Furthermore, since the data acquired by structural response sensors is in one-dimensional form, these methods often require converting the raw data into a two-dimensional matrix, leading to increased processing time, potential loss of original data information, and difficulty in processing data from multiple sensors simultaneously. Summary of the Invention

[0005] The purpose of this invention is to provide a method, device, and medium for structural damage identification that combines a convolutional autoencoder and a gradient boosting decision tree, which can improve the accuracy and speed of structural damage identification.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A structural damage identification method combining convolutional autoencoders and gradient boosting decision trees, the method comprising:

[0008] A data sample set for the target structure is constructed; the data sample set includes multiple data samples, which are structural response data from multiple structural response sensors with damage labels;

[0009] Based on the data sample set, an adaptive loss function is used to train and test the convolutional autoencoder to obtain the trained convolutional autoencoder.

[0010] The structure response data in the data sample set is reconstructed based on the trained convolutional autoencoder, and the output of the encoder in the trained convolutional autoencoder is extracted as the deep coding feature corresponding to the structure response data.

[0011] The deep coding features corresponding to the structural response data are used as input, and the damage labels corresponding to the structural response data are used as the real labels to construct training and testing samples.

[0012] The gradient boosting decision tree model is trained and tested based on the training and test samples to obtain the trained gradient boosting decision tree model.

[0013] Damage identification of the target structure is performed based on the trained convolutional autoencoder and the trained gradient boosting decision tree model. Specifically, data samples under unknown working conditions are processed by the trained convolutional autoencoder to extract features, and then the trained gradient boosting decision tree model is used for feature analysis to predict the damage condition and achieve damage identification.

[0014] Optionally, construct a data sample set for the target structure, specifically including:

[0015] A structural model of the target structure is constructed, and n structural response sensors are arranged on the structural model; the structural response sensors are used to acquire structural response data.

[0016] Damage conditions of the target structure can be simulated by replacing the preset elements in the structural model with components whose stiffness is less than the stiffness threshold or by reducing the cross-sectional area of ​​the preset elements in the structural model, or by reducing the elastic modulus of the preset elements in the structural model. In this way, different damage conditions can be simulated by changing the preset elements in the structural model, changing the amount of reduction in the cross-sectional area, or changing the amount of reduction in the elastic modulus.

[0017] Based on the n structural response sensors, structural response data under different damage conditions are obtained, and an n×l dataset is constructed, where l is the total length of all structural response data under different damage conditions.

[0018] The dataset is sliced ​​using a sliding window to obtain multiple n×m data samples, and a data sample set is constructed, where m is the length of the sliding window.

[0019] Optionally, the length of the sliding window can be an integer power of 2.

[0020] Optionally, the convolutional autoencoder includes an encoder and a decoder connected to each other. The output layer of the encoder is connected to the input layer of the decoder. The output layer of the encoder is used to output the deep encoded features. During training, the output of the decoder continuously approximates the real data, while the output (encoded block) of the encoder gradually extracts more complete data features.

[0021] Optionally, the encoder includes multiple convolutional modules connected in sequence; each convolutional module includes a convolutional layer, a batch normalization layer, and a Leaky_ReLU function.

[0022] The decoder includes multiple deconvolution modules connected in sequence, each deconvolution module comprising a deconvolution layer, a batch normalization layer, and a ReLU function.

[0023] Optionally, the adaptive loss function used to train the convolutional autoencoder is:

[0024]

[0025] Where, λ i The weighting coefficients are the structural response data of the i-th structural response sensor. loss i Let loss be the loss function for the structural response data of the i-th structural response sensor. i =MSE(X i ,X' i ), X i and X' i Let represent the structural response data collected by the i-th structural response sensor and its corresponding reconstructed data, respectively. n is the number of structural response sensors. abs() is the absolute value calculation function, mean() is the mean value calculation function, and MSE() is the mean square error calculation function.

[0026] Optionally, the trained convolutional autoencoder and the trained gradient boosting decision tree model are combined through a stacked model integration method. The gradient boosting decision tree model adopts a multi-layer tree structure, which is specifically used to analyze the abstract feature data extracted by the convolutional autoencoder to achieve efficient identification of structural damage.

[0027] Optionally, the gradient boosting decision tree model employs a distributed computing framework that allows the model to be executed in parallel on multiple processing nodes, thereby significantly improving the processing speed of feature analysis and damage identification.

[0028] A computer device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described above.

[0029] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the above-described method.

[0030] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0031] This invention provides a method, device, and medium for structural damage identification combining a convolutional autoencoder (CAE) and a gradient boosting decision tree (GFDS). The method includes: constructing a data sample set of a target structure; training and testing a CAE using an adaptive loss function on the data sample set to obtain a trained CAE; reconstructing structural response data in the data sample set based on the trained CAE, and extracting the encoder output from the trained CAE as the deep coding features corresponding to the structural response data; constructing training and testing samples by using the deep coding features corresponding to the structural response data as input and the damage labels corresponding to the structural response data as ground truth labels; training and testing a GFDS model based on the training and testing samples to obtain a trained GFDS model; and identifying damage to the target structure based on the trained CAE and the trained GFDS model. In this embodiment of the invention, a convolutional autoencoder is first used to extract features from the structural response data of multiple structural response sensors simultaneously. Then, the extracted deep coding features and corresponding damage labels are used as input to train a gradient boosting decision tree model, thereby mining the relationship between features and damage, thus realizing structural damage identification and improving the accuracy and speed of structural damage identification. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a flowchart illustrating a structural damage identification method combining a convolutional autoencoder and a gradient boosting decision tree, provided in an embodiment of the present invention.

[0034] Figure 2 The diagram shows the structure and sensor arrangement of a three-span continuous beam bridge provided in this embodiment of the invention.

[0035] Figure 3 This is a flowchart illustrating the specific operation of the structural damage identification method combining convolutional autoencoders and gradient boosting decision trees in a three-span continuous beam bridge structure, as provided in an embodiment of the present invention.

[0036] Figure 4 This is a comparison of the damage recognition results between the structural damage recognition method combining convolutional autoencoders and gradient boosting decision trees and the convolutional neural network method provided in this embodiment of the invention.

[0037] Figure 5 This is a comparison of the training and testing times of the structural damage identification method combining convolutional autoencoders and gradient boosting decision trees and the convolutional neural network method provided in this embodiment of the invention.

[0038] Figure 6 An internal structural diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] The purpose of this invention is to provide a method, device, and medium for structural damage identification that combines a convolutional autoencoder and a gradient boosting decision tree, which can improve the accuracy and speed of structural damage identification.

[0041] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] Example 1

[0043] This embodiment provides a structural damage identification method that combines a convolutional autoencoder and a gradient boosting decision tree. The method includes the following steps.

[0044] A data sample set for the target structure is constructed; the data sample set includes multiple data samples, which are structural response data from multiple structural response sensors with damage labels.

[0045] Based on the data sample set, an adaptive loss function is used to train and test the convolutional autoencoder to obtain the trained convolutional autoencoder.

[0046] The structure response data in the data sample set is reconstructed based on the trained convolutional autoencoder, and the output of the encoder in the trained convolutional autoencoder is extracted as the deep coding feature corresponding to the structure response data.

[0047] The deep coding features corresponding to the structural response data are used as input, and the damage labels corresponding to the structural response data are used as the real labels to construct training and testing samples.

[0048] The gradient boosting decision tree model is trained and tested based on the training and testing samples to obtain the trained gradient boosting decision tree model.

[0049] Damage identification of the target structure is performed based on the trained convolutional autoencoder and the trained gradient boosting decision tree model. Specifically, data samples under unknown working conditions are processed by the trained convolutional autoencoder to extract features, and then the trained gradient boosting decision tree model is used for feature analysis to predict the damage condition and achieve damage identification.

[0050] like Figure 1 and Figure 3 As shown, constructing the data sample set for the target structure involves the following steps:

[0051] S1. Build a structural model and arrange acceleration sensors.

[0052] Build an experimental or numerical model of the structure and arrange multiple acceleration sensors on the structure.

[0053] For example, this embodiment of the invention uses a three-span continuous beam bridge as an example for illustration. Figure 2 Figures (a) and (b) show schematic diagrams of a three-span continuous beam bridge and its sensor locations. Structural response signals were collected under simulated vehicle load excitation and then processed by... Figure 2 The sensor shown in (b) collects the vertical acceleration signal of the main beam, which is then transmitted via... Figure 1 The procedure shown is used for damage identification.

[0054] For example, taking a three-span continuous beam bridge as an example, an experimental model of the three-span continuous beam bridge structure was built, and five acceleration sensors were installed on the bridge.

[0055] S2. Collect the acceleration response of multiple parts of the structure under different working conditions.

[0056] The structure is subjected to load excitation, and the acceleration response of the structure under different damage conditions is collected to obtain structural response data. In this embodiment of the invention, damage is constructed by reducing the elastic modulus of a specified element, creating multiple damage conditions, including single damage and double damage.

[0057] Construct a data set from the original structural response data to obtain an \(n\times l\) matrix, where \(n\) is the number of sensors and \(l\) is the data length.

[0058] Exemplarily, taking a three-span continuous beam bridge as an example, \(n = 5\) acceleration sensors are installed on the three-span continuous beam bridge, and \(l = 200,000\) data points are collected for each working condition. Therefore, the size of the data set for each working condition is \(5\times200,000\).

[0059] S3. Perform data slicing to obtain data samples.

[0060] Use a sliding window to slice the original data set. The window length is \(m\) (\(m\lt l\)), and the sliding step is \(s\) (\(s\leq m\)), obtaining multiple data samples of size \(n\times m\).

[0061] Exemplarily, taking a three-span continuous beam bridge as an example, use the sliding window method for data slicing. The window length is selected as \(m = 512\), and the step size \(s = 128\). The window length is selected as an integer power of 2 to facilitate subsequent training and feature extraction of the convolutional autoencoder. The sliding step is selected as one-fourth of the window length, which not only ensures the difference of data samples but also plays a role in making full use of data and expanding the sample size. After data slicing, data samples of size \(5\times512\) are obtained.

[0062] Build, train, and test a convolutional autoencoder based on the data sample set to obtain a trained convolutional autoencoder, which specifically includes the following steps:

[0063] S4. Build, train, and test a convolutional autoencoder for feature extraction.

[0064] Design and build the network structure of the convolutional autoencoder. Put the data samples into the convolutional autoencoder for training and testing. When the loss function tends to zero and remains stable, extract the output of the encoder as the deep coding feature.

[0065] Use the deep coding feature corresponding to the structural response data as the input, and use the damage label corresponding to the structural response data as the true label to construct the training and testing samples of the gradient boosting decision tree.

[0066] In step S4, use the designed convolutional autoencoder. First, set hyperparameters such as the learning rate, number of training epochs, batch size, etc., and use a composite loss function, so that the convolutional autoencoder can simultaneously perform data reconstruction and feature extraction for multiple sensors and meet the accuracy requirements of all sensors.

[0067] In step S4, use the composite loss function as the loss function for training the convolutional autoencoder. The calculation formula of the composite loss function is where \(\lambda\) iThe weighting coefficients are the structural response data of the i-th structural response sensor. loss i Let loss be the loss function for the structural response data of the i-th structural response sensor. i =MSE(X i ,X' i ), X i and X' i Let represent the structural response data and reconstructed data collected by the i-th structural response sensor, respectively, where n is the number of structural response sensors, abs() is the absolute value calculation function, mean() is the mean value calculation function, and MSE() is the mean square error calculation function.

[0068] In step S4, when the reconstruction error represented by the composite loss function stabilizes and approaches zero, the coded block containing deep coding features is extracted as output.

[0069] For example, taking a three-span continuous beam bridge as an example, as shown in Table 1, the convolutional autoencoder network contains a total of ten layers, of which the encoder has five convolutional layers (Conv2d). Except for the first layer, each layer is batch normalized (BN) after convolution, and the activation function is Leaky_ReLU. The decoder uses five deconvolutional layers (UConv2d). Except for the last layer, each layer is batch normalized after deconvolution, and the activation function is ReLU.

[0070] Table 1. Structural parameters of the convolutional autoencoder

[0071]

[0072] Before training, we first set the hyperparameters. In this example, we set the batch size to 64, the learning rate to 0.01, the number of training epochs to 200, and the Adam optimizer to be selected. It uses the momentum optimization algorithm and has the advantage of adaptively adjusting the learning rate.

[0073] Based on the training and testing samples, the gradient boosting decision tree model is built, trained, and tested to obtain the trained gradient boosting decision tree model. The specific steps include the following:

[0074] S5. Build, train, and test the gradient boosting decision tree model.

[0075] A gradient boosting decision tree model was built using the Scikit-learn library in Python. The deep encoding features extracted from S4 and their corresponding damage labels were used as input. The model was trained and tested using a gradient boosting decision tree. First, the training samples were divided into training and testing sets. The gradient boosting decision tree model was trained and then tested on the testing set to evaluate its damage recognition performance.

[0076] In step S5, the encoded features and corresponding damage labels are used as sample data, and the gradient boosting decision tree algorithm is used for training and testing. The samples are divided into 80% for training and 20% for testing.

[0077] In step S5, the gradient boosting decision tree is trained using ensemble learning, combining multiple weak learners into a strong learner. Each iteration aims to fit the error from the previous iteration. This efficient training method learns from the feature data, thereby achieving damage detection.

[0078] In step S5, the specific training steps are as follows.

[0079] (1) Initialization: The model first creates a decision tree as a base learner for preliminary coarse prediction.

[0080] (2) Iterative training: In each iteration, the residuals of the previous model are first calculated, which is the difference between the true value and the current model's prediction. Then, a new decision tree is trained based on these residuals. The goal of this new tree is to predict the residuals of the previous step as accurately as possible.

[0081] (3) Adding new trees: The predictions of new trees are added to the previous model predictions, thereby gradually improving the predictive ability of the model. New trees are added in the direction of reducing the overall model error.

[0082] (4) Error evaluation and early stopping: After each iteration, the model's performance on the test set is evaluated. If the model's performance does not improve significantly in several consecutive iterations, or if the preset number of iterations is reached, the training process can be stopped early, which is called early stopping. Early stopping is an effective means of preventing overfitting.

[0083] (5) Output the final model: After several iterations, the training process stops when the model's performance reaches its optimal level or no longer improves. At this point, the gradient boosting decision tree model integrates multiple decision trees and can be used to predict new data.

[0084] In step S5, the LightGBM algorithm is specifically adopted. Compared with other gradient boosting decision tree algorithms, it has the advantages of being lightweight, fast, and supporting parallel computing.

[0085] In step S5, a multi-output classification algorithm is used to calculate the damage level at multiple damage locations simultaneously, so that multiple damage identification can be achieved with only one model.

[0086] The model's prediction results on the training and test sets are visualized, and the confusion matrix and ROC curve are plotted to evaluate the model's accuracy.

[0087] Damage identification of the target structure is performed based on the trained convolutional autoencoder and the trained gradient boosting decision tree model, specifically including:

[0088] S6. Perform damage identification.

[0089] Structural response data of the structure under unknown working conditions is collected to obtain the corresponding data sample set. Deep coding features are obtained through steps S1 to S4, and then analyzed by the trained and tested model in S5 to complete damage identification.

[0090] In this embodiment, both the training set and the test set can use structural acceleration signals collected from the actual structure, or simulation data generated by finite element simulation. When certain damage conditions of the actual structure are not convenient to construct, finite element simulation can be used for simulation.

[0091] When using actual structural data, it is necessary to first induce damage to the experimental structural components, such as by replacing low-stiffness components to simulate structural aging or reducing the cross-sectional area of ​​components to simulate structural wear. Loads can be applied using simulated vehicle loads or environmental excitations. Structural response data corresponding to each damage condition should be collected and labeled with damage tags.

[0092] When using finite element simulation data, damage can be simulated by reducing the elastic modulus of the material to decrease the element stiffness, or by reducing the element cross-sectional area. Simulated vehicle loads are applied to the numerical model, and structural response data corresponding to each damage condition is collected and labeled with damage tags.

[0093] Figure 4 This is a comparison chart of the damage recognition results of the Gradient Boosting Decision Tree (GBDT) model and the Convolutional Neural Network (CNN) method of this invention. Figure 4 As shown, under the same conditions, the recall rate of the identification results under most damage conditions is higher than that of the convolutional neural network method.

[0094] Figure 5 Figures (a) and (b) in the diagram compare the training and testing times of the method described in this invention and the convolutional neural network method, respectively. Figure 5As shown, the method of this invention achieves higher accuracy while requiring significantly less training time (convolutional autoencoder CAE and gradient boosting decision tree model GBDT) and testing time compared to convolutional neural network methods. In practical engineering applications and when dealing with massive amounts of data, it can significantly reduce the hardware and time costs required for computation, improve recognition efficiency by more than double, and enable real-time, efficient damage detection of structures.

[0095] The method of this invention can perform single-site damage identification and multi-site damage identification. The convolutional autoencoder, belonging to the category of self-supervised learning, can extract deep features from the original data; the gradient boosting decision tree is an ensemble learning algorithm that can be used to achieve high-precision and high-efficiency damage identification. Compared with other damage identification methods, it has advantages such as fast computation speed, no need for large amounts of data, high identification accuracy, and strong practicality.

[0096] Example 2

[0097] Embodiment 2 of the present invention provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method in Embodiment 1.

[0098] The computer device can be a database, and its internal structure diagram can be as follows: Figure 6 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The database stores pending transactions. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements the method in Embodiment 1.

[0099] Example 3

[0100] Embodiment 3 of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method in Embodiment 1.

[0101] It should be noted that the object information (including but not limited to object device information, object personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this invention are all information and data authorized by the object or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0102] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided by this invention may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided by this invention may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0103] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0104] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A structural damage identification method combining convolutional autoencoders and gradient boosting decision trees, characterized in that, The method includes: A data sample set for the target structure is constructed; the data sample set includes multiple data samples, which are structural response data from multiple structural response sensors with damage labels; Based on the data sample set, an adaptive loss function is used to train and test the convolutional autoencoder to obtain the trained convolutional autoencoder. The structure response data in the data sample set is reconstructed based on the trained convolutional autoencoder, and the output of the encoder in the trained convolutional autoencoder is extracted as the deep coding feature corresponding to the structure response data. The deep coding features corresponding to the structural response data are used as input, and the damage labels corresponding to the structural response data are used as the real labels to construct training and testing samples. The gradient boosting decision tree model is trained and tested based on the training and test samples to obtain the trained gradient boosting decision tree model. Damage identification is performed on the target structure based on the trained convolutional autoencoder and the trained gradient boosting decision tree model. The encoder in the convolutional autoencoder includes multiple convolutional modules connected in sequence; each convolutional module includes a convolutional layer, a batch normalization layer, and a Leaky_ReLU function. The decoder includes multiple deconvolution modules connected in sequence, and each deconvolution module includes a deconvolution layer, a batch normalization layer, and a ReLU function. The adaptive loss function used to train the convolutional autoencoder is: ； in, The weighting coefficients are the structural response data of the i-th structural response sensor. , Let i be the loss function of the structural response data of the i-th structural response sensor. , and Let represent the structural response data collected by the i-th structural response sensor and its corresponding reconstructed data, respectively, where n is the number of structural response sensors. For absolute value calculation functions, This is the mean calculation function. This is the function for calculating the mean square error; The trained convolutional autoencoder and the trained gradient boosting decision tree model are combined through a stacked model ensemble, with the gradient boosting decision tree model employing a multi-layered tree structure.

2. The structural damage identification method combining convolutional autoencoder and gradient boosting decision tree according to claim 1, characterized in that, The data sample set for constructing the target structure specifically includes: Construct a structural model of the target structure, and arrange the components on the structural model. n A structural response sensor; the structural response sensor is used to acquire structural response data; Damage conditions of the target structure can be simulated by replacing the preset elements in the structural model with components whose stiffness is less than the stiffness threshold or by reducing the cross-sectional area of ​​the preset elements in the structural model, or by reducing the elastic modulus of the preset elements in the structural model. In this way, different damage conditions can be simulated by changing the preset elements in the structural model, changing the amount of reduction in the cross-sectional area, or changing the amount of reduction in the elastic modulus. based on n The aforementioned structural response sensor obtains structural response data under different damage conditions, and constructs... The dataset, in which, This represents the total length of all structural response data under different damage conditions. The dataset is sliced ​​using a sliding window to obtain multiple... Data samples, construct a data sample set, where, The length of the sliding window.

3. The structural damage identification method combining convolutional autoencoder and gradient boosting decision tree according to claim 2, characterized in that, The length of the sliding window is an integer power of 2.

4. The structural damage identification method combining convolutional autoencoder and gradient boosting decision tree according to claim 1, characterized in that, The convolutional autoencoder includes an encoder and a decoder connected to each other. The output layer of the encoder is connected to the input layer of the decoder, and the output layer of the encoder is used to output the deep encoded features.

5. The structural damage identification method combining convolutional autoencoder and gradient boosting decision tree according to claim 1, characterized in that, The gradient boosting decision tree model employs a distributed computing framework.

6. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the method of any one of claims 1-5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-5.

Images