A bridge health monitoring abnormal data reconstruction method, system, equipment and product

Abstract

The present invention relates to a method, system, device and product for reconstructing abnormal data from bridge health monitoring. The method solves modal coordinates based on the structural response data set measured by effective measuring points of bridge health monitoring and the vibration mode matrix corresponding to the effective measuring points according to the modal superposition principle; estimates the reconstructed estimated value of the abnormal monitoring data; constructs a data reconstruction model architecture to constrain the training data reconstruction model with the trend of the reconstructed estimated value. The model extracts local temporal features of the data through a convolutional neural network module, dynamically quantifies the spatial correlation between measuring points through a channel attention mechanism module, fuses the local temporal features of the data and the spatial correlation between the measuring points, performs sequence analysis modeling based on the fused features through a bidirectional long short-term memory network module, and globally optimizes the hyperparameters of the model through a Bayesian optimization module; and obtains the reconstruction value of the abnormal data based on the trained data reconstruction model. The present invention has high reconstruction accuracy for abnormal data.

Description

Technical Field

[0001] The present invention relates to the technical field of structural health monitoring and digital signal processing, and in particular to a method, system, device and product for reconstructing abnormal data of bridge health monitoring. Background Art

[0002] As a key technical tool for the safe operation and maintenance of bridge structures, bridge structural health monitoring provides crucial data support for structural condition assessment and maintenance decision-making. The integrity and accuracy of monitoring data directly impact the reliability of structural condition assessments. However, in practical engineering applications, issues such as hardware failures and system transmission interruptions often lead to anomalies or even missing monitoring data, which seriously impacts the accuracy of bridge structural condition assessments. To address this critical issue, achieving high-precision reconstruction of anomaly data in bridge monitoring systems has become a hot topic of research.

[0003] Traditional data reconstruction methods, including regression, interpolation, fitting, and filling, analyze the statistical characteristics of existing data to infer the distribution of the overall data and estimate the missing parts. However, these methods have obvious limitations: first, when faced with massive amounts of monitoring data, dimensionality explosion is very likely to occur during the solution process, causing the computing system to crash; second, these traditional techniques have difficulty effectively capturing the correlation between the internal structure of the data and its characteristic attributes, making it difficult to ensure the final restoration quality.

[0004] With the rapid development of big data technology, deep learning algorithms are increasingly used in the field of structural health monitoring, and their powerful spatiotemporal feature extraction capabilities have demonstrated significant technical advantages.

[0005] For example, the paper "Anomaly Detection and Reconstruction of Bridge Monitoring Data Based on LSTM Neural Network" published in the journal "Journal of Wuhan University of Technology (Transportation Science and Engineering Edition)" uses LSTM neural network for time series prediction, identifies collective anomalies based on prediction error thresholds, and realizes data reconstruction and filling.

[0006] For example, the paper "Missing Data Reconstruction Algorithm for Bridge Monitoring Based on FSOM Neural Network" published in the journal "Electronic Design Engineering" uses FSOM neural network clustering and hierarchical data processing, combines spatiotemporal correlation analysis to calculate the missing data weights, and uses support vector regression to construct a reconstruction decision function to achieve high-precision data reconstruction with dynamic weight updates.

[0007] For example, the Chinese invention patent with publication number CN118133105A proposes a structural health monitoring data reconstruction method and system. It adopts the fusion of signal processing and deep learning, and a missing data reconstruction method based on compressed sensing (CS) and context encoder (CS-ECE). Through adversarial learning, it directly reconstructs time-frequency domain features from compressed signals, breaking through the sparsity limitations of traditional CS and achieving fast and high-precision reconstruction.

[0008] For example, the Chinese invention patent with publication number CN120197036A proposes an efficient time series analysis method based on a lightweight convolutional neural network (CNN), which reconstructs time domain data through raw data reading, sample standardization, adaptive time-frequency conversion, frequency domain component extraction and fitting.

[0009] For example, the Chinese invention patent with publication number CN120030312A combines causal convolutional neural networks and Transformer structures to propose a spatiotemporal joint feature reconstruction method, which can efficiently extract temporal features and spatial correlation features under a single architecture and reconstruct data in combination with the target sequence.

[0010] For example, the Chinese invention patent with publication number CN119357854A proposes a bridge strain time series data reconstruction method based on deep learning. By training a neural network model to learn normal data characteristics, abnormal monitoring data is reconstructed with high precision, and data repair and anomaly correction are achieved by combining time-frequency domain feature analysis.

[0011] However, existing research still lacks the ability to deeply integrate deep learning algorithms with the mechanical properties of bridge structures. The inherent mechanical response characteristics inherent in monitoring data are not fully explored, limiting the accuracy of structural condition assessments and inaccurate data reconstruction. This limitation can not only cause analysis results to deviate from actual engineering practices but also lead to significant safety misjudgments, hindering the reliable application of this technology in engineering practice.

[0012] Therefore, in the field of bridge health monitoring, it is imperative to take into account the structural characteristics of the bridge and the spatiotemporal correlation of the data to achieve high-precision data reconstruction.

[0013] For example, Chinese invention patent publication number CN119025788A proposes a bridge temperature monitoring data reconstruction method and system. This method eliminates the time lag effect of raw structural temperature data and uses a data decomposition model to decompose the structural temperature data to obtain structural temperature trend terms. Finally, a temporal convolutional network is used to capture time series dependencies and construct a bridge temperature monitoring data reconstruction model. However, this method is only applicable to temperature data reconstruction and has limited applicability to other monitoring indicators.

[0014] For example, the paper "Reconstructing Missing Data in Structural Monitoring by Fusion of EEMD and BiLSTM Deep Networks," published in the Journal of Chongqing University, combines the advantages of ensemble empirical mode decomposition (EEMD) and bidirectional long short-term memory (BiLSTM) networks in time series processing. Using EEMD to adaptively decompose time series data into multi-scale IMF components to stabilize the signal, the BiLSTM network is then used to reconstruct missing data with high precision. However, this method only addresses the nonlinear and non-stationary characteristics of monitoring data and fails to incorporate bridge structural characteristics as prior knowledge to constrain trends in deep learning models.

[0015] For example, Chinese invention patent publication number CN118193504A proposes an EMD-GRU-based method for reconstructing missing time series data from bridge sensors. This method uses empirical mode decomposition (EMD) to decompose the data into multi-scale intrinsic mode sequences. A GRU neural network, combined with grid search optimization, is then used to predict and reconstruct missing data with high precision. However, GRU neural networks are primarily adept at capturing temporal dependencies in time series and are less capable of capturing spatial correlations.

[0016] It can be seen that although the current bridge health monitoring abnormal data reconstruction method can reconstruct the data to a certain extent, the data reconstruction accuracy still needs to be improved. Summary of the Invention

[0017] Based on this, it is necessary to provide a bridge health monitoring abnormal data reconstruction method, system, equipment and product to address the problem that the accuracy of bridge health monitoring abnormal data reconstruction still needs to be improved.

[0018] In order to solve the above problems, the present invention adopts the following technical solutions:

[0019] In a first aspect, the present invention provides a method for reconstructing abnormal data from bridge health monitoring, comprising the following steps:

[0020] According to the modal superposition principle, the modal coordinates are solved based on the structural response data set measured at the effective measurement points of bridge health monitoring and the vibration mode matrix corresponding to the effective measurement points;

[0021] According to the modal coordinates and the vibration shape vectors of the abnormal measurement points, the reconstructed estimated values ​​of the abnormal monitoring data are estimated;

[0022] Constructing a data reconstruction model architecture and training the data reconstruction model, wherein the data reconstruction model uses the estimated trend of the reconstructed estimated value as a constraint; the data reconstruction model architecture includes a convolutional neural network module, a channel attention mechanism module, a bidirectional long short-term memory network module, and a Bayesian optimization module, wherein the convolutional neural network module is used to extract local features of the data time series, the channel attention mechanism module is used to dynamically quantify the spatial correlation between measurement points, the bidirectional long short-term memory network module is used to perform sequence analysis modeling based on fusion features, and the Bayesian optimization module is used to globally optimize the hyperparameters of the deep learning model, wherein the fusion features are features obtained by fusing the local features of the data time series and the spatial correlation between the measurement points;

[0023] The reconstructed value of the abnormal data is obtained according to the trained data reconstruction model.

[0024] In a preferred embodiment, the calculation formula of the modal coordinates is:

[0025]

[0026] in, represents the modal coordinates, Represents the vibration mode matrix of the effective measuring point, The displacement mode matrix with dimension m×n and the strain mode matrix with dimension k×n Composition, m represents the number of effective displacement measurement points, k represents the number of effective strain measurement points, n is the modal order, m+k≥n, represents the structural response data set of the effective measurement points, The displacement response matrix with dimension m×p and the strain response matrix with dimension k×p Composition, p represents the measuring point The amount of data collected within a certain period of time.

[0027] In a preferred embodiment, the formula for estimating the reconstructed estimated value of the abnormal monitoring data is:

[0028]

[0029] in, represents the reconstructed estimated value of the abnormal monitoring data, Represents the vibration mode matrix of abnormal measurement points with dimension s×n, s represents the number of abnormal measurement points, s≤m+k.

[0030] In a preferred embodiment, the channel attention mechanism module is also used to evaluate the importance of each feature channel, enhance the expressive power of key features through feature recalibration, and stimulate or compress corresponding channels for different tasks to extract useful features.

[0031] In a preferred embodiment, the hyperparameters include the number of BiLSTM units, learning rate, batch size, cycle, and number of Bayesian optimizations, and the objective function in the optimization process of the hyperparameters of the global optimization deep learning model adopts the root mean square error function.

[0032] In a preferred embodiment, the data reconstruction model architecture includes a feature fusion layer for fusing the local temporal features of the data and the spatial correlation between the measurement points to obtain fused features.

[0033] In a preferred embodiment, the input of the data reconstruction model includes: the estimated reconstructed value of the abnormal monitoring data, the data monitored by the effective measuring points, and the normal data monitored by the abnormal measuring points, and the output is the reconstructed value of the abnormal data.

[0034] In a second aspect, the present invention provides a bridge health monitoring abnormal data reconstruction system, comprising:

[0035] A calculation module is used to solve the modal coordinates based on the modal superposition principle, the structural response data set measured by the effective measuring points of the bridge health monitoring and the vibration mode matrix corresponding to the effective measuring points;

[0036] An estimation module is used to estimate the reconstructed estimated value of the abnormal monitoring data based on the modal coordinates and the vibration shape vector of the abnormal measurement point;

[0037] A model construction and training module is used to construct a data reconstruction model architecture and train the data reconstruction model; the data reconstruction model uses the estimated trend of the reconstructed estimated value as a constraint; the data reconstruction model architecture includes a convolutional neural network module, a channel attention mechanism module, a bidirectional long short-term memory network module, and a Bayesian optimization module; the convolutional neural network module is used to extract local features of the data time series, the channel attention mechanism module is used to dynamically quantify the spatial correlation between measurement points, and the bidirectional long short-term memory network module is used to perform sequence analysis modeling based on fusion features; the Bayesian optimization module is used to globally optimize the hyperparameters of the deep learning model; the fusion feature is a feature obtained by fusing the local features of the data time series and the spatial correlation between the measurement points;

[0038] The acquisition module is used to obtain the reconstructed value of the abnormal data according to the trained data reconstruction model.

[0039] In a third aspect, the present invention provides an electronic device comprising: a memory; one or more processors; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for executing any one of the methods for reconstructing abnormal data of bridge health monitoring according to the first aspect and its preferred embodiments.

[0040] In a fourth aspect, the present invention provides a computer program product, comprising a computer program or instructions, which, when executed by a processor, implements a bridge health monitoring abnormal data reconstruction method according to the first aspect and any one of its preferred embodiments.

[0041] The present invention provides a bridge health monitoring anomaly data reconstruction method, system, device, and product. Using the principle of modal superposition, the method obtains reconstructed estimates of bridge health monitoring anomaly data that take into account the structural characteristics of the bridge. This estimate is used as prior knowledge. Under the constraints of the trend of the reconstructed estimates, the method utilizes deep learning within a data reconstruction model that includes a convolutional neural network module, a channel attention mechanism module, a bidirectional long short-term memory network module, and a Bayesian optimization module to perform data reconstruction. The model deeply explores the spatiotemporal characteristics of the data while considering the structural characteristics of the bridge, and outputs high-precision reconstruction results with physically meaningful constraints. While maintaining computational efficiency, the present invention significantly enhances the model's sensitivity to key features, improving data reconstruction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic flow chart of a method in one embodiment of the present invention;

[0043] Figure 2 A finite element model of a real bridge in one embodiment of the present invention;

[0044] Figure 3 This is a layout diagram of displacement measurement points on a real bridge in one embodiment of the present invention;

[0045] Figure 4 This is one of the data reconstruction model architecture and training flow charts of the present invention;

[0046] Figure 5 This is an example diagram of the displacement reconstruction result 16 seconds after the missing measuring point D3 in one embodiment of the present invention;

[0047] Figure 6 Another example diagram of the displacement reconstruction result 16 seconds after the missing measuring point D3 in one embodiment of the present invention;

[0048] Figure 7 A schematic diagram of the structure of a system in one embodiment of the present invention;

[0049] Figure 8 FIG. 1 is a schematic diagram of an electronic device according to one embodiment of the present invention. DETAILED DESCRIPTION

[0050] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0051] Although the current bridge health monitoring abnormal data reconstruction method can reconstruct the data to a certain extent, it fails to integrate the dynamic characteristics of the bridge structure as prior knowledge into the deep learning model, and fails to take into account the spatiotemporal correlation of the bridge structure characteristics and the data. The data reconstruction accuracy needs to be improved. To this end, the present invention provides a bridge health monitoring abnormal data reconstruction method, see Figure 1 , including the following steps:

[0052] According to the modal superposition principle, the modal coordinates are solved based on the structural response data set measured by the effective measurement points of the bridge health monitoring and the vibration mode matrix corresponding to the effective measurement points;

[0053] According to the modal coordinates and the vibration shape vectors of the abnormal measurement points, the reconstructed estimated values ​​of the abnormal monitoring data are estimated;

[0054] Constructing a data reconstruction model architecture and training the data reconstruction model, wherein the data reconstruction model uses the estimated trend of the reconstructed estimated value as a constraint; the data reconstruction model architecture includes a convolutional neural network module, a channel attention mechanism module, a bidirectional long short-term memory network module, and a Bayesian optimization module, wherein the convolutional neural network module is used to extract local features of the data time series, the channel attention mechanism module is used to dynamically quantify the spatial correlation between measurement points, the bidirectional long short-term memory network module is used to perform sequence analysis modeling based on fusion features, and the Bayesian optimization module is used to globally optimize the hyperparameters of the deep learning model, wherein the fusion features are features obtained by fusing the local features of the data time series and the spatial correlation between the measurement points;

[0055] The reconstructed value of the abnormal data is obtained according to the trained data reconstruction model.

[0056] It is understandable that valid measuring points are non-abnormal measuring points. Usually, data anomalies mainly include six categories: outliers, missing values, duplicate values, drift values, gain values, and trend abnormal values. The abnormal data described in the embodiment of the present invention is any one, two or more of these categories, that is, the abnormal data to be reconstructed includes at least one of outlier data, missing data, duplicate data, drift data, gain data, and trend abnormal data. No limitation is made here. It is understandable that all abnormal data are suitable for data reconstruction by this method. The following uses data missing as an example to illustrate the bridge health monitoring abnormal data reconstruction method. Obviously, the present invention does not limit the abnormal data to missing data.

[0057] A bridge health monitoring abnormal data reconstruction method includes the following steps:

[0058] Step 1: According to the modal superposition principle, the modal coordinates are solved based on the structural response data set measured by the effective measuring points of bridge health monitoring (the data set includes displacement information, strain information, etc.) and the vibration mode matrix corresponding to the effective measuring points. The calculation formula is as follows:

[0059]

[0060] Where, Indicates time, Represents the vibration mode matrix of the effective measuring point, From the displacement mode matrix (dimension is m×n) and the strain mode matrix (dimension is k×n), where m and k represent the number of effective displacement and strain measurement points, respectively, and n is the modal order. There is a unique solution, which must satisfy =n, that is, the number of effective measuring points m+k≥n, Represents the rank function. represents the structural response data set of the effective measurement points, From the displacement response matrix (dimension is m×p) and the strain response matrix (dimension is k×p), where p represents the measurement point The amount of data collected within a certain period of time. Non-square matrix, need to be solved by generalized inverse matrix, first calculate Get an n×n square matrix, and then find The pseudo-inverse matrix , and will eventually and Multiplication to solve .

[0061] Step 2: Combine the modal coordinates obtained in step 1 and the vibration mode vector of the abnormal measurement point to estimate the reconstructed estimated value of the abnormal monitoring data , see the following formula, and obtain the estimated value of the structural response As prior knowledge, it provides structural response trend constraints for subsequent deep learning models (data reconstruction models).

[0062]

[0063] Where, The mode shape matrix of the abnormal measurement points is s×n, where s represents the number of abnormal measurement points. It is worth noting that s≤m+k is required to ensure the feasibility of reconstruction. It represents the reconstructed estimated value of the abnormal monitoring data, that is, the reconstructed estimated value of the structural response data of the abnormal measurement point.

[0064] Step 3: A BO-CNN-BiLSTM-SENet deep learning model (data reconstruction model) was constructed to perform data reconstruction and train the model. Under the trend constraint of the reconstruction estimates obtained using the modal superposition method, a convolutional neural network (CNN) module was used to extract local temporal features of the data, and a channel attention mechanism (SENet) module was used to dynamically quantify the spatial correlation between measurement points. The local temporal features of the data were fused with the spatial correlation between measurement points to generate fused features. The bidirectional long short-term memory (BiLSTM) module performed sequence analysis and modeling based on the fused features. A Bayesian optimization module was used to globally optimize hyperparameters to obtain the reconstructed values ​​of the missing data.

[0065] A feature fusion layer is used to fuse the local temporal features of the data and the spatial correlation between the measurement points. The feature fusion layer can be a separate module, that is, the data reconstruction model includes the feature fusion layer, and can belong to a bidirectional long short-term memory network module, that is, the bidirectional long short-term memory network module includes the feature fusion layer.

[0066] Step 4: Reconstruct the model based on the trained data to obtain the reconstructed value of the abnormal data.

[0067] The following describes in detail the precise reconstruction of abnormal data using an example embodiment of an application, which includes the following steps:

[0068] S101: According to the modal superposition principle, based on the vibration mode matrix corresponding to the effective measuring points and based on the structural response data set measured at the effective measuring points, the modal coordinates are solved.

[0069] The structural response data set (displacement, strain, etc.) measured by the effective measuring points is obtained in the bridge health monitoring system. In order to further illustrate and verify the high-precision reconstruction of abnormal measuring point data by the method, a numerical simulation method is used to first establish a finite element model of a real bridge. The finite element model is as follows: Figure 2 As shown, displacement measuring points D1 to D6 are arranged symmetrically upstream and downstream, and displacement measuring points are arranged symmetrically upstream and downstream at the 1 / 4 span (measuring point D1 and measuring point D4), the middle of the span (measuring point D2 and measuring point D5), and the 3 / 4 span (measuring point D3 and measuring point D6) of the bridge. Figure 3As shown in the figure. In the numerical simulation, a random seismic wave with a duration of 80 s, a mean of 0, and a standard deviation of 0.02 g was applied to the finite element model. Time history analysis was performed with a time step of 0.01 s, and the displacement response data of each measuring point for a total of 8000 time steps were obtained. To simulate the data missing situation, it is assumed that the data of measuring point D3 in the first 64 s (6400 time steps) are complete and correct, and the displacement response data of the last 16 s (1600 time steps) are missing. Therefore, the structural response data set measured by the above effective measuring points is It is the 80 s displacement response data matrix composed of measuring points D1, D2, D4, D5, and D6, and its dimension is 5×8000.

[0070] The mode shape matrix corresponding to the measuring point location is composed of the normalized modal shape values ​​of the first n vertical modes at each measuring point. This matrix can be obtained through field measurement or finite element numerical simulation. In this embodiment of the present invention, a total of five valid measuring points (D1, D2, D4, D5, and D6) are involved. Therefore, n ≤ 5. The modal shape values ​​of the first five vertical modes at each valid measuring point are obtained using the finite element method. See Table 1 for details. The resulting corresponding mode shape matrix is ​​a 5×5 matrix, and its specific form is shown below.

[0071]

[0072] Table 1

[0073]

[0074] The modal coordinates can be substituted into the following formula to obtain the solution.

[0075]

[0076] Step S102: Based on the modal coordinates obtained in step S101 The abnormal monitoring data is reconstructed and estimated by taking into account the vibration mode vectors of the missing measuring point D3. In the simulation scenario, the data for the first 64 seconds (6400 time steps) of measuring point D3 is complete, but the displacement response data for the last 16 seconds (1600 time steps) is missing. First, the first five-order vertical mode vibration mode matrices of measuring point D3 are determined. The specific form is shown below.

[0077]

[0078] By substituting the mode matrix into the reconstruction calculation formula, the reconstruction estimation value of the abnormal monitoring data based on the modal superposition principle and taking into account the structural characteristics of the bridge can be obtained.

[0079]

[0080] Step S103: Construct the BO-CNN-BiLSTM-SENet deep learning model architecture and reconstruct the model based on training data, such as Figure 4 As shown, data reconstruction is carried out to obtain the reconstructed values ​​of the missing data.

[0081] The convolutional neural network (CNN) module, under the trend constraint of the response reconstruction estimate obtained by the modal superposition method, deeply explores the local temporal characteristics of the data. It includes convolutional layers and pooling layers. The convolutional layer captures characteristic patterns through sliding convolution kernels, significantly reducing the parameter scale; the pooling layer (maximum pooling layer) compresses the feature dimension through the maximum pooling operation, retains key information, and improves the generalization of the model.

[0082] The channel attention mechanism (SENet) module dynamically quantifies the spatial correlation between measurement points, evaluates the importance of each feature channel, enhances the expressiveness of key features through feature recalibration, and stimulates or compresses corresponding channels for different tasks to extract useful features, thereby improving the generalization ability of the data reconstruction model and improving the accuracy of the data reconstruction model's prediction.

[0083] The Bidirectional Long Short-Term Memory (BiLSTM) module includes a forward LSTM layer and a backward LSTM layer, and furthermore, a feature fusion layer. The feature fusion layer integrates spatiotemporal features, while the forward and backward LSTM layers perform sequence analysis and modeling. The collaborative work of the forward and backward LSTM layers simultaneously captures both forward-looking and retrospective information in the sequence, significantly improving the accuracy and stability of long-term predictions.

[0084] The Bayesian Optimization (BO) module uses a Bayesian optimization algorithm to achieve global optimization of hyperparameters during the data reconstruction model optimization process. The BO module's processing includes: defining hyperparameter ranges, estimating current hyperparameters, calculating the objective function value, updating hyperparameters, and determining whether the preset maximum number of iterations has been reached. If not, the module returns to the step of estimating the current hyperparameters and re-estimates them. If so, the model is updated with the current hyperparameters to obtain the trained data reconstruction model. These hyperparameters include the number of BiLSTM units, the learning rate, the batch size, the number of cycles, and the number of Bayesian optimization cycles. The number of BiLSTM units determines the number of neurons in the BiLSTM hidden layer, affecting the model's complexity and learning ability. The learning rate determines the step size for updating the model's weights, and its setting must balance training stability and convergence speed. A high learning rate can lead to training instability or miss the optimal solution, while a low learning rate can slow the training process. The batch size determines the number of samples used in each training iteration, affecting the model's training speed and generalization ability. The cycle determines the number of times the model is trained and has a significant impact on its final performance. The number of optimization cycles is set empirically to ensure a sufficient parameter search. The optimization objective function uses the root mean square error function, which comprehensively considers the overall deviation between the predicted value and the true value to ensure the accuracy of the reconstruction results. The specific optimization ranges of each hyperparameter and the optimization results are shown in Table 2.

[0085] Table 2

[0086]

[0087] Training Data Reconstruction Model: This embodiment utilizes a conventional training method using a training set and a test set. The model inputs include the reconstructed estimated values ​​of the abnormal monitoring data, data monitored at valid measuring points, and normal data monitored at the abnormal measuring point (i.e., data when the data is normal, i.e., complete). It will be appreciated that if the abnormal measuring point has no measured data or no normal data, the inputs are solely the reconstructed estimated values ​​of the abnormal monitoring data and the data monitored at the valid measuring points. The output of the data reconstruction model is the reconstructed values ​​of the abnormal data. In this embodiment, the data monitored at the valid measuring points are complete and accurate 80-second displacement response data from five valid measuring points (measurement points D1, D2, D4, D5, and D6). The complete data at the abnormal measuring point is the data 64 seconds prior to measurement point D3.

[0088] Step S104: reconstruct abnormal data according to the trained data reconstruction model.

[0089] After adopting the data reconstruction method of this embodiment, the displacement reconstruction result 16s after the missing measuring point D3 is compared with the finite element numerical calculation result. Figure 5 and Figure 6 shown. Figure 5 The results of numerical calculation and the reconstruction values ​​obtained by the method of this embodiment, Figure 6 The data points of the reconstructed displacement values ​​are more closely clustered on the y=x reference line (the reference line is Figure 6 The red slash line in the figure is near the coefficient of determination R 2 The above phenomenon shows that the data reconstruction method of this embodiment effectively improves the accuracy of health monitoring data response reconstruction by taking into account the structural characteristics of the bridge and the spatiotemporal correlation of the data, and obtains accurate reconstruction values ​​of abnormal data.

[0090] See also Figure 7 The present invention provides a bridge health monitoring abnormal data reconstruction system 100, which includes: a calculation module 10, an estimation module 20, a model construction and training module 30, and an acquisition module 40.

[0091] The calculation module 10 is used to solve the modal coordinates based on the structural response data set measured by the effective measuring points of the bridge health monitoring and the mode shape matrix corresponding to the effective measuring points according to the modal superposition principle;

[0092] The estimation module 20 is used to estimate the reconstructed estimated value of the abnormal monitoring data based on the modal coordinates and the vibration shape vector of the abnormal measurement point;

[0093] The model construction and training module 30 is used to construct a data reconstruction model architecture and train the data reconstruction model; the data reconstruction model uses the estimated trend of the reconstructed estimated value as a constraint; the data reconstruction model architecture includes a convolutional neural network module, a channel attention mechanism module, a bidirectional long short-term memory network module, and a Bayesian optimization module. The convolutional neural network module is used to extract local features of the data time series, the channel attention mechanism module is used to dynamically quantify the spatial correlation between measurement points, the bidirectional long short-term memory network module is used to perform sequence analysis modeling based on fusion features, and the Bayesian optimization module is used to globally optimize the hyperparameters of the deep learning model. The fusion feature is a feature obtained by fusing the local features of the data time series and the spatial correlation between the measurement points.

[0094] The obtaining module 40 is used to obtain the reconstructed value of the abnormal data according to the trained data reconstruction model.

[0095] In this embodiment, the formula used by the calculation module 10 to solve the modal coordinates is: .

[0096] In this embodiment, the estimation module 20 uses the following estimation formula to estimate the reconstructed estimated value of the abnormal monitoring data: .

[0097] When the bridge health monitoring abnormal data reconstruction system 100 is specifically implemented, abnormal data reconstruction can be achieved by referring to a bridge health monitoring abnormal data reconstruction method in any of the above embodiments, and the specific implementation steps are not repeated here.

[0098] See also Figure 8 According to the method of the present invention, an electronic device can be implemented, and the electronic device includes: a memory; one or more processors; one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the fast iteration method of the positioning loss detection module according to any of the above embodiments.

[0099] The processor may be a central processing unit (CPU), or other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0100] The memory may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage devices. In addition, the memory may include any combination of computer-readable storage media, such as semiconductor memory chips, magnetic disks, and optical disks.

[0101] The memory stores executable codes, which, when processed by the processor, can enable the processor to execute part or all of the above-mentioned methods.

[0102] Those skilled in the art to which the present invention belongs can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0103] The present invention also provides a computer program product, including a computer program or instructions, which, when executed by a processor, implements the bridge health monitoring abnormal data reconstruction method described in any of the above embodiments.

[0104] The effects of the bridge health monitoring abnormal data reconstruction method, system, device and product of the present invention are:

[0105] This invention improves the accuracy of reconstructing abnormal bridge health monitoring data. Traditional bridge health monitoring data reconstruction methods fail to simultaneously consider the temporal or spatiotemporal correlations between bridge structural characteristics and data. This invention combines a modal superposition method that considers structural characteristics with a deep learning model that can deeply explore the spatiotemporal nature of data. This achieves a collaborative optimization of modal prior knowledge and data-driven advantages. While considering the structural characteristics of the bridge, it deeply explores the spatiotemporal characteristics of the data and outputs high-precision reconstruction results with physically meaningful constraints. While maintaining computational efficiency, this method significantly improves the model's sensitivity to key features and enhances data reconstruction accuracy.

[0106] This invention achieves efficient joint mining and extraction of spatiotemporal features of data. It innovatively constructs a BO-CNN-BiLSTM-SENet deep learning model architecture, forming a closed loop of "spatial feature extraction → temporal modeling → end-to-end regression." This enables all-round coupled modeling from local spatial responses to long-range temporal dynamics, significantly improving the model's sensitivity to key features while maintaining computational efficiency.

[0107] This invention also overcomes the limitations of data reconstruction at the same measurement points. It establishes a data reconstruction method system across multiple physical quantities. This method not only enables reconstruction based on displacement measurement point data, but also utilizes the responses of strain measurement points to repair data from faulty measurement points. This method fully leverages the advantages of densely distributed strain measurement points and stable data, significantly improving the method's applicability in engineering practice.

[0108] The invention has strong applicability, quick calculation, convenient operation and high promotion value.

[0109] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0110] The present invention is described with reference to flowcharts and / or block diagrams of methods, systems, devices, and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0111] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0112] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0113] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0114] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A bridge health monitoring abnormal data reconstruction method, characterized in that: The following steps are involved: According to the modal superposition principle, the modal coordinates are solved based on the structural response data set measured at the effective measurement points of bridge health monitoring and the vibration mode matrix corresponding to the effective measurement points; According to the modal coordinates and the vibration shape vectors of the abnormal measurement points, the reconstructed estimated values ​​of the abnormal monitoring data are estimated; Constructing a data reconstruction model architecture and training the data reconstruction model, wherein the data reconstruction model uses the estimated trend of the reconstructed estimated value as a constraint; the data reconstruction model architecture includes a convolutional neural network module, a channel attention mechanism module, a bidirectional long short-term memory network module, and a Bayesian optimization module, wherein the convolutional neural network module is used to extract local features of the data time series, the channel attention mechanism module is used to dynamically quantify the spatial correlation between measurement points, the bidirectional long short-term memory network module is used to perform sequence analysis modeling based on fusion features, and the Bayesian optimization module is used to globally optimize the hyperparameters of the deep learning model, wherein the fusion features are features obtained by fusing the local features of the data time series and the spatial correlation between the measurement points; The reconstructed value of the abnormal data is obtained according to the trained data reconstruction model.

2. A bridge health monitoring abnormal data reconstruction method according to claim 1, characterized in that: The calculation formula of the modal coordinates is: ； in, represents the modal coordinates, Represents the vibration mode matrix of the effective measuring point, The displacement mode matrix with dimension m×n and the strain mode matrix with dimension k×n Composition, m represents the number of effective displacement measurement points, k represents the number of effective strain measurement points, n is the modal order, m+k≥n, represents the structural response data set of effective measurement points, The displacement response matrix with dimension m×p and the strain response matrix with dimension k×p Composition, p represents the measuring point The amount of data collected within a certain period of time.

3. The bridge health monitoring abnormal data reconstruction method according to claim 2 is characterized in that: The formula for estimating the reconstructed estimated value of the abnormal monitoring data is: ； in, represents the reconstructed estimated value of the abnormal monitoring data, Represents the vibration mode matrix of abnormal measurement points with dimension s×n, s represents the number of abnormal measurement points, s≤m+k.

4. The bridge health monitoring abnormal data reconstruction method according to claim 1 is characterized in that: The channel attention mechanism module is also used to evaluate the importance of each feature channel, enhance the expressiveness of key features through feature recalibration, and stimulate or compress corresponding channels for different tasks to extract useful features.

5. The bridge health monitoring abnormal data reconstruction method according to claim 1 is characterized in that: The hyperparameters include the number of BiLSTM units, learning rate, batch size, cycle, and number of Bayesian optimizations. The objective function in the optimization process of the hyperparameters of the global optimization deep learning model adopts the root mean square error function.

6. The bridge health monitoring abnormal data reconstruction method according to claim 1 is characterized in that: The data reconstruction model architecture includes a feature fusion layer for fusing the local temporal features of the data and the spatial correlation between the measurement points to obtain fused features.

7. A bridge health monitoring abnormal data reconstruction method according to any one of claims 1 to 6, characterized in that: The input of the data reconstruction model includes: the estimated reconstructed estimated value of the abnormal monitoring data, the data monitored by the effective measuring point, and the normal data monitored by the abnormal measuring point, and the output is the reconstructed value of the abnormal data.

8. A bridge health monitoring abnormal data reconstruction system, characterized by: include: A calculation module is used to solve the modal coordinates based on the modal superposition principle, the structural response data set measured by the effective measuring points of the bridge health monitoring and the mode shape matrix corresponding to the effective measuring points; An estimation module is used to estimate the reconstructed estimated value of the abnormal monitoring data based on the modal coordinates and the vibration shape vector of the abnormal measurement point; A model construction and training module is used to construct a data reconstruction model architecture and train the data reconstruction model; the data reconstruction model uses the estimated trend of the reconstructed estimated value as a constraint; the data reconstruction model architecture includes a convolutional neural network module, a channel attention mechanism module, a bidirectional long short-term memory network module, and a Bayesian optimization module. The convolutional neural network module is used to extract local features of the data time series, the channel attention mechanism module is used to dynamically quantify the spatial correlation between measurement points, the bidirectional long short-term memory network module is used to perform sequence analysis modeling based on fusion features, and the Bayesian optimization module is used to globally optimize the hyperparameters of the deep learning model. The fusion features are features obtained by fusing the local features of the data time series and the spatial correlation between the measurement points. The acquisition module is used to obtain the reconstructed value of the abnormal data according to the trained data reconstruction model.

9. An electronic device, characterized in that: include: Memory; one or more processors; One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for executing any one of the bridge health monitoring abnormal data reconstruction methods according to claims 1-7.

10. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the method for reconstructing abnormal data of bridge health monitoring according to any one of claims 1 to 7 is implemented.

Images